Find a real problem. Fix it. Prove it.

Pick one dinner your household cooks anyway and lock it as the meal you will measure four times.

• Ask whoever cooks which dinner repeats most often, and write down the exact dish and the night it is usually made.
• List every item that goes into it on one line each, including oil, salt, tea and anything drunk alongside it.
• Photograph the raw ingredients laid out on the counter before cooking, with a spoon in frame for scale.
• Photograph one plate as it is served and again after eating, from the same standing position both times.
• Open a Google Sheets file named 'Sixty-Litre Dinner' with columns: item, raw grams per person, price paid, litres per kilogram, source, litres on the plate.

Measure the raw mass of every ingredient that ends up on one person's plate.

• Zero the scale with the empty bowl on it, then weigh each raw ingredient in grams before it goes in the pot.
• Where the pot serves five, weigh the whole pot's ingredients and divide every figure by five, typing the division into the cell so it can be checked.
• Measure oil, milk and water in a marked jug and write the millilitres, noting that oil is lighter than water and stating the conversion you used.
• Weigh the served plate as well and record both raw and cooked mass, because the published water figures attach to raw mass.
• Photograph the scale display for at least three items and file the photographs in the dated folder.

Work out what one plate cost in rupees from prices actually paid, not from memory.

• Collect the shop receipts, and where there are none walk to the shop and write the current price per kilogram for each item on a card.
• Ask whoever shops where each item was bought and write the shop name beside the price.
• Divide each price per kilogram by one thousand and multiply by the grams on the plate, in a formula cell, to get rupees per item.
• Add cooking fuel as a single line, marked clearly as an estimate, with the reason for the estimate written next to it.

Look up the published water footprint per kilogram for each item and convert it to litres on your plate.

• Open the Water Footprint Network product gallery and search each ingredient by name, one at a time.
• Where an ingredient is missing there, use the freshwater-use-per-kilogram table on Our World in Data's 'Environmental Impacts of Food Production' page.
• Copy the figure, its units and the page title into the source column — a number with no source beside it is struck out at marking.
• Multiply litres per kilogram by kilograms on the plate inside a formula cell, so a stranger can see the arithmetic rather than trust it.
• Sort the sheet by litres, then write one line, in the source's own words, on what the coefficient includes and excludes.

Measure the food from this household thrown away over one week and attach its full footprint to it.

• Keep one bowl by the sink for seven days; after every meal scrape uneaten food into it and weigh it before it goes in the bin.
• Weigh anything thrown out uncooked from the fridge as well, item by item, and write the reason it was thrown.
• Apply the same litres-per-kilogram coefficient to the wasted mass, because the water was already spent before the food was binned.
• Add two rows to the sheet: grams wasted this week and litres of water thrown away this week.

Finish the costed problem statement on one page and choose the single change you will make.

• Write four lines: whose problem this is, litres per plate, rupees per plate, and litres and rupees per year at this household's real frequency.
• Build one Datawrapper bar chart of litres per item, largest bar first, titled with the finding rather than with the word 'chart'.
• List three possible changes — swap an ingredient, cut a portion, change where an item is bought, stop one form of waste — with the litres each would save on paper.
• Choose one and write the sentence: 'I will change X to Y for three weeks and expect litres per plate to fall from A to B.'
• Show the page to whoever cooks and record their agreement as a dated note or a voice message before anything changes.

Write the protocol that keeps the after-measurement identical to the before-measurement.

• Write what stays fixed: same dish, same scale, same number of people, same shops, same weighing point in the cooking.
• Write what is allowed to move: only the one thing you named in step six.
• Put the three cooking dates in a shared calendar and tell everyone who eats the meal.
• Add three blank labelled rows to the sheet for weeks one, two and three before any data exists.
• Record a thirty-second voice note stating your prediction, and file it with today's date on it.

Cook the meal with the change in place and measure exactly the quantities you measured before.

• Weigh every raw item again in grams per person, using the same scale and the same divisor.
• Record the prices paid this week rather than reusing the Act 1 prices.
• Apply the same coefficients from the same sources — do not change source in the middle of an experiment.
• Weigh this meal's waste into the same bowl and add it to the running count.
• Write down what went wrong: a substitute ingredient, an extra guest, a forgotten weighing.

Repeat the identical measurement twice more and compare all three weeks against the baseline.

• Cook and measure on the two booked dates without changing anything except what the protocol allows.
• Calculate the mean litres per plate across the three after-weeks and set it against the baseline plate.
• Chart baseline against weeks one, two and three in Datawrapper as four bars on one axis, in litres per plate.
• List every confounder you can name: guests, a price rise, an ingredient out of stock, a different cook, a different appetite.
• State the change as a number and as a percentage, and state plainly whether three measurements are enough to be sure.

Publish the finding, defend it to real people, and record what the household will keep.

• Record a defence video of no more than three minutes: the plate, the litres, the sources, the change, the result and the limits.
• Show the one-page explainer to three people who eat this meal and write down what each of them said.
• Send the chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask the household on camera whether the change stays and why, and let them answer without prompting.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick one dinner your household cooks anyway and lock it as the meal you will measure four times.

• Ask whoever cooks which dinner repeats most often, and write the dish and its usual night at the top of a fresh page.
• List every item that goes into it on one line each, including oil, salt, tea and anything drunk alongside it.
• Rule six columns across the page: item, grams, rupees, litres per kilogram, source, litres on plate.
• Draw the plate as served and label what is on it, or photograph it if a phone is available.

Measure the raw mass of every ingredient that ends up on one person's plate.

• Zero the scale with the empty bowl on it, then weigh each raw ingredient in grams before it goes in the pot.
• Where there is no scale, measure by level cupfuls, then weigh one cupful once on a shop's scale and write that conversion at the top of the page.
• Where the pot serves five, divide every figure by five and write the division out longhand so it can be checked.
• Weigh the served plate too, and record raw and cooked mass separately, because the published figures attach to raw mass.

Work out what one plate cost in rupees from prices actually paid, not from memory.

• Collect the receipts, and where there are none walk to the shop and copy the current price per kilogram onto your card.
• Ask whoever shops where each item was bought and write the shop name beside the price.
• Do the arithmetic longhand in the margin: price per kilogram divided by one thousand, multiplied by grams on the plate.
• Add cooking fuel as one line, marked as an estimate, with the reason written beside it.

Obtain a published litres-per-kilogram table on paper and convert your masses to litres by hand.

• Ask one person with a connection, a college library or a print shop for the Water Footprint Network product figures, or the Our World in Data table of freshwater use per kilogram, printed or copied out.
• Write the source across the top of the page: the organisation, the table name and the date it was obtained.
• Copy only the ingredients you need, one per row, with units, and write the source again in the source column of each row.
• Multiply litres per kilogram by kilograms on the plate longhand, leaving the working visible on the page.
• Underline the largest single figure and write one line on what that coefficient includes and excludes, in the source's own words.

Measure the food this household throws away over one week and attach its full footprint to it.

• Keep one bowl by the sink for seven days; after every meal scrape uneaten food into it and weigh it before it goes in the bin.
• Weigh anything thrown out uncooked from the fridge as well, item by item, and write the reason it was thrown.
• Apply the same borrowed coefficient to the wasted mass, because the water was already spent before the food was binned.
• Rule two more lines at the foot of the page: grams wasted this week, litres thrown away this week.

Finish the costed problem statement on one page and choose the single change you will make.

• Write four lines: whose problem this is, litres per plate, rupees per plate, and litres and rupees per year at this household's real frequency.
• Draw the litres-per-item bar chart on squared paper, largest bar first, with the scale stated and the finding as the title.
• List three possible changes — swap an ingredient, cut a portion, change where an item is bought, stop one form of waste — with the litres each would save.
• Choose one and write: 'I will change X to Y for three weeks and expect litres per plate to fall from A to B.'
• Read the page aloud to whoever cooks and have them sign and date it if they agree to the change.

Write the protocol that keeps the after-measurement identical to the before-measurement.

• Write what stays fixed: same dish, same scale, same number of people, same shops, same weighing point in the cooking.
• Write what is allowed to move: only the one thing you named in step six.
• Rule three blank labelled rows for weeks one, two and three before any data exists, and write the three dates beside them.
• Write your prediction on the page, date it, and have one person witness it with their signature.

Cook the meal with the change in place and measure exactly the quantities you measured before.

• Weigh every raw item again in grams per person, using the same scale and the same divisor.
• Write down the prices paid this week rather than copying the Act 1 prices down the page.
• Use the same borrowed coefficient table — do not change source in the middle of an experiment.
• Weigh this meal's waste into the same bowl and add it to the running count.
• Write what went wrong: a substitute ingredient, an extra guest, a forgotten weighing.

Repeat the identical measurement twice more and compare all three weeks against the baseline.

• Cook and measure on the two booked dates, changing nothing except what the protocol allows.
• Add the three after-weeks and divide by three, longhand, to get the mean litres per plate.
• Draw four bars on squared paper to one stated scale: baseline, week one, week two, week three.
• List every confounder you can name: guests, a price rise, an ingredient out of stock, a different cook, a different appetite.
• Write the change as a number and as a percentage, and say plainly whether three measurements is enough to be sure.

Publish the finding, defend it to real people, and record what the household will keep.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read the one-page explainer to three people who eat this meal and write down what each of them said.
• Send or hand the chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask the household in front of a witness whether the change stays and why, and write the answer down word for word.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Agree who is being logged, on what terms, and build the empty log before a single drink is counted.

• Sit the household down together and read out one sentence: this counts drinks, not people, nobody is weighed and nobody is named.
• Give each person a label — adult 1, adult 2, person 3 — write the labels on a card, and tell everyone they may stop being logged at any time without saying why.
• Open a Google Sheets file named 'The Sugar You Did Not Order' with columns: date, time, person label, drink, millilitres, sugar grams per hundred millilitres, spoons added, source, grams, teaspoons.
• Write the seven dates of the baseline week into the sheet now, one block per day, including the days you expect to be quiet.
• Type the sentence 'this project is about a product and a habit, never about a person's body' into cell A1 and leave it there for the whole project.

Record every drink taken by every logged person for seven consecutive days, by volume.

• Keep the sheet open on a phone in the kitchen and enter each drink within a few minutes of it being poured, not at the end of the day from memory.
• Measure the household's usual cup, glass and bottle once with a marked jug, write the millilitres of each on a card taped inside a cupboard door, and log by vessel afterwards.
• Log everything with sugar in it or on it: tea, coffee, soft drinks, packaged and fresh juice, flavoured milk, lassi, squash, cordial, energy drinks, sweetened yoghurt drinks and sweetened lemonade.
• Log drinks taken outside the house too, marked with the letter O, and log plain water and unsweetened tea as zero rows so the week is complete.
• At the end of each day type a single line saying whether anything was missed and roughly how much.

Get grams of sugar per hundred millilitres for every packaged drink in the log, from the label itself where possible.

• Collect every bottle, carton, tin and sachet from the week, including empties from the bin, and lay them out on the counter.
• Photograph each nutrition panel close enough to read, and file the photographs in a dated Drive folder named after the drink.
• Copy the sugars figure per hundred millilitres into the sheet, and where the panel gives per serving instead, divide by the serving size in millilitres and multiply by a hundred inside a formula cell.
• For any bottle with no readable panel, search Open Food Facts by product name or barcode and paste the page title into the source column beside the number.
• For fresh juice, plain milk or a yoghurt drink with no packet, take the figure from USDA FoodData Central and name the entry you used.

Measure the sugar added at home, which no label will ever tell you.

• Put a small bowl on a kitchen scale, zero it, add ten level teaspoons of sugar counted out loud, weigh the total, divide by ten, and write the grams in one level teaspoon at the top of the sheet.
• Where there is no scale, use four grams for one level teaspoon and write on the page that this is a standard figure you did not measure yourself.
• Watch the tea being made for three days and count the level teaspoons that go in, asking whoever makes it to spoon as normally as they can while you count.
• Where sugar goes into a shared pot, count the spoons into the pot, count the cups poured out of it, and divide in a formula cell so the division can be checked.
• Photograph the scale display showing the ten-spoon weight and file it with the label photographs.

Turn the whole week into one number anybody understands: level teaspoons of sugar drunk.

• In a formula cell, multiply millilitres by grams per hundred millilitres and divide by a hundred to get grams of sugar for each packaged drink row.
• For home-added sugar, multiply spoons by the grams in one level teaspoon, and add both figures where a drink has label sugar and spooned sugar.
• Divide every grams figure by the grams in one level teaspoon to give teaspoons per row, and sum the column for the week.
• Build three totals with the spreadsheet's SUMIF: teaspoons for the week, teaspoons by drink type, and teaspoons by person label.
• Draw the teaspoons-by-drink bar chart with the spreadsheet chart editor, largest bar first, and title it with the finding rather than with the word chart.

Finish the costed problem statement on one page and choose the single drink you will swap.

• Write four lines: which household and how many people, level teaspoons drunk this week, the same figure multiplied out across a year, and the rupees the week's drinks cost from receipts or shop prices.
• Put the teaspoons-by-drink chart on the same page with the four-gram-per-teaspoon conversion printed under it.
• List three possible swaps with the teaspoons each would save on paper: replace one packaged drink with water or unsweetened tea, halve the spoons in one home-made drink, or change one bottle size.
• Choose one, and only one, and write the sentence: 'For one week we will swap X for Y and expect the household total to fall from A teaspoons to B.'
• Read the page to the person who actually drinks that drink, ask them to agree or refuse, and record their answer and the date before anything changes.

Write down what must stay identical in the second week so the two weeks can be compared at all.

• Write what is fixed: same seven weekdays, same people and labels, same cups and vessels, same log sheet, same label figures, same grams per teaspoon.
• Write what is allowed to move: only the one drink named in step six.
• Add seven blank dated day blocks to the sheet for the swap week before any data exists, and tell everyone which day it starts.
• Buy or prepare the replacement drink before day one, so nobody is stuck without a choice at the moment of habit.
• Type your prediction into the sheet with today's date and time on it, and leave it visible.

Log the swap week exactly as you logged the baseline week, changing nothing else.

• Enter every drink at the kettle again, in the same columns, with the same person labels.
• Log the replacement drink as its own row with its own label figure or spoon count, rather than entering a zero and moving on.
• Log the swapped drink as zero on the days it did not happen, and log it honestly on any day it came back.
• Write one line each evening on how the swap held: kept, partly kept, or abandoned, and why.
• If anyone asks to stop being logged, stop that person's rows the same minute, mark them withdrawn with the date, and carry on with the rest.

Set the two weeks side by side and report the difference honestly, including if there is none.

• Total the swap week in teaspoons the same way and put the two totals side by side in the sheet.
• Draw the two-bar chart, baseline against swap week, in level teaspoons per household per week, on one axis with the scale starting at zero.
• Draw the swapped drink on its own second chart, so a reader can see whether the swap worked and the total still did not move.
• List every confounder by name: a guest, a hot week, a fast, an illness, a person withdrawing, a price change, somebody drinking more of something else instead.
• State the change as teaspoons and as a percentage, and write one plain sentence on whether one week either side is enough to be sure.

Publish the finding, defend it to real people, and record what the household will keep.

• Record a defence video of no more than three minutes: the log, the teaspoon total, the sources, the one swap, the result and the limits, with no name and no person on camera.
• Show the one-page explainer to three people who drink these drinks and write down what each of them said.
• Send the two-bar chart to the expert you interviewed with one question: what would you have counted that I did not.
• Ask the household, without prompting them, whether the swap stays, and record the answer in their own words using labels rather than names.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Agree who is being logged, on what terms, and rule the empty sheet before a single drink is counted.

• Sit the household down together and read out one sentence: this counts drinks, not people, nobody is weighed and nobody is named.
• Give each person a label — adult 1, adult 2, person 3 — write the labels at the head of the wall sheet, and tell everyone they may stop being logged at any time without saying why.
• Tape a large sheet by the kettle, rule seven dated day blocks down it, and rule five columns across: time, person label, drink, millilitres, spoons added.
• Write the seven dates in now, including the days you expect to be quiet.
• Write across the top of the sheet: this project is about a product and a habit, never about a person's body.

