Make the busy weeks pay and survive the quiet ones: 10 AI skills for seasonal trading
peak-debrief
what the season really earned, in numbers you can act on next year
How the two work together
Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.
Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.
No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.
Prompt for Claude
--- name: peak-debrief description: Turns a finished peak into one dated file of numbers you can build next year's order and rota from: revenue net of VAT and net of tips, the true labour cost including on-costs and accrued holiday, the stock that did not sell and the stock that was thrown, gross margin per trading hour day by day, and the three decisions worth changing with a cash figure against each. Use in the fortnight after a bank holiday weekend, a festival, a summer season or a Christmas run, while the records are still to hand. --- # What the season really earned, in numbers you can act on next year You give this the period, your till exports, your payroll for those weeks, your purchase invoices and stock counts at both ends, and your tipping records. You get back one dated file: what came in after VAT and after tips, what it actually cost to serve, where the margin was made hour by hour, what the weather did and did not explain, and three named decisions with the money they were worth. It will not tell you the season was a success or a failure. It produces the numbers that let you decide, and it says out loud where a number could not be built because a record does not exist. ## What it does 1. **Fix the period and the comparison before you open a single file.** Write the start date, the end date, the trading days inside it and the comparison you will use, and do it first, because a period chosen after the numbers are in is a period chosen to flatter. The comparison is normally the same calendar period last year, adjusted for the fact that bank holidays move: Easter Monday fell on 6 April in 2026 and falls on 29 March in 2027, so a like for like Easter comparison is not a like for like date comparison. Where the venue traded different hours, opened a new area or ran a promotion, note it now as a stated difference rather than discovering it halfway through and adjusting to taste. 2. **Take revenue from the records you already have to keep, not from memory or a summary screen.** GOV.UK requires VAT registered businesses to keep records of "everything you buy and sell (including zero-rated, reduced and VAT exempt items)", copies of all invoices issued and received, and a VAT account showing "your total VAT sales", "your total VAT purchases", "the VAT you owe HM Revenue and Customs (HMRC)" and "the VAT you can reclaim from HMRC". Those records must be kept "at least 6 years", and under Making Tax Digital certain records must be kept digitally, including the VAT on goods and services supplied and received and the time and value of each supply. So the revenue figure comes from the VAT account and the till export together, net of VAT, split wet and dry, by day. A gross figure read off a summary screen and compared against a net figure from last year is the commonest error in this exercise. 3. **Take tips and service charge out of revenue entirely, because they are not yours.** Under the Employment (Allocation of Tips) Act 2023 and its statutory code, which "come into effect on Tuesday 1 October 2024", employers "must ensure that all tips are distributed to staff, at the latest, by the end of the month following the month in which the tips are paid by customers". The code requires "a written tipping policy when qualifying tips are paid at or are otherwise attributable to a place of business on more than an occasional and exceptional basis", and a tipping record covering all qualifying tips received and the amount allocated to each worker, which "must be maintained for a period of three years". A worker may request to view it once in any three month period. Pull the figure from that record. A peak that looks strong because a discretionary service charge was counted as turnover is a peak that did not happen. 4. **Build the true labour cost hour by hour from records you are already required to hold.** Regulation 59 of the National Minimum Wage Regulations 2015 requires an employer to keep records sufficient to establish that the worker is paid at least the minimum wage, in a form that enables the information for a pay reference period to be produced "in a single document", and those records "must be kept by the employer for a period of six years beginning with the day upon which the pay reference period immediately following that to which they relate ends". That retention was three years until it was substituted with six on 1 April 2021. Take gross pay from payroll, add employer National Insurance and pension, and add accrued holiday for irregular hours staff at "12.07% of actual hours worked in a pay period". Then divide by trading hour, not by day, because the wage cost of an empty Tuesday afternoon does not show up in a daily total. 