Record every drink taken by every logged person for seven consecutive days, by volume.

• Hang the pen on a string beside the sheet and mark each drink as it is poured, not at the end of the day from memory.
• Measure the household's usual cup, glass and bottle once with a jug, write the millilitres of each at the top of the sheet, and log by vessel afterwards.
• Log everything with sugar in it or on it: tea, coffee, soft drinks, packaged and fresh juice, flavoured milk, lassi, squash, cordial, energy drinks, sweetened yoghurt drinks and sweetened lemonade.
• Log drinks taken outside the house too, marked with the letter O, and log plain water and unsweetened tea as zero rows so the week is complete.
• Keep every bottle, carton and sachet in a box under the sink for the whole week instead of throwing it away.

Get grams of sugar per hundred millilitres for every packaged drink from the packets in your box.

• Empty the box onto the table and sort the packets into one pile per drink.
• Copy the sugars figure per hundred millilitres from each nutrition panel into a table in the notebook, one row per drink, with the brand and the pack size beside it.
• Where a panel gives sugar per serving, write the serving size in millilitres and do the conversion longhand in the margin so the working stays visible.
• For any drink with no panel at all, walk to the shop and copy the panel off a full bottle on the shelf, writing the shop and the date beside the figure.
• For fresh juice, plain milk or a home-made yoghurt drink, ask a pharmacist, a dietitian or a college library for a printed composition figure and write its source across the top of the table.

Measure the sugar added at home, which no label will ever tell you.

• Put a small bowl on a kitchen scale, zero it, add ten level teaspoons counted out loud, weigh the total, divide by ten longhand, and write the grams in one level teaspoon at the top of the sheet.
• Where there is no scale at home, weigh the ten spoons once on a shop's scale, or use four grams and write on the page that this is a standard figure you did not measure yourself.
• Watch the tea being made for three days and count the level teaspoons that go in, asking whoever makes it to spoon as normally as they can while you count.
• Where sugar goes into a shared pot, count the spoons into the pot, count the cups poured out of it, and write the division out longhand.
• Level every spoon against the rim with a knife before counting it, and say so on the page.

Turn the whole week into one number anybody understands: level teaspoons of sugar drunk.

• For each packaged drink row, multiply millilitres by grams per hundred millilitres and divide by a hundred, longhand in the margin, to get grams of sugar.
• For home-added sugar, multiply spoons by the grams in one level teaspoon, and add both figures where a drink carries label sugar and spooned sugar.
• Divide every grams figure by the grams in one level teaspoon to get teaspoons, and add the column up twice, once from the top and once from the bottom, to check it.
• Rule three summary tables at the foot of the sheet: teaspoons for the week, teaspoons by drink type, and teaspoons by person label.
• Draw the teaspoons-by-drink bar chart on squared paper, largest bar first, with the scale stated and the finding as the title.

Finish the costed problem statement on one page and choose the single drink you will swap.

• Write four lines: which household and how many people, level teaspoons drunk this week, the same figure multiplied out across a year, and the rupees the week's drinks cost from receipts or shop prices.
• Copy the teaspoons-by-drink chart onto the same page with the grams-per-teaspoon conversion printed under it.
• List three possible swaps with the teaspoons each would save: replace one packaged drink with water or unsweetened tea, halve the spoons in one home-made drink, or change one bottle size.
• Choose one, and only one, and write: 'For one week we will swap X for Y and expect the household total to fall from A teaspoons to B.'
• Read the page aloud to the person who actually drinks that drink and have them sign and date it if they agree, or record their refusal and choose another swap.

Write down what must stay identical in the second week so the two weeks can be compared at all.

• Write what is fixed: same seven weekdays, same people and labels, same cups and vessels, same ruled sheet, same label figures, same grams per teaspoon.
• Write what is allowed to move: only the one drink named in step six.
• Rule and date a fresh wall sheet for the swap week before any data exists, and tell everyone which day it starts.
• Buy or prepare the replacement drink before day one, so nobody is stuck without a choice at the moment of habit.
• Write your prediction at the foot of the page, date it, and have one person witness it with a signature.

Log the swap week exactly as you logged the baseline week, changing nothing else.

• Mark every drink at the kettle again, in the same columns, with the same person labels.
• Give the replacement drink its own row with its own label figure or spoon count, rather than writing a zero and moving on.
• Write the swapped drink as zero on the days it did not happen, and write it honestly on any day it came back.
• Write one line each evening on how the swap held: kept, partly kept, or abandoned, and why.
• If anyone asks to stop being logged, stop that person's rows the same minute, mark them withdrawn with the date, and carry on with the rest.

Set the two weeks side by side and report the difference honestly, including if there is none.

• Total the swap week in teaspoons the same way, adding the column twice, and write the two totals side by side.
• Draw the two bars on squared paper to one stated scale starting at zero, baseline against swap week, in level teaspoons per household per week.
• Draw the swapped drink on its own small chart, so a reader can see whether the swap worked and the total still did not move.
• List every confounder by name: a guest, a hot week, a fast, an illness, a person withdrawing, a price change, somebody drinking more of something else instead.
• Write the change as teaspoons and as a percentage, and add one plain sentence on whether one week either side is enough to be sure.

Publish the finding, defend it to real people, and record what the household will keep.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone with no person in shot.
• Read the one-page explainer to three people who drink these drinks and write down what each of them said.
• Hand or post the two-bar chart to the expert you interviewed with one question: what would you have counted that I did not.
• Ask the household, without prompting them, whether the swap stays, and write the answer down word for word using labels rather than names.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Choose one spot you can stand on every evening for a month and fix it so it never moves.

• Stand where you can see the western skyline from a window, a courtyard or a walled roof, and check that an adult in the household can see or hear you there.
• Mark the standing position with chalk and tape a card at your eye height on the frame beside you, so height as well as position is fixed.
• Open OpenStreetMap, find the building, and copy the latitude and longitude to five decimal places into your sheet.
• Photograph the western skyline from the mark and draw over the photograph the wall top, ridge or roof line the sun goes behind.
• Write one line on what blocks the true horizon and roughly how high it stands above your eye height.

Get a clock you can trust to the second and record the evidence that it is trustworthy.

• Open time.is on the device you will read each evening and let it report the offset between that device and network time.
• If the device is more than two seconds out, correct it in the settings, reload the page, and photograph the reading showing the corrected offset.
• Write the device name, the offset in seconds and the date into the sheet, and repeat this check on the first evening of each week.
• Agree the rules aloud with an adult in the household: which spot, which time, and that you never look at the sun or use any lens.

Record what the printed calendar, the app and an astronomical computation each state for your fourteen dates.

• Photograph the wall calendar or printed timetable page covering the next fourteen days, so the source is fixed and cannot be edited later.
• Write the app's name and version, then copy its stated time for each of the fourteen dates into the sheet.
• Open Stellarium Web, set the location to the coordinates from step one, and read the computed time for each of the same dates.
• Write down which city, zone or point each printed source says it was computed for; where it does not say, write 'not stated'.
• Build the columns: date, calendar time, app time, computed time, observed time, difference in minutes, sky condition.

Observe the actual moment from the fixed spot for fourteen evenings and record it to the minute.

• Stand on the chalk mark ten minutes before the printed time, with an adult in the household knowing you are there.
• Keep your eyes on the wall top, ridge or horizon edge and never on the disc itself, and use no phone screen, camera, glasses or lens of any kind.
• Note the minute the last light leaves that edge, reading the clock you proved in step two.
• Write or photograph the clock reading immediately, before you step off the mark.
• Record the sky in three words — clear, hazy, cloud on the horizon, dust — and mark any evening you could not see the edge as unusable rather than guessing.

Calculate the signed difference in minutes against each published source and describe how it behaves.

• Subtract the calendar time from the observed time for every usable date in a formula cell, keeping the sign so early and late stay apart.
• Repeat against the app time and against the computed time, giving three difference columns on the same rows.
• Calculate the mean, the largest and the smallest difference for each source, and count how many evenings each is out by three minutes or more.
• Chart the fortnight in Datawrapper as signed difference in minutes against date, with a line drawn at zero.
• Write one plain paragraph on why an error exists at all: the table is computed for a city centre or a broad zone, while you stand at your own longitude, your own altitude and behind your own wall, and a wall or a hill changes the moment the sun disappears from where you are standing.

Finish the costed problem statement and commit to publishing a corrected table for this one spot.

• Write four lines: whose problem this is, the mean difference in minutes at this spot, the worst evening of the fortnight, and the minutes of mistiming per person per year at that mean.
• Count the people who use the same table at or near this spot — the household, the floor, the neighbours — and multiply to get the yearly total minutes across all of them.
• Write one line without hedging: this measures visibility from one spot, it is an observation and not a verdict, and questions of religious practice belong to a qualified scholar.
• Choose the fix and write it as one sentence: 'I will publish a corrected table for this spot and get at least five people to use it for fourteen days.'
• Show the page to an adult in the household and record their reaction in their own words.

Turn the measured offset into a corrected thirty-day table with its evidence and its limits printed on it.

• Read the computed times for the next thirty dates at your coordinates from Stellarium Web and put them in the sheet.
• Apply your measured mean offset in a formula cell, so the correction is visible arithmetic rather than a hand-typed number.
• Add a header block: the exact spot, the coordinates, the dates observed, the mean offset, the range and the number of usable evenings behind it.
• Add a footer block: the horizon this table assumes, the plain statement that it does not hold for anyone standing elsewhere, and the sentence that this is an observation and not a ruling.
• Give it to one person who was not involved, ask them what it means, and correct anything they misread.

Put the corrected table in the hands of at least five people and agree how their use will be recorded.

• Give or send the table to five people who stand at or near your spot, one at a time, explaining the one number and its limits in under a minute.
• Ask each of them directly whether they will use it for fourteen days, and write the answer down, including every refusal.
• Agree with each user how use gets recorded: a tick sheet on the wall, a daily message, or one question you ask at a fixed time.
• Write the user ledger: name, role, date given, agreed recording method, and their first reaction word for word.
• Record a thirty-second voice note stating what adoption you expect, dated before the fortnight starts.

Observe fourteen more evenings against the corrected table and count who actually used it.

• Observe from the same chalk mark at the same eye height under the same safety rules, recording the observed minute exactly as before.
• Subtract the corrected table time from the observed time each evening to get the residual error in minutes.
• Collect every user's tick sheet or message thread on the agreed day and count days used, not days given.
• Ask every user the same two questions and write the answers word for word: did it change what you did, and will you keep using it.
• Write down every reason a user stopped: forgot, did not trust it, moved away, disagreed with the idea.

Publish the finding, defend it to real people, and record what happens to the table now.

• Record a defence video of no more than three minutes: the spot, the mean error, the sources, the corrected table, the residual, the adoption and the limits.
• Show the one-page explainer to three people who use the table and write down what each of them said.
• Send the fortnight chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask the users on camera whether the table stays on the wall and why, and let them answer without prompting.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Choose one spot you can stand on every evening for a month and fix it so it never moves.

• Stand where you can see the western skyline from a window, a courtyard or a walled roof, and check an adult in the household can see or hear you there.
• Mark the standing position with chalk and tape a card at your eye height on the frame beside you, so height as well as position is fixed.
• Draw the western skyline on the first page of the notebook and mark on it the wall top, ridge or roof line the sun goes behind.
• Pace out and write down where the spot sits: which room, which side of the building, how many steps from a named door.
• Write one line on what blocks the true horizon and roughly how high it stands above your eye height.

Set your watch from a clock known to be accurate and record how far it drifts.

• Find a clock known to be accurate — one set from a network, a radio time signal, a station clock or a television news clock — and name it in the notebook.
• Set the watch against it to the second, and write the date, the time and the clock you used.
• Re-check the watch against that same clock every third evening and write the drift in seconds each time, including the zeros.
• Agree the rules aloud with an adult in the household: which spot, which time, and that you never look at the sun or through any lens.

Copy out what the calendar and a second published source state for your fourteen dates.

• Copy the wall calendar or printed timetable times for the next fourteen dates into the first column, exactly as printed.
• Find a second published source — a newspaper's daily times, a mosque notice, a second printed calendar — and copy its times for the same dates.
• Write down which city, zone or point each source says it was computed for; where it does not say, write 'not stated'.
• Write where each source came from and on what date you copied it, so the copy can be traced back.
• Rule the remaining columns for observed time, difference in minutes, sky condition and notes.

Observe the actual moment from the fixed spot for fourteen evenings and record it to the minute.

• Stand on the chalk mark ten minutes early, with an adult in the household knowing you are there.
• Keep your eyes on the wall top, ridge or horizon edge and never on the disc itself, and use no glasses, binoculars, camera or lens of any kind.
• Note the minute the last light leaves that edge, reading the watch you set in step two.
• Write the reading in the notebook while still standing on the mark, before you move.
• Record the sky in three words, and mark any evening you could not see the edge as unusable rather than guessing.

Calculate the signed difference in minutes against both published sources and describe how it behaves.

• Subtract the calendar time from the observed time for every usable date, longhand in the margin, keeping the sign so early and late stay apart.
• Repeat against the second published source, filling the second difference column on the same rows.
• Add the differences and divide by the number of usable evenings, longhand, then underline the largest and the smallest.
• Draw the fortnight on squared paper: difference in minutes up the side, date along the bottom, a ruled line at zero, and the scale stated.
• Write one plain paragraph on why an error exists at all: the table is computed for a city centre or a broad zone, while you stand at your own longitude, your own altitude and behind your own wall, and a wall or a hill changes the moment the sun disappears from where you are standing.

Finish the costed problem statement and commit to publishing a corrected table for this one spot.

• Write four lines: whose problem this is, the mean difference in minutes at this spot, the worst evening of the fortnight, and the minutes of mistiming per person per year at that mean.
• Count the people who use the same table at or near this spot and multiply longhand to get the yearly total minutes across all of them.
• Write one line without hedging: this measures visibility from one spot, it is an observation and not a verdict, and questions of religious practice belong to a qualified scholar.
• Choose the fix and write it as one sentence: 'I will publish a corrected table for this spot and get at least five people to use it for fourteen days.'
• Read the page aloud to an adult in the household and write their reaction down in their own words.

Write out a corrected thirty-day table by hand, with its evidence and its limits printed on it.

• Copy the printed source's times for the next thirty dates into a fresh ruled page, one date per line.
• Add your measured mean offset to each line longhand, and write both the original and the corrected time so the arithmetic stays visible.
• Write a header block: the exact spot, how to find it, the dates observed, the mean offset, the range and the number of usable evenings behind it.
• Write a footer block: the horizon this table assumes, the plain statement that it does not hold for anyone standing elsewhere, and the sentence that this is an observation and not a ruling.
• Give it to one person who was not involved, ask them what it means, and correct anything they misread.

Put the corrected table in the hands of at least five people and agree how their use will be recorded.

• Copy the table out five times by hand or at a print shop, and give one to each of five people who stand at or near your spot.
• Explain the one number and its limits in under a minute to each of them, and ask directly whether they will use it for fourteen days.
• Write the answer down, including every refusal, and agree how use gets recorded: a tick sheet pinned beside the table, or one question you ask at a fixed time.
• Write the user ledger: name, role, date given, recording method, and their first reaction word for word.
• Write your expected adoption on a page, date it, and have one person sign as witness before the fortnight starts.

Observe fourteen more evenings against the corrected table and count who actually used it.

• Observe from the same chalk mark at the same eye height under the same safety rules, recording the observed minute exactly as before.
• Subtract the corrected table time from the observed time each evening, longhand, to get the residual error in minutes.
• Collect every user's tick sheet on the agreed day and count days ticked, not copies handed out.
• Ask every user the same two questions and write the answers word for word: did it change what you did, and will you keep using it.
• Write down every reason a user stopped: forgot, did not trust it, moved away, disagreed with the idea.

Publish the finding, defend it to real people, and record what happens to the table now.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read the one-page explainer to three people who use the table and write down what each of them said.
• Hand or send the drawn chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask the users in front of a witness whether the table stays on the wall and why, and write the answers word for word.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Walk your house, shop or classroom and list every object that has failed more than once, then choose one.

• Walk one lap of each room with the sheet open and write down every object that has been repaired, tied, taped, glued, wired or propped.
• Ask three people who use these things daily how often each one fails, and record their answers as times per month, never as the word 'often'.
• Photograph each candidate where it sits, close enough to show the joint or the point that gives.
• Score each candidate in three columns: times per month, minutes lost per failure, and whether it can be tested without touching anything electrical, gas, structural or above your reach.
• Choose the highest-scoring object that passes the safety column, and write one sentence saying why you chose it over the runner-up.

Collect physical evidence that this object has failed before, instead of trusting anybody's memory.

• Photograph every repair on it: old glue, a second screw hole, wire, tape, a replaced part, a bent bracket, a worn groove.
• Count the separate repair marks and write the count down, because each mark is a past failure somebody already paid for.
• Ask the three users when each repair happened and write the nearest month against each mark.
• Ask whoever fixed it last what they used and what it cost, and check whether the receipt or the leftover material still exists.
• Measure the object with a tape and record the dimensions that matter: thickness at the failure point, length of the arm, diameter of the fixing.

Produce the failure on purpose in a controlled test, and start a fourteen-day log of failures in ordinary use.

• Define one use in writing — one full open and close, one lift with the bag loaded to its usual weight, one turn of the handle — and do not change the definition afterwards.
• Run one hundred of those identical uses at a steady, ordinary speed with the same load each time, marking a tally every time it slips, jams, loosens, tears or comes apart.
• Load only with weight you can hold in one hand, placed so it can only fall onto bare floor, with feet and hands clear and your face turned away from anything under tension.
• Film thirty seconds of the test and one close-up of the exact point that gives, so a stranger can see the failure rather than take your word.
• Tape a tally card beside the object and begin a fourteen-day count of failures in ordinary use, one line per failure with date, time and who was using it.

Work out what a single failure costs and turn it into rupees and hours per year.

• Price the repair from a receipt or a shop visit: the material, the part, and what a tradesman charges to come, each with the shop or the name beside it.
• Price a full replacement of the object at two shops and record both prices with the shop names.
• Time the last failure end to end with the people involved: minutes to notice, to find tools, to fix, to walk to the shop, and the minutes the thing was unusable.
• Multiply cost per failure by the failure rate the three users gave you, and again by the rate your own tally implies, so two independent estimates sit side by side.
• Put the arithmetic in visible formula cells: failures per year times rupees per failure, and failures per year times minutes per failure divided by sixty.

Work out where the load concentrates and why the same point gives every time.

• Photograph the failure point close up, then draw the object in Excalidraw and trace the path the force takes, from where the hand or the weight pushes to where the object is held.
• Mark three places on the drawing: the narrowest section, the sharpest corner, and the single fixing carrying most of the load, because failures start at those three.
• State plainly whether the material gave or the joint gave, since a fixing working loose is a different problem from a bar bending.
• Check whether the same point failed at every repair mark from step two, and write yes or no with the evidence.
• Put the one question you cannot answer yourself to the tradesman you interview, and write their answer down word for word.

Finish the costed problem statement on one page and choose the single change you will make.

• Write five lines: whose problem this is, failures per hundred uses, failures per fortnight, rupees per year, hours per year.
• Build one Google Sheets bar chart of rupees per year for your object beside the two runners-up from step one, titled with the finding rather than with the word 'chart'.
• List three possible fixes using only what is already in the house or the shop: a second fixing, a packing plate, a stop that limits travel, a doubled strap, a washer at the wear point.
• Choose one and write the sentence: 'I will change X to Y and expect failures per hundred uses to fall from A to B and failures per fortnight from C to D.'
• Show the page to whoever owns the object, get their agreement to the change, and record it as a dated note before you touch anything.

Draw the change to real dimensions, fit it, and write the protocol that keeps the after-test identical to the before-test.

• Draw the reinforcement in Tinkercad using the dimensions measured in step two, showing the new fixing or plate in position on the object.
• Write what stays fixed: same definition of one use, same load, same tester, same place, same speed, same hundred-count, same definition of a failure.
• Fit the change with what is already in the house; if a hand tool is needed, an adult holds the work, and nothing is cut, drilled or heated by you alone.
• Photograph the object before fitting and after fitting from the same standing position with the same item in frame for scale.
• Record a thirty-second voice note stating your prediction and file it with today's date, before any after-data exists.

Run the identical hundred-use test on the changed object and count failures exactly as before.

• Use the same load, the same speed, the same tester and the same definition of one use, checked line by line against the protocol before starting.
• Tally every failure of the same kind, and record separately anything new that started going wrong because of the change.
• Film thirty seconds of the changed object under test, framed the same way as the first video.
• Inspect the change afterwards and photograph it: has it loosened, bent, worn, or shifted position.
• Write down what went wrong on the day: a distracted count, a load that slipped, a helper who joined in halfway.

Log failures in ordinary use for fourteen more days and compare like with like.

• Tape a fresh tally card in the same place and log every failure with date, time, user and minutes lost, exactly as in the baseline fortnight.
• Check the object once a week for the change working loose, photograph it, and write down what you find.
• Chart four bars on one axis in Google Sheets: failures per hundred uses before and after, failures per fortnight before and after.
• List every confounder you can name: colder or wetter weather, a lighter user, the object being used less because everybody knows it is being watched, a second repair somebody made without telling you.
• Recalculate rupees and hours per year with the same cost model, and state whether the fix saved more than the time it took to make.

Publish the finding, defend it to real people, and record whether the change stays on the object.

• Record a defence video of no more than three minutes: the object, the failure point, the two before-counts, the change, the two after-counts and the limits.
• Show the one-page explainer to three people who use the object and write down what each of them said.
• Send the chart and the drawing to the tradesman you interviewed with one question: what would you have reinforced that I did not.
• Ask the owner of the object on camera whether the change stays and why, and let them answer without prompting.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Walk your house, shop or classroom and list every object that has failed more than once, then choose one.

• Walk one lap of each room with the notebook open and write down every object that has been repaired, tied, taped, glued, wired or propped.
• Ask three people who use these things daily how often each one fails, and write their answers as times per month, never as the word 'often'.
• Sketch each candidate on its own line, marking the joint or the point that gives.
• Rule three columns beside the list: times per month, minutes lost per failure, and safe to test yes or no, against electrical, gas, structural or above your reach.
• Choose the highest-scoring object that passes the safety column and write one sentence saying why you chose it over the runner-up.

Collect physical evidence that this object has failed before, instead of trusting anybody's memory.

• Draw the object large on one page and mark every repair on the drawing: old glue, a second screw hole, wire, tape, a replaced part, a worn groove.
• Count the separate repair marks and write the count down, because each mark is a past failure somebody already paid for.
• Ask the three users when each repair happened and write the nearest month against each mark on the drawing.
• Ask whoever fixed it last what they used and what it cost, and note whether the receipt or leftover material still exists.
• Measure the object with a tape and write the dimensions on the drawing: thickness at the failure point, length of the arm, diameter of the fixing.

Produce the failure on purpose in a controlled test, and start a fourteen-day log of failures in ordinary use.

• Write down the definition of one use — one full open and close, one lift with the bag loaded to its usual weight, one turn of the handle — and keep that definition for the rest of the project.
• Run one hundred of those identical uses at a steady, ordinary speed with the same load, and have a second person mark a tally every time it slips, jams, loosens, tears or comes apart.
• Use a known load you can hold in one hand, such as filled bottles counted as one kilogram each, placed so it can only fall onto bare floor, with feet and hands clear and your face turned away.
• Draw the exact point that gives, large, immediately after the test, and have your counter sign and date the tally sheet.
• Tape a tally card beside the object and begin a fourteen-day count of failures in ordinary use, one line per failure with date, time and who was using it.

Work out what a single failure costs and turn it into rupees and hours per year.

• Walk to the hardware shop and copy the price of the material, the part and the tradesman's callout charge onto a card with the shop name.
• Ask at two shops what a full replacement of the object costs and write both prices down.
• Sit with the people involved and time the last failure end to end: minutes to notice, to find tools, to fix, to walk to the shop, and the minutes the thing was unusable.
• Do the arithmetic longhand in the margin: failures per year times rupees per failure, and failures per year times minutes per failure divided by sixty.
• Repeat the arithmetic twice, once with the users' failure rate and once with the rate your own tally implies, and leave both on the page.

Work out where the load concentrates and why the same point gives every time.

• Draw the object large on squared paper and draw arrows along the path the force takes, from where the hand or the weight pushes to where the object is held.
• Mark three places on the drawing: the narrowest section, the sharpest corner, and the single fixing carrying most of the load, because failures start at those three.
• Write plainly whether the material gave or the joint gave, since a fixing working loose is a different problem from a bar bending.
• Compare with the repair marks from step two and write yes or no against the question: did it fail at the same place every time.
• Put the one question you cannot answer yourself to the tradesman you interview, and write their answer down word for word.

Finish the costed problem statement on one page and choose the single change you will make.

• Write five lines: whose problem this is, failures per hundred uses, failures per fortnight, rupees per year, hours per year.
• Draw a bar chart on squared paper to a stated scale, rupees per year for your object beside the two runners-up from step one, with the finding as the title.
• List three possible fixes using only what is already in the house or the shop: a second fixing, a packing plate, a stop that limits travel, a doubled strap, a washer at the wear point.
• Choose one and write: 'I will change X to Y and expect failures per hundred uses to fall from A to B and failures per fortnight from C to D.'
• Read the page aloud to whoever owns the object and have them sign and date it if they agree to the change.

Draw the change to real dimensions, fit it, and write the protocol that keeps the after-test identical to the before-test.

• Draw the reinforcement on squared paper to a stated scale, using the dimensions measured in step two, with the new fixing or plate shown in position.
• Write what stays fixed: same definition of one use, same load, same tester, same counter, same place, same speed, same hundred-count.
• Fit the change with what is already in the house; if a hand tool is needed, an adult holds the work, and nothing is cut, drilled or heated by you alone.
• Draw or photograph the object before fitting and after fitting from the same position, with the same item beside it for scale.
• Write your prediction on the page, date it, and have one person witness it with their signature before any after-data exists.

Run the identical hundred-use test on the changed object and count failures exactly as before.

• Read the protocol aloud before starting and check the load, the speed, the tester and the definition of one use line by line.
• Have the same counter mark the tally, and have them sign and date the sheet at the end.
• Record separately anything new that started going wrong because of the change, in its own column.
• Inspect the change afterwards and draw it: has it loosened, bent, worn or shifted position.
• Write down what went wrong on the day: a distracted count, a load that slipped, a helper who joined in halfway.

Log failures in ordinary use for fourteen more days and compare like with like.

• Tape a fresh tally card in the same place and log every failure with date, time, user and minutes lost, exactly as in the baseline fortnight.
• Check the object once a week for the change working loose, and write down what you find each time.
• Draw four bars on squared paper to one stated scale: failures per hundred uses before and after, failures per fortnight before and after.
• List every confounder you can name: colder or wetter weather, a lighter user, the object being used less because everybody knows it is being watched, a second repair somebody made without telling you.
• Redo the rupees and hours arithmetic longhand with the new failure rate and state whether the fix saved more than the time it took to make.

Publish the finding, defend it to real people, and record whether the change stays on the object.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read the one-page explainer to three people who use the object and write down what each of them said.
• Hand the chart and the drawing to the tradesman you interviewed with one question: what would you have reinforced that I did not.
• Ask the owner of the object, in front of a witness, whether the change stays and why, and write the answer word for word.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Locate your meter, work out what its display means, and take the first dated reading without touching anything.

• Stand in front of the meter in daylight and photograph the whole face, then the number panel close up, keeping both hands at your sides.
• Write down the meter number printed on the front and check it matches the meter number printed on your bill.
• Identify which figure is the units total in kWh; on a digital meter press nothing and wait for the display to cycle to that screen by itself.
• Note whether the final digit is a decimal, and whether the meter shows separate peak and off-peak totals.
• Open a Google Sheets file named 'Meter log' with columns: date, time, reading, load-shedding hours since last reading, generator or UPS used, notes.

Build the baseline by reading the meter at the same two clock times every day for a week.

• Choose two fixed times, for example 07:00 and 22:00, and set two daily alarms for them.
• At each alarm photograph the display and type the reading into the sheet within one minute, with the actual time to the minute.
• Record load-shedding start and end times every day, and any hours the house ran on a generator, a UPS or solar.
• Subtract in formula cells to get units used in the day block and units used in the night block.
• Add one word for the day's weather and a note on whether the house was fuller or emptier than usual.

Turn your household's own bill into a rupee-per-unit rate you can defend.

• Photograph this month's bill and last month's, then open your distribution company's online bill page and pull twelve months of billed units.
• Circle on the photograph: units billed, the slab breakdown, fixed charges, every tax line, and the total payable.
• Calculate two rates — total payable divided by units billed, and the marginal slab rate your next unit falls into, taken from the NEPRA tariff schedule.
• Enter both rates in the sheet and write one sentence naming which rate you will use for the appliance costing and why.

Measure what one appliance draws by running it alone against the meter.

• Agree one appliance with the household: an air conditioner, a fridge, a water pump, a geyser, an iron, a heater, or the lights on one floor.
• Copy the wattage and voltage printed on the appliance's own label, without moving it, opening it or removing any cover.
• Agree a quiet window, then switch off every other load at the wall switches the household already uses, leaving only the one appliance running.
• Read and photograph the meter, start the stopwatch, run the appliance for a measured thirty to sixty minutes, then read and photograph the meter again.
• Divide units used by hours run to get units per hour, then repeat the whole test on a second day to see whether the figure holds.

Log how many hours the appliance actually runs and turn that into rupees per month.

• Keep a tally sheet beside the appliance and mark every start and stop time for seven days.
• For a fridge or a pump that cycles on its own, sit with it for a full hour and time how many minutes in that hour it actually runs, twice at different times of day.
• Multiply units per hour by hours per week, then by 4.33 weeks, then by the rupee rate you chose.
• Set that figure against the whole-house units from step two and state what percentage of the house this one appliance is.

Finish the costed problem statement on one page and choose the single change you will make.

• Write five lines: whose problem this is, units per month, rupees per month, rupees per year at today's tariff, and share of the whole house.
• Chart the seven day-blocks against the seven night-blocks in Datawrapper, titled with the finding rather than with the word 'chart'.
• List three candidate changes that cost nothing: a thermostat or fan setting, a timing shift, a door seal pressed back, a bulb already owned, or one habit.
• Choose one and write: 'I will change X for three weeks and expect this appliance to fall from A units per week to B.'
• Show the page to whoever pays the bill and record their agreement, dated, before anything changes.

Write the protocol that makes the after-measurement comparable with the before.

• Write what stays fixed: the same two clock times, the same meter, the same rupee rate, the same appliance, the same tally method.
• Write what is allowed to move: only the one thing you named in step six.
• Prepare twenty-one blank dated rows in the sheet before any data exists.
• Write the confounders you already expect: weather, guests, load-shedding hours, a different person using the appliance.
• Record a dated prediction of the after figure, in units and in rupees, before anything is changed.

Run the change for seven days and read the meter exactly as you read it before.

• Take both daily readings at the same two clock times and photograph each display.
• Keep the appliance tally sheet running exactly as it ran in Act 1.
• Repeat the isolation test once under the new setting and record units per hour again.
• Log load-shedding and generator hours daily, and mark the week as disturbed if those hours moved sharply against the baseline week.
• Write down every slip: a missed reading, a night the change was forgotten, a guest who reversed it.