5. **Cost the stock twice: what did not sell, and what was thrown away.** Take an opening and a closing count and the purchase invoices between them, and compute what was actually consumed rather than what was bought. Then take the waste separately, because those are different mistakes. Over-ordering leaves stock on the shelf, which is cash you can still recover. Spoilage, over-preparation and plate waste are cash already gone, and WRAP's Guardians of Grub tools separate them: its bin labels and 7-Day Tracking Sheet split the streams, and its Food Tracking Calculator "automatically calculates what the food thrown away is costing you". If the peak was not measured at the time, say so plainly and record it as a gap rather than estimating, because an estimated waste figure will be quoted next year as though somebody weighed it. 6. **Test the weather excuse against published data instead of accepting or dismissing it.** The Met Office publishes free UK and regional series covering "Max temp Min temp Mean temp Sunshine Rainfall Rain days ≥1.0mm Days of air frost" as "monthly, seasonal and annual values" for the UK, the four nations and regions including the Midlands and East Anglia. Pull the values for your region for both periods and put them next to the takings. Do not compute a correlation and do not claim causation from a handful of monthly points on a series covering several counties. Put the two columns side by side and let a person who knows the site read them. Half the time the excuse holds and half the time it does not, and knowing which stops a venue rebuilding a menu that was never the problem. 7. **Rebuild the peak as gross margin per trading hour, day by day, and look at the shape.** For each trading hour: net revenue, cost of sales, labour cost, and the margin left. This is the step that changes decisions, because a peak is never uniformly busy and the totals hide it. Almost every venue finds one or two sessions that lost money with a full rota on, usually a Sunday evening or the Monday lunch after the big Saturday. Order the hours by margin and look at the bottom five. Those five hours are next year's opening times, and the conversation is much easier when it is a table rather than an opinion about how it felt. 8. **Name three decisions you would change and put a cash figure on each.** Not ten. Three, each written as a decision somebody actually made, with the number it cost or earned and the person or role who would make it differently. "We ordered 40 kilos and used 26, at £7.40 a kilo, so £103.60 sat in the bin" is a decision. "Food costs were too high" is a mood. Where a figure cannot be built from the records, write the decision anyway and name the record that would have made it measurable, because that missing record is itself one of next year's changes. Three decisions with numbers get implemented. A list of fourteen observations gets read once. 9. **Write the file so next year's order and rota can be built by opening it, and archive it to the retention you already owe.** One dated file per peak, with the period, the comparison, the per hour table, the stock and waste figures, the three decisions, and a list of the numbers that could not be built and why. Keep it with the underlying records, which you must hold anyway: VAT records for at least six years, minimum wage records for six years, and the tipping record for three. Next year's order and rota both start by opening it, and correcting last year's numbers takes an hour where building them from nothing takes a fortnight. ## Then it checks 1. The period, the trading days and the comparison period are stated with their dates, and any difference between the two periods in hours, areas or promotions is written down rather than adjusted for silently. 2. Every revenue figure is net of VAT and net of tips and service charge, and the file names the record each figure came from. 3. The labour cost includes gross pay, employer National Insurance, pension and accrued holiday for irregular hours staff, and is expressed per trading hour as well as per day. 4. Stock consumed is built from an opening count, purchases and a closing count, and waste is reported separately from unsold stock, with any unmeasured waste recorded as a gap rather than estimated. 5. The climate figures sit beside the takings as two columns with no causal claim made, and the region and period they cover are named. 