Repeat two more weeks, then check your finding against the next printed bill.

• Take the same two readings daily for a further fourteen days without altering anything else.
• Calculate mean units per week across the three after-weeks and set it against the baseline week.
• When the next bill arrives, compare its billed units with your own meter arithmetic and explain any gap in writing.
• Chart baseline against weeks one, two and three on one axis, in units and again in rupees.
• Write the confounder paragraph: temperature, load-shedding hours, people in the house, and anything else that moved.

Publish the finding, defend it to real people, and record what the household will keep.

• Record a defence video of no more than three minutes: the meter, the number, the method, the change, the result and the limits.
• Show the one-page explainer to three neighbouring households on the same tariff and write down what each of them said.
• Send the chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask whoever pays the bill, on camera, whether the change stays and why, and let them answer without prompting.
• Complete the relations ledger with ten names, roles, dates and replies.

Locate your meter, work out what its display means, and take the first dated reading without touching anything.

• Stand in front of the meter in daylight and copy every figure on the face into the notebook exactly as it appears, hands at your sides.
• Copy the meter number printed on the front and check it against the meter number printed on the bill.
• Identify which figure is the units total in kWh; on a digital meter press nothing and wait for the display to cycle to that screen by itself.
• Sketch the meter face and label what each figure means, marking the decimal digit if there is one.
• Rule the log columns: date, time, reading, difference, load-shedding hours, notes.

Build the baseline by reading the meter at the same two clock times every day for a week.

• Choose two fixed times, for example 07:00 and 22:00, and agree with the household that somebody reminds you at each.
• At each time copy the reading into the log with the actual clock time, in ink, without correcting earlier entries.
• Write load-shedding start and end times every day, and any hours run on a generator, a UPS or solar.
• Subtract longhand in the difference column to get day-block units and night-block units.
• Add one word for the weather and a note on whether the house was fuller or emptier than usual.

Turn your household's own printed bill into a rupee-per-unit rate you can defend.

• Lay this month's and last month's bills side by side and circle in pen: units billed, the slab breakdown, fixed charges, every tax line, and the total payable.
• Copy twelve months of billed units from the consumption history box printed on the bill itself into your notebook.
• Divide total payable by units billed, longhand, to get the average rate, and copy the slab rate your next unit falls into from the tariff table printed on the bill.
• Write one sentence naming which rate you will use for the appliance costing and why.

Measure what one appliance draws by running it alone against the meter.

• Agree one appliance with the household: an air conditioner, a fridge, a water pump, a geyser, an iron, a heater, or the lights on one floor.
• Copy the wattage and voltage printed on the appliance's own label, without moving it, opening it or removing any cover.
• Agree a quiet window, then switch off every other load at the wall switches the household already uses, leaving only the one appliance running.
• Copy the meter reading and the clock time, run the appliance for a measured thirty to sixty minutes, then copy the reading and the time again.
• Divide units used by hours run longhand to get units per hour, then repeat the whole test on a second day to see whether the figure holds.

Tally how many hours the appliance actually runs and turn that into rupees per month.

• Keep a tally card beside the appliance and mark every start and stop time for seven days.
• For a fridge or a pump that cycles on its own, sit with it for a full hour with the watch and count the minutes it actually runs, twice at different times of day.
• Multiply units per hour by hours per week, then by 4.33 weeks, then by your chosen rate, showing every multiplication on the page.
• Set the figure against the whole-house units from step two and write what percentage of the house this one appliance is.

Finish the costed problem statement on one page and choose the single change you will make.

• Write five lines: whose problem this is, units per month, rupees per month, rupees per year at today's tariff, and share of the whole house.
• Draw the seven day-blocks and seven night-blocks as paired bars on squared paper, scale stated, titled with the finding.
• List three candidate changes that cost nothing: a thermostat or fan setting, a timing shift, a door seal pressed back, a bulb already owned, or one habit.
• Choose one and write: 'I will change X for three weeks and expect this appliance to fall from A units per week to B.'
• Read the page aloud to whoever pays the bill and have them sign and date it if they agree to the change.

Write the protocol that makes the after-measurement comparable with the before.

• Write what stays fixed: the same two clock times, the same meter, the same rupee rate, the same appliance, the same tally card.
• Write what is allowed to move: only the one thing you named in step six.
• Rule twenty-one blank dated rows in the notebook before any data exists.
• Write the confounders you already expect: weather, guests, load-shedding hours, a different person using the appliance.
• Write your prediction of the after figure, date it, and have one person witness it with a signature.

Run the change for seven days and read the meter exactly as you read it before.

• Take both daily readings at the same two clock times and write them in ink with the real time.
• Keep the tally card running exactly as it ran in Act 1.
• Repeat the isolation test once under the new setting and record units per hour again.
• Write load-shedding and generator hours daily, and mark the week as disturbed if those hours moved sharply against the baseline week.
• Write down every slip: a missed reading, a night the change was forgotten, a guest who reversed it.

Repeat two more weeks, then check your finding against the next printed bill.

• Take the same two readings daily for a further fourteen days without altering anything else.
• Add the three after-weeks and divide by three, longhand, then set the mean against the baseline week.
• When the next bill arrives, compare its billed units with your own arithmetic and write out any gap and the reason for it.
• Draw four bars on squared paper to one stated scale: baseline, week one, week two, week three, labelled in units and in rupees.
• Write the confounder paragraph: temperature, load-shedding hours, people in the house, and anything else that moved.

Publish the finding, defend it to real people, and record what the household will keep.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read the one-page explainer to three neighbouring households on the same tariff and write down what each of them said.
• Hand or post the chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask whoever pays the bill, in front of a witness, whether the change stays and why, and write the answer word for word.
• Complete the relations ledger with ten names, roles, dates and replies.

Pick one small, self-contained, neutral thing you can teach completely in ten minutes, and write the script.

• List five candidate topics you could teach fully in ten minutes: a knot, the rules of a card game, how a lock works, the parts of a bicycle brake, five words of a language nobody in the room speaks.
• Strike out any topic that appears on somebody's syllabus this term, and any topic touching belief, politics, family or money.
• Write the ten-minute script word for word in Google Docs, including the examples and the exact order, so the second group hears the identical lesson.
• Read the script aloud against the phone stopwatch twice, and cut or pad until it lands between nine and eleven minutes both times.
• Write one sentence on why this topic is safe to test: it touches no mark, no belief and nobody's private life.

Build the test once, with its marking key, so it can be given twice without changing a word.

• Write five short questions that can only be answered by somebody who received this lesson, one mark each, with no half marks anywhere.
• Write the marking key beside them, stating exactly what earns a mark and what does not, decided before a single answer exists.
• Build the questions once in Google Forms with a participant-code box at the top, no name field, and the question order fixed.
• Give the form to one person who has not had the lesson; their score tells you how much of your test can simply be guessed.
• Put both test moments in the calendar now: immediately after teaching, and exactly seventy-two hours later to the hour.

Recruit at least five people, take consent in words, and give everybody a code instead of a name.

• Find at least five people who can give you ten minutes now and five more minutes in three days, and ask each one face to face.
• Read the consent wording aloud to each person: this is a study about forgetting, not about you; your score is never shown to anyone; you can stop at any moment and I will destroy your answers.
• Give each person a code, tell them what it is, and write the code-to-name page separately from every other file.
• Record consent as a tick, a code and a date, and note that consent will be asked for again at the second test.
• Write one neutral line per person on things that could affect the result: how much of the topic they already knew, how comfortably they read, whether the lesson is in their first language.

Deliver the ten minutes from the script and test everybody immediately.

• Teach from the script without adding examples, without answering extra questions, and time every session on the stopwatch.
• Hand out the form the moment the lesson ends, with no discussion in between and no talking between participants.
• Watch that each person enters their code rather than their name, and check the responses carry no names before you go further.
• Mark every answer against the key and enter the score against the code in Google Sheets.
• Tell everybody truthfully that there is nothing to revise and that they should carry on as normal.

Give the identical five questions again after exactly three days, unannounced, and measure the drop.

• Wait the full seventy-two hours without reminding anybody, sending the material or discussing the topic in front of them.
• Ask each person again whether they are willing to be tested, and record anybody who declines as a withdrawal, removing their first score.
• Give the identical form, identical questions, identical order, and mark to the same key you wrote in step two.
• Ask each person one question afterwards and write the answer down: have you thought about this or talked about it since.
• Calculate the drop for each code, then the group means, then the second mean as a percentage of the first.

Finish the costed problem statement on one page and choose the single change you will test.

• Write five lines: who was taught, the immediate mean, the three-day mean, the percentage lost, and what that loss is worth in minutes of teaching per week and in hours or rupees per year at the rate teaching is actually paid for here.
• Draw the group's two points as one line in Datawrapper, labelled with the number of people and the exact gap in hours, titled with the finding rather than with the word 'chart'.
• List three possible changes: a two-minute recap at day one, one practice question at day one, the learner explaining the lesson back at the end, a written summary in their own words.
• Choose one and write the sentence: 'The second group gets X and nothing else changes; I expect retention to rise from A per cent to B per cent.'
• Write two lines underneath: what must stay identical for the comparison to hold, and the one difference between the groups you already know you will have to report.

Recruit and consent a second, comparable group of at least five, and record honestly how they differ from the first.

• Recruit at least five people who had no contact with the first group's lesson and who have not heard the topic discussed.
• Read the same consent wording aloud, record consent by code and date, and give this group codes in a separate series so the two are never mixed up.
• Write the same one-line profile for each person: prior knowledge, reading comfort, language of the lesson.
• Set the two group profiles side by side and write down every way the groups are not alike, in a section titled 'limits' rather than buried in a footnote.
• Fix both test moments in the calendar and the exact moment the one change will be delivered.

Deliver the identical ten minutes to the second group and test immediately with the identical questions.

• Read the same script word for word, timed on the same stopwatch, adding nothing and answering no extra questions.
• Deliver the change here if the change belongs at teaching time, such as the learner explaining it back or writing a summary in their own words, and time that separately.
• Hand out the identical form the moment the lesson ends, with codes and no names.
• Mark against the same key, entering scores in the same sheet under the second group's codes.
• Compare the two immediate means and write one line on whether the groups started level; if they did not, say so now rather than at the end.

Deliver the change at the agreed moment, retest at seventy-two hours, and put the two curves side by side.

• Deliver the change exactly as written and at the agreed hour, timing it and recording who received it and who was missed.
• Test at exactly seventy-two hours from the lesson, asking consent again and recording any withdrawal.
• Mark to the same key, then chart both groups in Datawrapper as two lines from immediate to seventy-two hours, each labelled with its number of people.
• State the difference in retained percentage between the groups as a number, and say plainly whether five people per group can settle it.
• List every confounder: different people, a different day of the week, one group taught in a noisier room, somebody who discussed the topic with a friend, a lesson that ran a minute long.

Publish the finding without naming anybody, defend it to real people, and record what you would keep.

• Record a defence video of no more than three minutes: the topic, both groups, the two curves, the one change, the difference and the limits, using codes and never names.
• Show the one-page explainer to three people who teach or train and write down what each of them said.
• Send the two-line chart to the teacher or trainer you interviewed with one question: what would you have controlled that I did not.
• Tell every participant their own result privately if they want it, tell them nobody else has seen it, and destroy the code-to-name page in front of one witness.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick one small, self-contained, neutral thing you can teach completely in ten minutes, and write the script.

• List five candidate topics you could teach fully in ten minutes: a knot, the rules of a card game, how a lock works, the parts of a bicycle brake, five words of a language nobody in the room speaks.
• Strike out any topic that appears on somebody's syllabus this term, and any topic touching belief, politics, family or money.
• Write the ten-minute script word for word in the notebook, including the examples and the exact order, so the second group hears the identical lesson.
• Read the script aloud twice against a clock with a second hand and cut or pad until it lands between nine and eleven minutes both times.
• Write one sentence on why this topic is safe to test: it touches no mark, no belief and nobody's private life.

Build one master test sheet with its marking key, so the identical paper can be given twice.

• Write five short questions that can only be answered by somebody who received this lesson, one mark each, with no half marks anywhere.
• Write the marking key on its own page, stating exactly what earns a mark and what does not, before a single answer exists.
• Copy the master sheet out by hand once per person for the first test and again for the second, with a code box at the top and no name line.
• Give one copy to a person who has not had the lesson; their score tells you how much of your test can simply be guessed.
• Write both test moments on the calendar now: immediately after teaching, and exactly seventy-two hours later to the hour.

Recruit at least five people, take consent in words, and give everybody a code instead of a name.

• Find at least five people who can give you ten minutes now and five more minutes in three days, and ask each one face to face.
• Read the consent sentence at the top of the sheet aloud to each person: this is a study about forgetting, not about you; your score is never shown to anyone; you can stop at any moment and I will destroy your answers.
• Give each person a code, tell them what it is, and write the code-to-name key on a folded page kept apart from the sheets.
• Record consent as a tick, a code and a date, and note that consent will be asked for again at the second test.
• Write one neutral line per person on things that could affect the result: prior knowledge, reading comfort, the language of the lesson.

Deliver the ten minutes from the script and test everybody immediately.

• Teach from the script without adding examples, without answering extra questions, and time every session against the clock.
• Hand out the copied sheets the moment the lesson ends, with no discussion in between and the participants seated apart.
• Check each sheet carries a code and not a name before you collect it, and put the collected sheets straight into the envelope.
• Mark every sheet against the key page and write the score against the code in the notebook, never against a name.
• Tell everybody truthfully that there is nothing to revise and that they should carry on as normal.

Give the identical five questions again after exactly three days, unannounced, and measure the drop.

• Wait the full seventy-two hours without reminding anybody, handing out anything or discussing the topic in front of them.
• Ask each person again whether they are willing, and record anybody who declines as a withdrawal, removing their first score and destroying their sheet.
• Hand out the second set of copies from the same master, identical questions in the identical order, and mark to the same key.
• Ask each person one question afterwards and write the answer down: have you thought about this or talked about it since.
• Work out the drop for each code longhand, then the group means, then the second mean as a percentage of the first.

Finish the costed problem statement on one page and choose the single change you will test.

• Write five lines: who was taught, the immediate mean, the three-day mean, the percentage lost, and what that loss is worth in minutes of teaching per week and in hours or rupees per year at the rate teaching is actually paid for here.
• Draw the group's two points on squared paper to a stated scale, labelled with the number of people and the exact gap in hours, with the finding as the title.
• List three possible changes: a two-minute recap at day one, one practice question at day one, the learner explaining the lesson back at the end, a written summary in their own words.
• Choose one and write: 'The second group gets X and nothing else changes; I expect retention to rise from A per cent to B per cent.'
• Read the page aloud to one person who was not involved, and write down the first thing they did not understand.

Recruit and consent a second, comparable group of at least five, and record honestly how they differ from the first.

• Recruit at least five people who had no contact with the first group's lesson and who have not heard the topic discussed.
• Read the same consent sentence aloud, record consent by code and date, and use a separate letter series so the two groups are never mixed up.
• Write the same one-line profile for each person: prior knowledge, reading comfort, language of the lesson.
• Rule two columns on one page, one group each, and write every way the groups are not alike under a heading that says 'limits'.
• Write both test moments and the exact moment of the change on the calendar, and copy out the sheets for both tests from the same master.

Deliver the identical ten minutes to the second group and test immediately with the identical sheets.

• Read the same script word for word, timed against the same clock, adding nothing and answering no extra questions.
• Deliver the change here if it belongs at teaching time, such as the learner explaining it back or writing a summary in their own words, and time that separately.
• Hand out the copied sheets the moment the lesson ends, collect them into the envelope, and check every sheet carries a code.
• Mark against the same key page and write the scores in the notebook under the second group's codes.
• Compare the two immediate means longhand and write one line on whether the groups started level; if they did not, say so now rather than at the end.

Deliver the change at the agreed moment, retest at seventy-two hours, and draw the two curves side by side.

• Deliver the change exactly as written and at the agreed hour, timing it and recording who received it and who was missed.
• Test at exactly seventy-two hours from the lesson, asking consent again and recording any withdrawal.
• Mark to the same key, then draw both groups on one sheet of squared paper as two lines from immediate to seventy-two hours, each labelled with its number of people.
• Write the difference in retained percentage between the groups as a number, and say plainly whether five people per group can settle it.
• List every confounder: different people, a different day of the week, one group taught in a noisier room, somebody who discussed the topic with a friend, a lesson that ran a minute long.

Publish the finding without naming anybody, defend it to real people, and record what you would keep.

• Deliver the defence out loud in under three minutes to three people who sign and date the page: the topic, both groups, the two curves, the one change, the difference and the limits, using codes and never names.
• Read the one-page explainer to three people who teach or train and write down what each of them said.
• Hand the two-curve page to the teacher or trainer you interviewed with one question: what would you have controlled that I did not.
• Tell every participant their own result privately if they want it, tell them nobody else has seen it, and burn or tear up the code page in front of one witness.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick the one purse you will track for twenty-eight days and get written permission if it is not your own.

• Choose one of three: your own pocket money, a household grocery float, or a small shop's daily till.
• Write one sentence defining an outflow: any money leaving that purse, of any amount, including coins and including money spent for somebody else.
• Where the purse is not yours, ask the holder face to face and write their permission on a dated page they sign, stating plainly that they may stop the log on any day without giving a reason.
• Where it is a shop till, ask the owner for that same permission in writing, and write on the same page that the shop will never be named in anything you publish.
• Photograph the signed page and file it in a dated Google Drive folder, then open a Google Sheets file named 'Where The Money Went'.

Get the guess written down before a single rupee is logged.

• Ask the person who holds the purse to write what they think leaves it in a normal week: one total, then a figure against each of your categories.
• Write your own separate guess in the same shape, even where it is your own purse and your own money.
• Put both in a Google Doc named 'Day one guess', close it, and do not open it again until day fourteen.
• Open File then Version history once, so the file itself carries the creation date rather than your word for it.
• Write today's date and the words 'not to be opened until' plus the day-fourteen date at the top of the log sheet.

Build a form and a sheet that record one outflow in under fifteen seconds.

• Write your category list on one page, between six and ten categories, each one a thing a stranger could sort a purchase into without asking you.
• Build a Google Form with four questions: amount in rupees, category chosen from your list, who decided it as a role rather than a name, and one line on what it was for.
• Link the form to the sheet so every submission lands as a row with the time stamped automatically.
• Save the form to your phone home screen and test it with three invented rows, then delete those three rows before day one begins.
• Add a column headed 'estimated' and tick it for any amount you had to reconstruct rather than see.

Record every outflow from that purse at the moment it happens, for fourteen consecutive days.

• Open the form and submit the row before the money is out of your hand, not later that evening.
• Where you genuinely could not log at the moment, enter it the same day and tick the estimated column.
• Record the decider as a role — me, the purse holder, another member of the household, the shop owner — and never as a name.
• Enter a day with nothing spent as a row saying zero, so that day exists in the count.
• Write one line each night in the failure log: anything missed, anything reconstructed, anything you were told about after the event.

Total the fortnight by category, then open the sealed guess and measure the gap.

• Use SUMIF, or insert a pivot table, to total rupees against each category across the fourteen days, then divide by two for rupees per week.
• Build one bar chart in the chart editor: rupees per week by category, largest bar first, titled with the finding rather than the word 'chart'.
• Only now open the day-one guess and put the guessed figure beside the actual figure for every category, in two columns.
• Calculate the gap in rupees per week for every category and sort by that gap, largest first.
• Optional, and only if you want it: add two columns marking each of your own outflows as needed, chosen or neither, and a note on whether anything carried interest or a late fee. This is your own reflection on your own spending, never a verdict on anybody else's.

Finish the costed problem statement on one page and choose the single change you will make.

• Write four lines: whose purse this is, total rupees out per week, the category with the largest gap between believed and actual, and what that gap comes to per year at this rate.
• Put the guessed bars beside the actual bars on one axis and give the chart the finding as its title.
• List three candidate changes — a written limit on one category, a separate envelope for one category, one category logged where everybody can see it daily, or one habit dropped — with the rupees each might save.
• Choose one and write the sentence: for fourteen days I will change X and expect this category to fall from A to B rupees per week.
• Read the page to the purse holder and record their agreement as a dated note or a voice message before anything changes.

Write the protocol that keeps the second fortnight measurable against the first.

• Write what stays fixed: the same purse, the same categories, the same form, the same at-the-moment rule, the same fourteen days.
• Write what is allowed to move: only the one change you named in step six.
• Write the change itself as an instruction somebody could follow without you in the room, and pin it where the purse lives.
• Add fourteen blank dated rows on a second sheet tab before any data exists.
• Record a thirty-second voice note stating your prediction, and file it in the dated folder.

Run the change and log the first seven days exactly as you logged the baseline.

• Submit every outflow through the same form, at the moment it happens, into the second fortnight's rows.
• Total these seven days by category and compare them against the matching seven days of the baseline rather than against the whole fortnight.
• Write down every occasion the change was not kept, and why, without softening it.
• Ask the purse holder once, on day seven, whether the change is annoying, and write the answer word for word.
• Do not alter the change mid-week; where it turns out to be unworkable, write that down and keep logging.

Finish the second fortnight and compare the two fortnights honestly.

• Log the remaining seven days with nothing altered.
• Total the second fortnight by category and set it beside the first, per week, in the same two-column shape you used for the guess.
• Chart before against after in the chart editor: one pair of bars per category, largest change first.
• List every confounder you can name: a wedding, a festival, a price rise, a guest, an illness, a different pay date, and the fact that logging money tends to reduce it on its own.
• State the change in rupees per week and as a percentage, and say plainly whether twenty-eight days is enough to be sure.

Publish the finding, defend it to real people, and record what the purse holder will keep.

• Record a defence video of no more than three minutes: the purse described without naming anyone, the weekly total, the gap, the change, the result and the limits.
• Check the video and every page for card numbers, account numbers, anybody's income, and the shop's name where a shop was involved, and remove all of it before anything is shared.
• Show the one-page explainer to three people this purse actually pays for, and write down what each of them said.
• Send the chart to the expert you interviewed with one question: what would you have categorised differently.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick the one purse you will track for twenty-eight days and get written permission if it is not your own.

• Choose one of three: your own pocket money, a household grocery float, or a small shop's daily till.
• Write at the top of a fresh page the one sentence that defines an outflow: any money leaving that purse, of any amount, including coins and including money spent for somebody else.
• Where the purse is not yours, ask the holder face to face and have them sign and date a page saying they agree and may stop the log on any day without giving a reason.
• Where it is a shop till, get the owner's permission in writing on that same page, and write on it that the shop will never be named in anything you publish.
• Fold the signed page into the evidence envelope and write the date on the outside.

Get the guess written down before a single rupee is logged.

• Ask the person who holds the purse to write, in their own hand, what they think leaves it in a normal week: one total and a figure against each category.
• Write your own guess on a separate slip in the same shape, even where it is your own purse and your own money.
• Put both slips in an envelope, seal it, and have both of you sign across the flap so it cannot be opened and rewritten.
• Write on the front: today's date, and 'not to be opened until' with the day-fourteen date beside it.
• Hand the envelope to a third person to hold for the fortnight where you think you will be tempted to look.

Build a card system that records one outflow in under fifteen seconds.

• Write your category list on one page, between six and ten categories, each one a thing a stranger could sort a purchase into without asking you.
• Rule a pocket card into five columns: date and time, amount, category, who decided as a role, what it was for.
• Make fourteen cards at once, one per day, each headed with its date, and keep them with a pen where the purse lives.
• Write three invented rows on a spare card to check the columns work in your hand, then tear that card up.
• Add a small box headed 'estimated' on every card, to tick against any amount you reconstructed rather than saw.

Record every outflow from that purse at the moment it happens, for fourteen consecutive days.

• Write the row before the money is out of your hand, not later that evening.
• Where you genuinely could not write it at the moment, write it the same day and tick the estimated box.
• Write the decider as a role — me, the purse holder, another member of the household, the shop owner — and never as a name.
• Write a zero on a day nothing left the purse, so that day exists in the count.
• Each night put the day's card in the dated envelope and write one line on anything missed or reconstructed.

Total the fortnight by category, then open the sealed envelope and measure the gap.

• Add each category across the fourteen cards longhand, twice, and accept a total only when the two runs agree.
• Divide every category total by two for rupees per week, leaving the working visible on the page.
• Only now cut open the sealed envelope, in front of the purse holder, and write the guessed figure beside the actual figure for every category.
• Draw the bars on squared paper to a stated scale: guessed bar and actual bar side by side for each category, largest gap first.
• Optional, and only if you want it: mark each of your own outflows N for needed, C for chosen or X for neither, and write one line on whether anything carried interest or a late fee. This is your own reflection on your own spending, never a verdict on anybody else's.

Finish the costed problem statement on one page and choose the single change you will make.

• Write four lines: whose purse this is, total rupees out per week, the category with the largest gap between believed and actual, and what that gap comes to per year at this rate.
• Draw the paired guessed-against-actual bars on squared paper with the scale stated and the finding as the title.
• List three candidate changes — a written limit on one category, a separate envelope for one category, one category written up where everybody can see it daily, or one habit dropped — with the rupees each might save.
• Choose one and write: for fourteen days I will change X and expect this category to fall from A to B rupees per week.
• Read the page aloud to the purse holder and have them sign and date it where they agree to the change.

Write the protocol that keeps the second fortnight measurable against the first.

• Write what stays fixed: the same purse, the same categories, the same card, the same at-the-moment rule, the same fourteen days.
• Write what is allowed to move: only the one change you named in step six.
• Write the change on a card as an instruction somebody could follow without you in the room, and pin it where the purse lives.
• Rule and date fourteen fresh blank cards before any data exists, and put them with the pen.
• Write your prediction on the page, date it, and have one person witness it with a signature.

Run the change and log the first seven days exactly as you logged the baseline.

• Write every outflow on the day's card, at the moment it happens, using the same five columns.
• Total these seven days by category and compare them against the matching seven days of the baseline rather than against the whole fortnight.
• Write down every occasion the change was not kept, and why, without softening it.
• Ask the purse holder once, on day seven, whether the change is annoying, and write the answer word for word.
• Do not alter the change mid-week; where it turns out to be unworkable, write that down and keep logging.

Finish the second fortnight and compare the two fortnights honestly.

• Log the remaining seven days with nothing altered.
• Total the second fortnight by category longhand, twice, and divide by two for rupees per week.
• Draw the paired bars on squared paper to the same scale you used before: before and after for each category, largest change first.
• List every confounder you can name: a wedding, a festival, a price rise, a guest, an illness, a different pay date, and the fact that logging money tends to reduce it on its own.
• Write the change in rupees per week and as a percentage, and say plainly whether twenty-eight days is enough to be sure.

Publish the finding, defend it to real people, and record what the purse holder will keep.

• Deliver the defence aloud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read every page for card numbers, account numbers, anybody's income, and the shop's name where a shop was involved, and cross all of it out before the pages leave your hand.
• Read the one-page explainer to three people this purse actually pays for, and write down what each of them said.
• Hand or post the bar chart to the expert you interviewed with one question: what would you have categorised differently.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick one real rule in a place you already belong that most people break, and write it as something a stranger could observe.

• List five rules you already live under: at home, in a shared room, on your street, at a ground, in a queue.
• Cross out any rule you cannot watch from a shared or public place, and any that would mean watching one particular person.
• Write the survivor in one sentence: the rule is X, one instance is Y, and an instance counts as followed when Z.
• Find out whether it is written anywhere, and copy the written version into a Google Doc; where it exists only as something people say, write down who says it and when it started, as far as anybody can tell.
• Ask three people who live under it whether they knew it existed, and write down all three answers.

Fix the time, the place, the session length and the counting rule, then run one pilot session.

• Choose one fixed time of day and one fixed shared or public place where instances happen often enough to count.
• Choose a fixed session length between ten and thirty minutes and write it down; every session from now on is exactly that long.
• Build the Google Sheet with one row per session: date, start time, instances followed, instances not followed, rate as a formula.
• Run one pilot session with a paper tally, timed on your phone clock, and mark that row clearly as a pilot so it never enters the ten.
• Write the edge cases the pilot exposed — an instance half-done, two at once, one you could not see clearly — and fix the answer to each before session one.

Run ten identical observation sessions and tally every instance.

• Turn up at the same time and the same place for every session, and start counting when the timer starts, not before.
• Mark one stroke in followed or not followed for every instance, and nothing else — no names, no descriptions, no photographs.
• Enter both totals into the sheet the same day, so the rate calculates itself while you still remember the session.
• Where a session could not happen, write the date and the reason in the failure log rather than moving that session to another time.
• Where anybody asks what you are doing, say plainly that you are counting how often a rule is followed for a project, and stop if they ask you to.

Find out who carries the cost of the breaking, and in what.

• List everybody the breaking affects: whoever clears up, whoever waits, whoever pays, whoever is made less safe.
• Build a five-question Google Form asking what the breaking costs them — minutes lost, rupees spent, something made less safe, something made less fair, and one line in their own words — and collect no names.
• Ask at least five of them, face to face wherever you can, and write their answers into the same spreadsheet.
• Where the cost is minutes, time it yourself once with the phone stopwatch rather than trusting the estimate.
• Write down anybody who says it costs them nothing, because that answer matters as much as the rest.

Turn the compliance rate and the answers into a cost per week.

• Multiply instances not followed per session by the number of sessions in a week, in a visible formula cell, to get instances not followed per week.
• Multiply that by the minutes or the rupees one instance costs, taken from the answers and from your own timing.
• Where the cost is safety or fairness rather than money, write it as a count of instances plus one sentence in the words the affected people used, and do not invent a rupee figure for it.
• Build one bar chart of compliance per session across the ten sessions, with the two-week rate drawn across it as a line.
• Write the limits: how many instances you saw in total, how many you might have missed, and what a different hour of the day would probably have shown.

Finish the costed problem statement and draft the changed rule everybody will be asked to agree to.

• Write four lines: the rule, the compliance rate with the instance count behind it, who carries the cost, and the cost per week.
• List three candidate changes — make the rule easier to follow, move where or when it applies, change what it asks for, or drop it because nobody ever consented to it — with what each would need from whom.
• Draft the changed rule in a Google Doc as one short page: the rule, why it exists, how to change it, how to appeal a decision under it, and the sentence that anybody may leave this group at any time.
• Check the draft against the hard line: it may not punish, shame, exclude, fine or cost anybody money, and it may not name a person. Rewrite anything that does.
• Write the prediction: compliance will move from A per cent to B per cent across two weeks.

Put the draft to everybody it affects, rewrite it until they agree, and get it signed.

• Share the Google Doc in suggestion mode with everybody affected, or read it aloud to anybody who is not online, and ask each of them for one objection.
• Change the rule to answer at least one objection, and keep the comment history showing what changed and why.
• Ask each person to sign and date the final page, and record how many declined as a number, without naming any of them and without pressing anybody for a reason.
• Say aloud, and write on the sheet, that anybody may ask to change it, may appeal a decision under it, and may leave the group at any time.
• Pin the signed sheet where the rule applies and photograph it for the evidence log.

Run five sessions under the new rule, changing nothing about how you count.

• Keep the same time, the same place, the same session length, the same tally sheet and the same counter.
• Count five sessions and enter both totals into the sheet the same day.
• Write down every occasion the rule was not kept as a situation, never as a person.
• Ask two people on day seven whether the new rule is workable, and write their answers word for word without naming them.
• Do not remind anybody of the rule during a session; a count you influenced is not a count.

Finish the ten sessions and compare the two fortnights honestly.