6. Exactly three decisions are named, each with a cash figure or an explicit statement of the missing record, and the file carries its build date and the date each source was read. Any check fails: name it, redo that step once. Failed twice: say what is wrong and stop. ## Rules - Public information only. - Never invent a fact, a number or a quote. - Anything sent in someone's name says whose name it is. A summary shared with the team or an investor goes out under the owner's name and goes to them first. - Never state a sector average gross margin, a typical labour cost percentage, an industry wastage rate or a benchmark spend per head. No such published figure exists for a single independent venue, and a debrief that measures a business against a borrowed number produces action against a fiction. - Never estimate a number that a record would have given. Where the record does not exist, write the gap and name the record, because an estimate written this year becomes a fact next year and nobody remembers it was a guess. - Never count a discretionary service charge or a tip as turnover, and never net a tip against a wage cost. The tipping record exists precisely so that those figures stay separate, and the statutory code gives workers a right to see it. - Never claim the weather caused a change in takings. Put the two columns side by side and stop. A regional monthly series cannot speak to one postcode and a correlation from a handful of points will be quoted later as though it meant something. - Refuse to produce a verdict on whether the season was a success. Produce the numbers and the three decisions. A verdict is the owner's, and a document that supplies one replaces the thinking it was meant to support. - This output is a working document prepared for the owner to check against their own tills, payroll and invoices, and for their accountant to check before it informs any tax, pricing or employment decision. It is not accounting advice, it is not a set of statutory accounts and it does not certify that any record meets a legal requirement. ## Built from - GOV.UK, "Keeping VAT records", https://www.gov.uk/charge-reclaim-record-vat/keeping-vat-records, no publication date shown on the page, read 16 September 2026: the records that must be kept, the VAT account contents, the at least six years retention and the Making Tax Digital requirement in step 2. - Department for Business and Trade, "Code of practice on fair and transparent distribution of tips", https://www.gov.uk/government/publications/distributing-tips-fairly-statutory-code-of-practice/code-of-practice-on-fair-and-transparent-distribution-of-tips-html-version, published 29 July 2024, read 16 September 2026: the 1 October 2024 commencement, the payment deadline, the written policy duty and the three year tipping record in step 3. - Employment (Allocation of Tips) Act 2023, https://www.legislation.gov.uk/ukpga/2023/13/contents, read 16 September 2026: the statutory frame behind the code, including how and when tips must be dealt with, troncs, information and enforcement, referenced in step 3. - The National Minimum Wage Regulations 2015, regulation 59, https://www.legislation.gov.uk/uksi/2015/621/regulation/59, retention substituted from three to six years with effect from 1 April 2021 by S.I. 2021/329, read 16 September 2026: the single document requirement and the six year retention behind step 4. - GOV.UK, "Holiday pay and entitlement reforms from 1 January 2024", https://www.gov.uk/government/publications/simplifying-holiday-entitlement-and-holiday-pay-calculations/holiday-pay-and-entitlement-reforms-from-1-january-2024, last updated 1 April 2024, read 16 September 2026: the 12.07% accrual added to the labour cost in step 4. - WRAP, "Guardians of Grub: Resources", https://guardiansofgrub.com/resources/, no publication date shown on the page, read 16 September 2026: the bin labels, 7-Day Tracking Sheet and Food Tracking Calculator that separate spoilage, preparation and plate waste in step 5. - Met Office, "UK and regional series", https://www.metoffice.gov.uk/research/climate/maps-and-data/uk-and-regional-series, no publication date shown on the page, read 16 September 2026: the free variables, regions and monthly resolution set beside the takings in step 6. - GOV.UK, "UK bank holidays", https://www.gov.uk/bank-holidays, last updated 24 November 2025, read 16 September 2026: the moving Easter and substitute dates that make the comparison in step 1 a matter of checking rather than assuming.
Prompt for Codex
# peak-debrief ## You are given The start date, end date and trading days of one finished peak at a UK hospitality business, and the comparison period already chosen by Claude with the reason it is comparable. The till export for both periods, by transaction or trading hour, with the date, the hour, the gross value, the VAT amount, whether the line is wet or dry, the cover count and any service charge line. The VAT account figures. The tipping record for the period, showing qualifying tips received and the amount allocated to each worker. The payroll export for the pay periods covering the peak, with, per person, gross pay, employer National Insurance, employer pension, hours worked, and their employment pattern already judged as exactly `fixed hours` or `irregular hours`. Opening and closing stock counts with the date and time each was taken. Purchase invoices between the two counts with supplier, date, line, quantity and net value. The waste log if one was kept, with