• Run the last five sessions with nothing altered.
• Calculate the second fortnight's rate and set it beside the first, in percentage points, with the total instances behind each.
• Chart the two rates as paired bars in the chart editor, with the number of instances written on each bar.
• Recalculate the cost per week at the new rate and state the difference in minutes or rupees.
• List every confounder: a holiday, an examination week, the weather, a new person in the group, the fact that people knew they were being counted, and the fact that you asked them to sign something.

Publish the finding, defend it to the people it affects, and record what happens to the rule now.

• Record a defence video of no more than three minutes: the rule, the baseline rate, who paid and how much, the changed rule, the after rate and the limits.
• Read every page and watch the video for anything that identifies a person, and take it out before anything is shared.
• Show the one-page explainer to three people the breaking costs, and write down what each of them said.
• Send the chart to the expert you interviewed with one question: what would you have written into the rule that I did not.
• Ask the group on the record whether the rule stays, changes again or ends, and write the answer down as they said it.

Pick one real rule in a place you already belong that most people break, and write it as something a stranger could observe.

• List five rules you already live under: at home, in a shared room, on your street, at a ground, in a queue.
• Cross out any rule you cannot watch from a shared or public place, and any that would mean watching one particular person.
• Write the survivor at the top of a fresh page in one sentence: the rule is X, one instance is Y, and an instance counts as followed when Z.
• Find out whether it is written anywhere, and copy the written version onto the page; where it exists only as something people say, write down who says it and when it started, as far as anybody can tell.
• Ask three people who live under it whether they knew it existed, and write down all three answers.

Fix the time, the place, the session length and the counting rule, then run one pilot session.

• Choose one fixed time of day and one fixed shared or public place where instances happen often enough to count.
• Choose a fixed session length between ten and thirty minutes and write it down; every session from now on is exactly that long.
• Rule twenty tally sheets in advance, each headed with the place, the session length and a blank date, two columns only: followed, not followed.
• Run one pilot session with a watch, and write 'pilot' across that sheet so it never enters the ten.
• Write the edge cases the pilot exposed — an instance half-done, two at once, one you could not see clearly — and fix the answer to each before session one.

Run ten identical observation sessions and tally every instance.

• Turn up at the same time and the same place for every session, and start counting when the watch says so, not before.
• Mark one stroke in followed or not followed for every instance, and nothing else — no names, no descriptions, no sketches of people.
• Total both columns at the foot of the sheet before you leave, and work out the rate longhand in the margin.
• Where a session could not happen, write the date and the reason in the failure log rather than moving that session to another time.
• Where anybody asks what you are doing, say plainly that you are counting how often a rule is followed for a project, and stop if they ask you to.

Find out who carries the cost of the breaking, and in what.

• List everybody the breaking affects: whoever clears up, whoever waits, whoever pays, whoever is made less safe.
• Write five questions on a card — minutes lost, rupees spent, something made less safe, something made less fair, and one line in their own words — and take no names.
• Ask at least five of them face to face and write each answer on its own slip, numbered rather than named.
• Where the cost is minutes, time it yourself once with a watch rather than trusting the estimate.
• Write down anybody who says it costs them nothing, because that answer matters as much as the rest.

Turn the compliance rate and the answers into a cost per week.

• Multiply instances not followed per session by the number of sessions in a week, longhand, to get instances not followed per week.
• Multiply that by the minutes or the rupees one instance costs, taken from the slips and from your own timing, leaving the working visible.
• Where the cost is safety or fairness rather than money, write it as a count of instances plus one sentence in the words the affected people used, and do not invent a rupee figure for it.
• Draw the per-session compliance bars on squared paper to a stated scale, with the two-week rate ruled across as a line.
• Write the limits: how many instances you saw in total, how many you might have missed, and what a different hour of the day would probably have shown.

Finish the costed problem statement and draft the changed rule everybody will be asked to agree to.

• Write four lines: the rule, the compliance rate with the instance count behind it, who carries the cost, and the cost per week.
• List three candidate changes — make the rule easier to follow, move where or when it applies, change what it asks for, or drop it because nobody ever consented to it — with what each would need from whom.
• Draft the changed rule on one page: the rule, why it exists, how to change it, how to appeal a decision under it, and the sentence that anybody may leave this group at any time.
• Check the draft against the hard line: it may not punish, shame, exclude, fine or cost anybody money, and it may not name a person. Rewrite anything that does.
• Write the prediction at the foot of the page and date it: compliance will move from A per cent to B per cent across two weeks.

Put the draft to everybody it affects, rewrite it until they agree, and get it signed.

• Read the draft aloud to everybody affected, in one sitting where you can, and ask each of them for one objection.
• Write every objection on the back of the page, then rewrite the rule to answer at least one of them and read the new version aloud.
• Copy the final version out cleanly with a signature and date line for each person, and record how many declined as a number, without naming any of them and without pressing anybody for a reason.
• Say aloud, and write on the sheet, that anybody may ask to change it, may appeal a decision under it, and may leave the group at any time.
• Pin the signed sheet where the rule applies and keep the drafts, objections and all, in the envelope.

Run five sessions under the new rule, changing nothing about how you count.

• Keep the same time, the same place, the same session length, the same headed sheets and the same counter.
• Count five sessions and total both columns at the foot of each sheet before you leave.
• Write down every occasion the rule was not kept as a situation, never as a person.
• Ask two people on day seven whether the new rule is workable, and write their answers word for word without naming them.
• Do not remind anybody of the rule during a session; a count you influenced is not a count.

Finish the ten sessions and compare the two fortnights honestly.

• Run the last five sessions with nothing altered.
• Add the ten sheets, work out the second fortnight's rate longhand, and set it beside the first in percentage points with the instance count behind each.
• Draw the two rates as a pair of bars on squared paper to a stated scale, writing the instance count on each bar.
• Recalculate the cost per week at the new rate and write the difference in minutes or rupees.
• List every confounder: a holiday, an examination week, the weather, a new person in the group, the fact that people knew they were being counted, and the fact that you asked them to sign something.

Publish the finding, defend it to the people it affects, and record what happens to the rule now.

• Deliver the defence aloud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read every page for anything that identifies a person and cross it out before the pages leave your hand.
• Read the one-page explainer to three people the breaking costs, and write down what each of them said.
• Hand or post the bar chart to the expert you interviewed with one question: what would you have written into the rule that I did not.
• Ask the group in front of a witness whether the rule stays, changes again or ends, and write the answer down word for word.

Log every forwarded message, voice note, image and clip that reaches your family and neighbourhood groups for seven days.

• Open a Google Sheets file named 'The Lie This Street Believes' with columns: date, group code, format, the claim in one sentence, times seen.
• Give every group a code letter and every sender nothing at all, so no name, number or profile picture ever enters the file.
• Type each claim out in your own words in one sentence rather than screenshotting it, and record the format — text, voice note, image, video, link — because the format decides how you check it.
• When the same claim arrives again, add one to the times-seen count instead of opening a new row.
• Where you must keep an image to check later, save it to a dated Drive folder cropped so the sender's name, number and picture are outside the frame.

Choose the single claim that appeared most often and write it as one sentence that can be shown true or false.

• Sort the sheet by times seen, largest first, and read the top five aloud.
• Strike out anything about religion, politics, a sect, an ethnic group or a named person, and write 'out of bounds' beside each one.
• From what is left, take the most practical and local: a remedy, a price, a scam, a claim about a school, a water supply or an electricity bill.
• Rewrite it with no hedging, in the words people actually use, so that evidence can settle it — for example 'boiling this cures that', or 'the bill rises if you do this'.
• Write one line on what somebody would do, spend or avoid if they believed it, because that is what you will cost later.

Test the claim against three sources that did not copy one another, and record exactly how each check was done.

• Search Google Fact Check Explorer for the claim in the words people use, then again in the other language it circulates in, and record every verdict with its publisher and date.
• Search AFP Fact Check for the same claim, then follow each result back to the material it cites — if two of your sources meet at the same original post, they count as one.
• For a photograph or a video frame, run it through Google Images reverse image search, sort by oldest, and write down the earliest date and place the picture appears.
• Go to one source with direct authority: a pharmacist or a doctor for a remedy, the utility's own published tariff for a bill claim, the school office for a school claim, and write down what they said and on what date.
• Fill a check table in Google Docs with four columns: source, what it says, how you checked it, date checked.

Put the same neutrally worded question to twenty people and record what each of them believes.

• Write the question once and never change a word: 'Have you heard that [claim]? Do you think that is true, false, or are you not sure?'
• Practise saying it flat, with no tone and no expression, on two people first, and leave those two out of the count.
• Ask each of the twenty on their own, so nobody hears anybody else's answer before giving their own.
• Record every answer in Sheets as true, false or not sure against a code number, never a name, and count anybody who declines in a separate refusal column.
• Ask one follow-up only — 'where did you first hear it' — and write the reply in their own words.

Find out what believing this claim has already cost the people who believe it.

• Go back to those who answered 'true' and ask what they did about it: bought something, avoided something, stopped taking a medicine, changed shop, warned a relative.
• Put a rupee figure on anything bought or avoided, taken from a price you check yourself at the shop or the pharmacy counter, not from memory.
• Record any harm that is not money in one line each and with no names: a treatment delayed, a shop or a person blamed, an argument in the family.
• Multiply the average cost by the number who answered true to get a cost across the twenty, and label it plainly as an estimate from a small sample.
• Where a believer says the claim cost them nothing, record that as a zero rather than leaving the cell empty.

Finish the costed problem statement on one page and choose exactly what your sixty-second correction will be.

• Write four lines: the claim in its own words, the verdict with its three checks, the percentage of twenty who say it is true, and what it has cost so far.
• Build one chart in Sheets of believe, do not believe and not sure, titled with the finding rather than with the word chart, and print the refusal count underneath it.
• Choose the format the same people actually use: a voice note if the claim arrived as a voice note, a card if it arrived as an image, a short video if it arrived as a clip.
• Write the prediction as one sentence: 'I will share a sixty-second correction with the same twenty people and expect belief to fall from A per cent to B per cent in seven days.'
• Show the page to one person who believed the claim, ask what would make them stop forwarding it, and write their answer down word for word.

Build a correction of sixty seconds or less that is polite, sourced, short and easy to forward.

• Script it in this order: what the message says, what is actually true, the one source that shows it, and what to do instead. Read it aloud against a watch and cut until it fits sixty seconds.
• Correct the message and never the sender: write 'this message is wrong' and never 'whoever sent this is lying'. Name nobody.
• Build the card or the sixty-second video in Canva on the free plan, in the language the claim circulates in, with the source printed on the final frame.
• Show the draft to two people first — one who believed the claim and one who did not — and write down every word each of them changed.
• Check it reads and plays on a small phone screen with the sound off before it goes anywhere.

Send the correction into the same groups the claim came from and log exactly where it went.

• Post it in each group it belongs in with one polite sentence of your own and no accusation in it.
• Send it directly to each of the twenty people you asked, one at a time, so that the sharing and the count cover the same people.
• Log the date, the group code, roughly how many people that group holds, and anything said back, in a fresh tab of the sheet.
• Ask three people to forward it on, and record whether they actually did.
• If any group turns hostile, stop posting there immediately, leave the argument alone, and write down what was said and what you did about it.

Seven days later, put the identical question to the same twenty people and compare.

• Wait a full seven days from the day you shared it, and record the exact date of each re-asking.
• Read the question from the same page, word for word, exactly as it was asked the first time.
• Ask 'did you see the correction' only after they have answered, never before it.
• Match every answer to its code number, and record anybody you cannot reach as unreachable rather than assuming they did not change.
• Chart before against after in Sheets as two bars for the percentage saying true, and state the change in percentage points and in whole people.

Publish the finding, defend it to real people, and record honestly what the correction did and did not do.

• Record a defence video of no more than three minutes: the claim, the three checks, the before percentage, the correction, the after percentage and the limits.
• Show the one-page explainer to three people the claim affected and write down what each of them said.
• Send the before-and-after chart to the expert you interviewed with one question: what would you have checked that I did not.
• State plainly whether the correction moved anybody, including if it moved nobody, and name every reason it might not have.
• Complete the relations ledger with ten entries — names for the expert, the peers and the people you shared with, code numbers for anybody whose belief you recorded.

Log every claim you hear repeated for seven days, in conversation, at the shop, at the gate, and every message read aloud to you.

• Rule five columns across a fresh page: date, group code, format, the claim in one sentence, times heard.
• Give every group and every gathering a code letter, and write no name, no house and no telephone number anywhere on the page.
• Write each claim in your own words in one sentence, and mark how it arrived — said aloud, read aloud from a phone, a picture shown to you, a card handed over.
• When the same claim comes round again, add a stroke to the times-heard tally rather than starting a new row.
• Carry the notebook with you and write it down the same hour you hear it, because a claim recalled at night has already changed shape.

Choose the single claim heard most often and write it as one sentence that can be shown true or false.

• Count the tallies, list the top five on a clean page, and read them aloud.
• Cross out anything about religion, politics, a sect, an ethnic group or a named person, and write 'out of bounds' beside each.
• From the rest, take the most practical and local: a remedy, a price, a scam, a claim about a school, a water supply or an electricity bill.
• Rewrite it with no hedging, in the words people actually use, so evidence can settle it either way.
• Write one line on what somebody would do, spend or avoid if they believed it.

Test the claim against three sources that did not copy one another, and write down exactly how each check was done.

• Go to a person with direct authority — a pharmacist or a doctor for a remedy, the utility office for a bill claim, the school office for a school claim — and write their answer and the date at the doorway before you forget it.
• Get a printed source into your hand: a medicine leaflet, the published tariff sheet, a public notice, a newspaper cutting from a library, and copy the exact line and its date into the notebook.
• Trace the chain: ask the person who repeated it who they heard it from, then ask that person the same, until the chain stops, and write where it stopped.
• Rule a check table with four columns — source, what it says, how you checked it, date checked — and fill one row per check.
• If two of your three checks turn out to come from the same origin, say so and find a third that does not.

Put the same neutrally worded question to twenty people and record what each of them believes.

• Write the question once on a card: 'Have you heard that [claim]? Do you think that is true, false, or are you not sure?' and read it from the card every single time.
• Practise it flat, with no tone, on two people first, and leave those two out of the count.
• Ask each of the twenty alone, and write the answer on a numbered slip — true, false or not sure — with no name on the slip.
• Keep a separate refusal tally, and thank anybody who declines without asking twice.
• Ask one follow-up only — 'where did you first hear it' — and write the reply in their own words on the same slip.

Find out what believing this claim has already cost the people who believe it.

• Return to the slips marked true, find those people again, and ask what they did about it: bought, avoided, stopped a medicine, changed shop, warned a relative.
• Walk to the shop or the pharmacy counter and write down the real price of anything bought or avoided, rather than trusting a remembered figure.
• Write any harm that is not money in one line each with no names: a treatment delayed, a shop or a person blamed, an argument in the family.
• Add the costs, divide by the number who answered true, and write the total across the twenty with the words 'estimate from a small sample' beside it.
• Where a believer says it cost them nothing, write a zero rather than leaving the line blank.

Finish the costed problem statement on one page and choose exactly what your sixty-second correction will be.

• Write four lines: the claim in its own words, the verdict with its three checks, the percentage of twenty who say it is true, and what it has cost so far.
• Draw the belief chart on squared paper — believe, do not believe, not sure — to a stated scale, with the refusal count written underneath.
• Choose the format that reaches these people: sixty seconds said out loud face to face, a hand-copied card, or both together.
• Write the prediction: 'I will give a sixty-second correction to the same twenty people and expect belief to fall from A per cent to B per cent in seven days.'
• Read the page to one person who believed the claim, ask what would make them stop repeating it, and write the answer word for word.

Build a correction of sixty seconds or less that is polite, sourced, short and easy to repeat.

• Write it in this order: what the message says, what is actually true, the one source that shows it, and what to do instead.
• Read it aloud against a watch with a second hand and cut words until it fits inside sixty seconds every time.
• Correct the message and never the sender: say 'this message is wrong', never 'whoever sent this is lying', and name nobody.
• Copy it onto plain cards by hand, or take one card to a photocopy shop, with the source written on the back of every copy.
• Say it to two people first — one who believed the claim, one who did not — and write down every word each of them changed.

Give the correction to the same people and the same gatherings the claim came from, and log where it went.

• Say the sixty seconds face to face to each of the twenty, one at a time, and hand over a card afterwards.
• Deliver it once in each gathering it belongs in, with one polite sentence of your own and no accusation in it.
• Log the date, the group code, roughly how many people heard it, and anything said back, in a ruled table.
• Ask three people to pass the card on, and record whether they actually did.
• If any gathering turns hostile, stop at once, do not argue, leave, and write down what was said and what you did.

Seven days later, put the identical question to the same twenty people and compare.

• Wait a full seven days from delivery, and write the exact date beside each re-asking.
• Read the question from the same card, word for word, exactly as it was asked the first time.
• Ask 'did you hear the correction' only after they have answered, never before.
• Write each answer on the slip that already carries that person's number, so before and after sit on one slip.
• Draw two bars on squared paper for the percentage saying true, before and after, and write the change in percentage points and in whole people underneath.

Publish the finding, defend it to real people, and record honestly what the correction did and did not do.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read the one-page explainer to three people the claim affected and write down what each of them said.
• Hand the before-and-after chart to the expert you interviewed with one question: what would you have checked that I did not.
• Write plainly whether the correction moved anybody, including if it moved nobody, and name every reason it might not have.
• Complete the relations ledger with ten entries — names for the expert, the peers and the people you shared with, numbers for anybody whose belief you recorded.

Pick one small, harmless, everyday habit of your own and check it against the exclusion list before anything else.

• Write down five ordinary things you do without deciding to: checking the phone, snoozing the alarm, opening a new tab, checking the lock, picking up the remote, saying one filler word.
• Strike out anything touching food, eating, weight, appearance, sleep restriction, medication, self-harm, or anything you feel bad or worried about, and strike out anything that is somebody else's habit.
• From what survives, choose the one that happens often enough to count and matters little enough to log calmly, and write one sentence on why you chose it.
• Write the safety line at the top of your working document: I log only my own habit, I name nobody, and if the logging feels bad I stop and change habit.
• Guess, in writing and dated, how many times a day it happens, so your guess is on record before the data exists.

Turn the habit into an event so clearly defined that two people watching would count the same number.

• Write the definition as one sentence a stranger could apply: for example, one occurrence begins when the screen lights in my hand and ends when it goes dark, and a second unlock within thirty seconds counts as the same occurrence.
• Write the boundary rules that stop double counting: the minimum gap between two occurrences, and what happens if the habit is interrupted halfway.
• Build a Google Form with four questions: where I am, who is present as alone or one other or two or more, what happened in the minute before as a short free-text line, and how the occurrence ended.
• Save the form to your phone home screen as an icon and open it once to time yourself filling it in.
• Test the definition for one hour on an ordinary afternoon, count out loud, and rewrite any rule that turned out to be ambiguous.

Record every occurrence for seven consecutive days, within a minute of it happening.

• Tap the form the moment the occurrence ends, before you do anything else, and never at the end of the day from memory.
• In the free-text box, write what physically happened in the minute before in five words or fewer: phone buzzed, walked into kitchen, finished a task, sat down, conversation stopped.
• Each night open the response sheet, count the day's rows, and type one line saying how complete the day was and roughly what you missed.
• On any day you know you missed a run of occurrences, mark that whole day as partial rather than repairing it from memory.
• Do not try to change the habit this week; if you notice yourself doing it less because of the logging, write that down as its own line.

Log seven more days the same way and time a sample of occurrences so the count can be turned into minutes.

• Keep logging identically; do not add or remove a form question mid-fortnight, and if you must, log the change with its date and time.
• On three of the seven days, start the phone stopwatch at the beginning of ten occurrences and record the seconds each one lasted.
• Take the mean of those timed occurrences and write it as seconds per occurrence, with the number of occurrences it was based on beside it.
• Compare the count on the two weekends with the count on the ten weekdays and note the difference in one line.
• Check the sheet has fourteen dated days with no gaps, and mark any missing day as missing rather than deleting it.

Count what came immediately before, and find the single cue present in most occurrences.

• Read every entry in the before-column and sort them into no more than six named categories, writing the rule you used to sort them.
• Build a pivot table counting occurrences per category, and write each category's share of the total as a percentage.
• Do the same for place, for hour of day, and for whether you were alone or with others, so you know which of the four columns actually separates the data.
• Draw the occurrences-by-cue bar chart in the spreadsheet chart editor, largest bar first, titled with the finding.
• Draw the fourteen-day daily count as a line, and write one sentence on whether the count drifted downwards simply because you were logging it.

Finish the costed problem statement on one page and choose the one cue you will remove or change.

• Write four lines: the habit as defined, occurrences per day averaged over fourteen days, minutes per day and hours per year from your timed sample, and the top cue with its percentage.
• Put the cue bar chart and the fourteen-day line on the same page, with the sample size for the timing printed under them.
• List three ways the top cue could be removed or changed physically rather than by willpower: move an object, silence a notification, change a route, change what sits on a surface, change the order of two things.
• Choose one and write the sentence: 'For fourteen days I will change X and expect the daily count to fall from A to B.'
• Compare that prediction with the guess you dated in step one, and write one line on how wrong the guess was.

Write down what must stay identical across the second fortnight so the two halves can be compared at all.

• Write what is fixed: same definition, same boundary rules, same four form questions, same within-a-minute recording rule, same fourteen-day length.
• Write what is allowed to move: only the one cue named in step six.
• Make the cue change physically today and photograph the before and after of whatever you moved, silenced or rearranged.
• Add fourteen blank dated rows to the sheet before any data exists, and mark the change day on the chart axis now.
• Type your prediction into the sheet with today's date and time, and leave it visible for the whole fortnight.

Log the first seven days after the cue change exactly as you logged the baseline.

• Tap the same form within a minute of every occurrence, with the same four questions and the same categories.
• Keep writing the before-column even though you think you know the answer, because the interesting result is a new cue replacing the old one.
• Each night write one line on whether the cue change held, was partly held, or was undone, and why.
• If you undid the change on any day, log that day normally and mark it as cue restored rather than dropping it.
• Stop and write it down if the logging starts to feel bad; changing habit at this point is a recorded finding, not a failed project.

Log the last seven days, then set twenty-eight days side by side and report the difference honestly.

• Finish days eight to fourteen, then calculate the mean occurrences per day for each fortnight in a formula cell.
• Chart all twenty-eight daily counts as one line with a vertical mark on the change day, so a reader sees the shape rather than only two numbers.
• Rebuild the cue bar chart for the second fortnight and set it beside the first, to show whether the old cue vanished and something else took its place.
• List every confounder by name: exam weeks, travel, illness, a holiday, a new phone, weather, a changed timetable, and the plain fact that you were watching yourself.
• State the change as occurrences per day, as a percentage and as minutes per week, and write one sentence on whether fourteen days either side is enough to be sure.

Publish the finding, defend it to real people, and record what you will keep.

• Record a defence video of no more than three minutes: the definition, the count, the cue, the change, the twenty-eight-day line, the result and the limits, naming nobody.
• Show the one-page explainer to three people and write down what each of them said, including anyone who did not believe the count.
• Send the twenty-eight-day chart to the expert you interviewed with one question: what would you have recorded that I did not.
• Write one paragraph on whether you are keeping the cue change, and why, without turning it into advice for anybody else.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick one small, harmless, everyday habit of your own and check it against the exclusion list before anything else.

• Write down five ordinary things you do without deciding to: checking the phone, snoozing the alarm, opening a new tab, checking the lock, picking up the remote, saying one filler word.
• Strike out anything touching food, eating, weight, appearance, sleep restriction, medication, self-harm, or anything you feel bad or worried about, and strike out anything that is somebody else's habit.
• From what survives, choose the one that happens often enough to count and matters little enough to log calmly, and write one sentence on why you chose it.
• Write the safety line at the top of the notebook: I log only my own habit, I name nobody, and if the logging feels bad I stop and change habit.
• Write and date your guess at how many times a day it happens, and have one person witness the guess with a signature.

Turn the habit into an event so clearly defined that two people watching would count the same number.

• Write the definition as one sentence a stranger could apply, including where the occurrence starts and where it ends.
• Write the boundary rules that stop double counting: the minimum gap between two occurrences, and what happens if the habit is interrupted halfway.
• Rule a pocket card into four columns — time, place, alone or one other or two or more, what happened in the minute before — and make seven of them, one per day.
• Read the definition aloud to one person, describe three borderline events, and see whether they count them the way you do.
• Test the card for one hour on an ordinary afternoon and rewrite any rule that turned out to be ambiguous.

Record every occurrence for seven consecutive days, within a minute of it happening.

• Keep the card and pen in the same pocket every day and mark the row the moment the occurrence ends, never at the end of the day from memory.
• In the fourth column write what physically happened in the minute before in five words or fewer: phone buzzed, walked into kitchen, finished a task, sat down, conversation stopped.
• Each night count the rows on the card, write the day's total in the calendar box, and add one line on how complete the day was.
• Mark any day you know you missed a run of occurrences as partial, rather than repairing it from memory.
• Put each finished card in the envelope with its date on it, and start a fresh card each morning.

Log seven more days the same way and time a sample of occurrences so the count can be turned into minutes.

• Keep the card design identical; if you must change a column, write the change on the card with its date and hour.
• On three of the seven days, use a watch or timer on the next ten occurrences that happen and write the seconds each one lasted on the back of the card.
• Add the ten timings and divide by ten longhand to get seconds per occurrence, writing the sample size beside it.
• Add the two weekend totals and compare them with the ten weekday totals in one written line.
• Check the calendar has fourteen filled boxes with no gaps, and write missing in any box that has no card behind it.

Count what came immediately before, and find the single cue present in most occurrences.

• Lay the fourteen cards out and read every entry in the fourth column aloud, sorting them into no more than six named categories with the sorting rule written down.
• Tally each category with five-bar gates on a fresh sheet, add the tallies, and write each category's share of the total as a percentage.
• Tally place, hour of day, and whether you were alone or with others the same way, so you know which column actually separates the data.
• Draw the occurrences-by-cue bars on squared paper, largest bar first, to a stated scale, titled with the finding.
• Draw the fourteen daily totals as a line and write one sentence on whether the count drifted downwards simply because you were logging it.

Finish the costed problem statement on one page and choose the one cue you will remove or change.

• Write four lines: the habit as defined, occurrences per day averaged over fourteen days, minutes per day and hours per year from your timed sample, and the top cue with its percentage.
• Copy the cue bar chart and the fourteen-day line onto the same page, with the timing sample size written underneath.
• List three ways the top cue could be removed or changed physically rather than by willpower: move an object, silence a notification, change a route, change what sits on a surface, change the order of two things.
• Choose one and write: 'For fourteen days I will change X and expect the daily count to fall from A to B.'
• Read your step-one guess back and write one line on how wrong it was.

Write down what must stay identical across the second fortnight so the two halves can be compared at all.

• Write what is fixed: same definition, same boundary rules, same four card columns, same within-a-minute rule, same fourteen-day length.
• Write what is allowed to move: only the one cue named in step six.
• Make the cue change physically today and draw or describe on the page exactly what you moved, silenced or rearranged, with the time you did it.
• Rule fourteen fresh cards and fourteen calendar boxes before any data exists, and draw the change line on the calendar now.
• Write your prediction at the foot of the page, date it, and have one person witness it with a signature.

Log the first seven days after the cue change exactly as you logged the baseline.

• Mark the same four columns within a minute of every occurrence, using the same categories.
• Keep writing the fourth column even though you think you know the answer, because the interesting result is a new cue replacing the old one.
• Each night write one line on whether the cue change held, was partly held, or was undone, and why.
• If you undid the change on any day, log that day normally and write cue restored across the top of that card.
• Stop and write it down if the logging starts to feel bad; changing habit at this point is a recorded finding, not a failed project.

Log the last seven days, then set twenty-eight days side by side and report the difference honestly.

• Finish days eight to fourteen, then add each fortnight's totals and divide by fourteen longhand to get the two means.
• Draw all twenty-eight daily totals as one line on squared paper to a stated scale, with a vertical mark on the change day.
• Tally the second fortnight's cues the same way and draw the second bar chart beside the first, to show whether the old cue vanished and something else took its place.
• List every confounder by name: exam weeks, travel, illness, a holiday, weather, a changed timetable, and the plain fact that you were watching yourself.
• Write the change as occurrences per day, as a percentage and as minutes per week, and add one sentence on whether fourteen days either side is enough to be sure.

Publish the finding, defend it to real people, and record what you will keep.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone, naming nobody.
• Read the one-page explainer to three people and write down what each of them said, including anyone who did not believe the count.
• Hand or post the twenty-eight-day chart to the expert you interviewed with one question: what would you have recorded that I did not.
• Write one paragraph on whether you are keeping the cue change, and why, without turning it into advice for anybody else.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Choose one stain that keeps coming back in this house and start counting how often it happens.

• Ask whoever does the washing which stain returns most — cooking oil, tea, ink, turmeric or grease — and write their answer down word for word.
• Count how many garments or cloths carry that stain right now and photograph them together beside a ruler.
• Start a fourteen-day household stain log with columns: date, item stained, product used, quantity used, rupees, minutes spent, result.
• Photograph the shelf of every cleaning product already in the house with the labels facing the camera, and write down each pack price and each pack quantity.
• Open a Google Sheets file named 'The Cheapest Clean' holding the stain log on one tab and the product list on another.

Cut one cloth into five identical patches and put an identical stain on every one.

• Take one old cotton item nobody needs and cut five squares of the same size from the same part of it, so weave and colour match across all five.
• Apply the stain with the same tool every time — one level quarter-teaspoon, or three drops from the same dropper — to the centre of each square.
• Stain all five within the same five minutes, so every patch is the same age when it is treated.
• Lay the five flat on a covered surface, label them one to five in pencil on the back so the front gives nothing away, and leave them the same number of hours to dry.
• Photograph all five together beside a ruler with the lamp already in the fixed position you will use for every later photograph.

Write the protocol that holds everything still except the one method under test.

• Write the fixed conditions down before anything is mixed: same water volume in millilitres, same water temperature from the same tap, same contact time in minutes, same scrub count, same brush.
• Write the five-point scale out in words — five for no stain visible, four for a faint shadow, three for clearly reduced, two for barely changed, one for unchanged — and stick it on the wall.
• Set the photograph rig: same lamp, same distance, same background sheet, same time of day, no flash, and mark the phone position on the floor with tape.
• Read every hazard label aloud with an adult present and agree the rule out loud: one product per bowl, never two, and nothing heated.
• List the five methods, one per row, each with the exact dose it will be given.

Treat the five patches under identical conditions, one method each, and photograph the result.

• Set out five separate bowls, one per method, with gloves on, a window open and an adult in the room before anything is opened.
• Time every treatment on the same timer to the same contact time, and count the scrubs aloud so the count is on the recording.
• Rinse each patch with the same volume of water from the same tap and hang all five to dry in the same place for the same hours.
• Photograph the five dried patches in the fixed rig, together and then one by one, and file them in a dated Google Drive folder.
• Have three people score every patch against the wall scale with the labels hidden, and record all three scores separately rather than agreeing one.

Work out what one use of each method actually costs in rupees.

• Weigh or measure each product before and after its test, or count the drops or level spoons used, and record the quantity consumed per test.
• Divide the pack price by the pack quantity in the same units, in a formula cell, to get rupees per millilitre or rupees per gram.
• Multiply by the quantity used to get rupees per use, then add water and any electricity as one line marked clearly as an estimate with the reason beside it.
• Open each patch photograph in Photopea and read the grey value at the centre of the stain with the eyedropper, recording it beside the human scores as a second, independent measure.
• Chart rupees per use against mean score in Datawrapper, one labelled point per method, titled with the finding rather than the word 'chart'.

Finish the costed problem statement and choose the method the household will switch to.

• Write four lines: whose problem this is, how often the stain happens, what the current method costs per use and per year at that frequency, and what the winner would cost instead.
• Multiply rupees per use by the stain count from the fourteen-day log, scale it to a year in a formula cell, and write the difference as one rupee figure.
• Lay the five patches, or their photographs, in a row on the page with the five cost figures printed under them.
• Choose the winner on cost and score together and write: 'The household will use X for one month, and I expect rupees per week to fall from A to B and minutes per week from C to D.'
• Show the page to whoever does the washing and record their agreement, or their refusal, as a dated note or a voice message.

Write the rules that keep the month's measurement identical to the fortnight's.

• Write what stays fixed: same log columns, same scale, same timer, same scorer, same tap, same brush.
• Write what is allowed to move: only the product used to treat the stain.
• Set the winning product aside in one container, weigh it or mark its level on day one, and photograph that starting mark.
• Add four blank labelled rows to the sheet for weeks one to four, with the dates beside them, before any data exists.
• Record a thirty-second voice note stating your prediction, dated before the month begins.

Run the household on the winning method for one week and log every stain event the day it happens.

• Log each event on the day: item, product, quantity used, minutes spent scrubbing, and the score after treatment.
• Weigh or mark the container at the end of the week and compare it against the sum of the logged doses.
• Photograph one treated item in the same fixed rig, so the month has a visual record beside the patch photographs.
• Write down what went wrong: a forgotten entry, somebody reaching for the old bottle, a stain treated twice, an event nobody told you about.

Finish the month, normalise both periods and compare them honestly.

• Keep logging every stain event for three more weeks, changing nothing the protocol does not allow.
• Divide the baseline fortnight and the switched month each by their number of weeks, in formula cells, to get rupees per week and minutes per week.
• Chart the two periods in Datawrapper: two bars for rupees per week, two bars for minutes per week, on stated axes.
• List every confounder you can name: fewer stains this month, a cook away from home, a price change, warmer water, somebody scrubbing harder than before.
• State the change as a rupee figure, as a percentage and in minutes, and say plainly whether one month is enough to be sure.

Publish the finding, defend it to real people, and record what the household will keep.

• Record a defence video of no more than three minutes: the stain, the controls, the five patches, the costs per use, the switch, the month and the limits.
• Show the one-page explainer to three people who wash clothes in this house or nearby and write down what each of them said.
• Send the cost-against-score chart to the expert you interviewed with one question: what would you have controlled that I did not.
• Ask the household on camera whether the switch stays and why, and let them answer without prompting.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Choose one stain that keeps coming back in this house and start counting how often it happens.

• Ask whoever does the washing which stain returns most — cooking oil, tea, ink, turmeric or grease — and write their answer down word for word.
• Count how many garments or cloths carry that stain right now and draw or list them on the first page.
• Rule a fourteen-day household stain log with columns: date, item stained, product used, quantity used, rupees, minutes spent, result.
• Walk to the shelf and copy every cleaning product's name, pack price and pack quantity off its label into the notebook.
• Write the household's usual method for this stain at the top of the page, in the words of the person who does it.

Cut one cloth into five identical patches and put an identical stain on every one.

• Take one old cotton item nobody needs and cut five squares of the same size from the same part of it, so weave and colour match across all five.
• Apply the stain with the same tool every time — one level quarter-teaspoon, or three drops from the same dropper — to the centre of each square.
• Stain all five within the same five minutes, so every patch is the same age when it is treated.
• Lay the five flat on a covered surface, label them one to five in pencil on the back only, and leave them the same number of hours to dry.
• Lay tracing paper over each dried patch and trace the outline of the stain, then count the squares it covers on squared paper and write that count down.

Write the protocol that holds everything still except the one method under test.

• Write the fixed conditions down before anything is mixed: same water volume in the jug, same water from the same tap, same contact time, same scrub count, same brush.
• Draw the five-point scale by hand: five shaded reference squares running from unchanged to no stain visible, each labelled in words, and pin it on the wall.
• Write how a patch gets scored: three people look at the front only, with the pencil labels turned away, and each writes a number without speaking.
• Read every hazard label aloud with an adult present and agree the rule out loud: one product per bowl, never two, and nothing heated.
• List the five methods, one per row, each with the exact dose it will be given.

Treat the five patches under identical conditions, one method each, and record the result twice over.

• Set out five separate bowls, one per method, with gloves on, a window open and an adult in the room before anything is opened.
• Time every treatment on the same watch to the same contact time, and count the scrubs aloud so somebody else can hear the count.
• Rinse each patch with the same jugful from the same tap and hang all five to dry in the same place for the same hours.
• Trace each dried patch again on fresh tracing paper and count the squares the remaining stain covers, giving a measured area beside the human score.
• Have three people score every patch against the drawn wall scale with the labels turned away, and write all three scores separately rather than agreeing one.

Work out what one use of each method actually costs in rupees.

• Measure each product before and after its test with the kitchen scale, the jug or a counted number of drops or level spoons, and record the quantity consumed.
• Divide the pack price by the pack quantity longhand in the margin, leaving the working visible, to get rupees per millilitre or per gram.
• Multiply by the dose used to get rupees per use, then add water and fuel as one line marked as an estimate with the reason beside it.
• Draw the result on squared paper: rupees per use up the side, mean score along the bottom, one labelled point per method, with the scale stated.
• Underline the cheapest method that scored four or five, and underline twice the most expensive method that scored two or below.

Finish the costed problem statement and choose the method the household will switch to.

• Write four lines: whose problem this is, how often the stain happens, what the current method costs per use and per year at that frequency, and what the winner would cost instead.
• Multiply rupees per use by the stain count from the fourteen-day log longhand, scale it to a year, and write the difference as one rupee figure.
• Pin the five patches themselves in a row across the page with the five cost figures written under them.
• Choose the winner on cost and score together and write: 'The household will use X for one month, and I expect rupees per week to fall from A to B and minutes per week from C to D.'
• Read the page aloud to whoever does the washing and have them sign and date it if they agree to the switch.

Write the rules that keep the month's measurement identical to the fortnight's.

• Write what stays fixed: same log columns, same drawn scale, same watch, same scorer, same tap, same brush.
• Write what is allowed to move: only the product used to treat the stain.
• Set the winning product aside in one container, weigh it or mark its level on day one with a pen line, and write that starting reading down.
• Rule four blank labelled rows for weeks one to four, with the dates beside them, before any data exists.
• Write your prediction on the page, date it, and have one person witness it with their signature.

Run the household on the winning method for one week and log every stain event the day it happens.

• Log each event on the day: item, product, quantity used, minutes spent scrubbing, and the score against the drawn scale.
• Read the container level or weight at the end of the week and compare it against the sum of the logged doses.
• Trace one treated item's stain before and after, so the month carries a measured record beside the patch tracings.
• Write down what went wrong: a forgotten entry, somebody reaching for the old bottle, a stain treated twice, an event nobody told you about.

Finish the month, normalise both periods and compare them honestly.

• Keep logging every stain event for three more weeks, changing nothing the protocol does not allow.
• Divide the baseline fortnight by two and the switched month by four, longhand, to get rupees per week and minutes per week for each.
• Draw four bars on squared paper to a stated scale: rupees per week before and after, minutes per week before and after.
• List every confounder you can name: fewer stains this month, a cook away from home, a price change, warmer water, somebody scrubbing harder than before.
• Write the change as a rupee figure, as a percentage and in minutes, and say plainly whether one month is enough to be sure.

Publish the finding, defend it to real people, and record what the household will keep.

• Deliver the defence out loud in under three minutes to three people who sign and date the page, or record it on a borrowed phone.
• Read the one-page explainer to three people who wash clothes in this house or nearby and write down what each of them said.
• Hand or send the cost-against-score chart to the expert you interviewed with one question: what would you have controlled that I did not.
• Ask the household in front of a witness whether the switch stays and why, and write the answer down word for word.
• Complete the relations ledger with ten names, roles, dates and what each person replied.

Pick one street or one building, tell a named adult where you will be, and fix the single question you will ask everyone.

• Sketch the street or the building with every household as a numbered box, and write no house number and no name on the sketch.
• Write the question once and never change it: 'What is the one thing here that wastes your time or your money that nobody has ever asked you about.'
• Tell a named adult the street, the hours you will be there and when you will be back, and for anyone under eighteen arrange for that adult to come along and stay within sight.
• Write the opening line you will say at every door: who you are, what the question is for, how long it takes, and that they are free to say no.
• Build a Google Form with three fields: household code, the answer typed word for word, and a refusal box.

Put the same open question to fifteen to twenty households and write every answer down word for word.

• Call in daylight, at the doorway, the gate or the shared space, and never step inside for any reason.
• Read the question exactly as written and then stop talking, however long the silence runs.
• Type the answer verbatim into the form, in the words used, including the repetition and the hesitation — do not tidy it and do not shorten it.
• Where an answer is long, read it back and ask whether you have written what they meant, then correct it in front of them.
• Log every refusal as a refusal against a household code and move on without asking twice.

Sort the verbatim answers into groups and find the problem raised unprompted most often.

• Pull the form responses into Sheets, one answer per row, and read all of them through once before labelling anything.
• Label each answer with a group name taken from the residents' own words rather than a category you invented.
• Count how many households raised each group, and separately how many raised it first, before anything else in their answer.
• Ask one person who did not do the survey to group the same answers blind, then compare their grouping with yours row by row.
• Chart the groups in Datawrapper, largest first, with the count printed on each bar.

Turn the top problem into minutes per household per week or rupees per household per month, measured wherever it can be measured.

• Return to five households who raised it and ask two fixed questions: how long does it take each time, and how many times a week does it happen.
• Time it yourself wherever it can be timed — the walk, the queue, the wait, the second trip — three separate times, and record all three timings rather than an average.
• Check any price yourself at the shop or the office rather than accepting a remembered figure.
• Put every household's figure in its own row and calculate the median as well as the mean, because one extreme household drags a mean a long way.
• Mark every figure as measured or reported, and never add the two into one total without saying which is which.

Scale the per-household cost to the whole street each month and check that total with residents themselves.

• Multiply the median per-household figure by the number of households on the street and then by the weeks in a month, with the arithmetic sitting in visible formula cells.
• State how many households you actually asked against how many exist, because every household you did not ask is an assumption and must be labelled one.
• Convert minutes into hours at the end rather than in the middle, and round exactly once.
• Read the total back to three residents and write down whether each said it sounds high, low or about right.
• Write one line naming what would make this total wrong and roughly by how much.

Finish the costed problem statement on one page and agree one small change with at least three neighbours.

• Write four lines: whose problem this is, how many households raised it unprompted, the cost per household, and the total hours or rupees the street loses each month.
• List three changes that need nothing bought: a shared rota, a sign, a bin moved, a step repaired, one pooled purchase.
• Take the page back to the households who raised it, ask which change they would actually take part in, and write down the refusals as carefully as the agreements.
• Get three neighbours to take a role each, recorded by household code, with what they agreed to do and on which days written beside the code.
• Write the prediction as one sentence: 'We will do X for fourteen days and expect minutes per household per week to fall from A to B.'

Put the change physically in place and fix the rules that keep the after-measurement identical to the before.

• Ask whoever is responsible for the space before anything is moved, repaired or fixed to a wall, and record the permission by role rather than by name.
• Make the thing itself: type and print the rota, letter the sign in Docs, move the bin, repair the step with what is already to hand.
• Write what stays fixed: same coded households, same question wording, same watch, same three-timing rule, same time of day.
• Put both dates in a shared calendar — the start and the re-measurement fourteen days later — and tell the three neighbours both of them.
• Photograph the sign, the rota or the repair with nobody in the frame and file it in a dated Drive folder.

Keep the change running for fourteen days and log what actually happens to it, including the days it fails.

• Check the rota, the sign or the repair every second day and write one line on its condition with the date.
• Log every time it is ignored, removed or complained about, with the date and no name attached.
• Ask the three neighbours once a week whether they are still doing their part, and record the honest answer rather than the polite one.
• Where it breaks, put it back once and record that you did, then leave it alone if it breaks again.
• Write a dated failure entry each week whether or not anything went wrong.

After fourteen days, measure the same quantity in the same households, in the same way, and compare.

• Return to the same coded households in the same order and at the same time of day, including those who chose not to join the change - but never to a household that asked not to be called on again.
• Time the same activity with the same watch and the same three-timing rule, recording all three timings again.
• Read the frequency question from the page word for word, exactly as it was asked the first time.
• Recalculate minutes per household per week and hours across the street per month using the identical arithmetic in the same cells.
• Chart before against after in Datawrapper as two bars, and list every confounder: weather, a holiday, a repair by somebody else, a household that moved, a shop that closed.

Publish the finding back to the street, defend it, and record whether the change stays.

• Record a defence video of no more than three minutes: the question, what people actually said, the cost, the change, the after number and the limits.
• Put the one-page explainer where the street can read it, with permission from whoever owns that wall or board, and with no names on it.
• Send the before-and-after chart to the expert you interviewed with one question: what would you have measured that I did not.
• Ask the three neighbours on record whether the change stays now that the fortnight has ended, and write each answer word for word.
• Complete the relations ledger with ten entries — names for the expert, the peers and the people you shared with, codes for every resident.

Pick one street or one building, tell a named adult where you will be, and fix the single question you will ask everyone.

• Draw the street or the building on one page with each household as a numbered box, and write no house number and no name anywhere on it.
• Write the question on a card, once: 'What is the one thing here that wastes your time or your money that nobody has ever asked you about.'
• Tell a named adult the street, the hours and your return time, and for anyone under eighteen arrange for that adult to come along and stay within sight.
• Write the doorway opening line on the back of the card: who you are, what the question is for, how long it takes, and that they are free to say no.
• Rule the notebook into four columns: household code, the answer word for word, refused, time of day.

Put the same open question to fifteen to twenty households and write every answer down word for word.

• Call in daylight, at the doorway, the gate or the shared space, and never step inside for any reason.
• Read the question from the card exactly as written, then stop talking however long the silence runs.
• Write the answer in their words, including repetition and hesitation, and never in your own summary of it.
• Where an answer is long, read it back and ask whether you have written what they meant, then correct it in front of them.
• Mark every refusal in the refused column against a household code and move on without asking twice.

Sort the verbatim answers into groups and find the problem raised unprompted most often.

• Copy each answer onto its own slip of paper and read all of them through once before sorting anything.
• Lay the slips out on a table and push them into piles, giving every pile a name taken from the residents' own words.
• Count the slips in each pile, and count separately how many raised that problem first, before anything else in their answer.
• Shuffle the slips and ask one person who did not do the survey to make their own piles blind, then compare pile by pile.
• Draw the problems-by-count bar chart on squared paper, largest bar first, with the count written on each bar.

Turn the top problem into minutes per household per week or rupees per household per month, measured wherever it can be measured.

• Return to five households who raised it and ask two fixed questions: how long does it take each time, and how many times a week does it happen.
• Time it yourself with the watch wherever it can be timed — the walk, the queue, the wait, the second trip — three separate times, writing all three down.
• Walk to the shop or the office and write down any price yourself rather than trusting a remembered figure.
• Rule one row per household, work out the median and the mean longhand in the margin, and show both.
• Mark every figure M for measured or R for reported, and never add the two into one total without saying which is which.

Scale the per-household cost to the whole street each month and check that total with residents themselves.

• Multiply the median per-household figure by the number of households and then by the weeks in a month, leaving the working visible on the page.
• Write how many households you asked against how many exist, and label the rest plainly as an assumption.
• Convert minutes into hours at the very end rather than in the middle, and round exactly once.
• Read the total back to three residents and write down whether each said it sounds high, low or about right.
• Write one line naming what would make this total wrong and roughly by how much.