the date, the stream and the weight or value. The Met Office regional climate series for both periods. The stated differences between the periods, each in the owner's own words. And the three decisions the owner would change, each in their own words with the role who would make it differently. You do not choose the comparison period, you do not judge whether the season went well, and you do not select the three decisions. ## Produce Write into a `./peak-debrief-output/` folder: 1. `revenue-by-hour.csv` with these columns in this order: `date`, `trading_hour`, `period`, `gross_takings_gbp`, `vat_gbp`, `service_charge_gbp`, `tips_gbp`, `net_revenue_excl_tips_gbp`, `wet_net_gbp`, `dry_net_gbp`, `covers`, `average_spend_gbp`, `source_record`. `period` is exactly one of `peak`, `comparison`. `net_revenue_excl_tips_gbp` always excludes VAT, service charge and tips. `source_record` names the record the row was built from. 2. `labour-by-hour.csv` with columns: `date`, `trading_hour`, `period`, `people_on`, `paid_hours`, `gross_pay_gbp`, `employer_ni_gbp`, `employer_pension_gbp`, `accrued_holiday_hours`, `accrued_holiday_cost_gbp`, `total_labour_cost_gbp`. `accrued_holiday_hours` is 12.07% of paid hours for `irregular hours` people only and `0` for `fixed hours` people. Hours with people rostered and no revenue row still appear. 3. `margin-by-hour.csv` with columns: `date`, `trading_hour`, `period`, `net_revenue_excl_tips_gbp`, `cost_of_sales_gbp`, `total_labour_cost_gbp`, `gross_margin_gbp`, `margin_per_hour_rank`, `loss_making`. `margin_per_hour_rank` is 1 for the highest margin hour of the peak upward. `loss_making` is `yes` where `gross_margin_gbp` is below zero. 4. `stock-movement.csv` with columns: `item`, `opening_count_units`, `opening_count_datetime`, `purchases_units`, `purchases_net_gbp`, `closing_count_units`, `closing_count_datetime`, `consumed_units`, `consumed_net_gbp`, `unsold_carried_forward_units`, `unsold_value_gbp`, `count_gap`. `consumed_units` is opening plus purchases minus closing. `count_gap` is `yes` where either count is missing or undated. 5. `waste.csv` with columns: `date`, `stream`, `weight_kg`, `value_gbp`, `measured_or_missing`. `stream` is exactly one of `spoilage`, `preparation`, `plate waste`, `not separated`. `measured_or_missing` is exactly one of `measured`, `not measured`. Never write a value in a `not measured` row. Every `not measured` date is listed in `gaps.md`. 6. `weather-beside-takings.csv` with columns: `period`, `month`, `year`, `region`, `mean_temp_c`, `max_temp_c`, `rainfall_mm`, `rain_days_1mm_or_more`, `sunshine_hours`, `net_revenue_excl_tips_gbp`. Two columns of figures side by side and nothing more. No correlation column, no derived index, no commentary. 7. `three-decisions.md` with exactly three numbered entries, each carrying the decision verbatim as the owner wrote it, the role who would make it differently, and either the cash figure computed from the CSVs with the rows it came from, or the sentence naming the record that would have made it measurable. Never a fourth entry and never a reworded one. 8. `period-summary.csv` with columns: `metric`, `peak_value`, `comparison_value`, `variance`, `variance_percent`, `stated_difference_applies`. Metrics exactly: `net_revenue_excl_tips_gbp`, `wet_net_gbp`, `dry_net_gbp`, `covers`, `average_spend_gbp`, `cost_of_sales_gbp`, `total_labour_cost_gbp`, `gross_margin_gbp`, `loss_making_hours`, `waste_value_gbp`. `stated_difference_applies` names any supplied difference between the periods that touches that metric. 9. `gaps.md` as a numbered list of: any revenue figure that could not be separated from VAT, service charge or tips; any person with no employment pattern; any stock item with a missing or undated count; every date with no waste measurement; any month with climate data but no takings or the reverse; any of the three decisions with neither a cash figure nor a named missing record; any stated difference that could not be applied; and any figure that could not be computed because an input was missing. ## Rules - Codex measures, computes, reconciles and records. It never chooses the comparison period, never judges whether the season was a success, never selects or rewords the three decisions, and never writes a recommendation. - Never write a revenue figure that includes VAT, a service charge or a tip. Where the till export does not separate them, leave the derived column empty and list the rows in `gaps.md`. A gross figure compared against a net figure is the error this file exists to prevent. - Never net a tip against a wage cost and never move a tip figure into revenue or into labour. Tips are reported from the tipping record in their own column only. - Never estimate a waste figure, a stock count or a cover count that was not recorded. Write `not measured` and list the date. An estimate written here becomes a fact next year. - Never state that weather caused a change in takings, compute a correlation coefficient or build a weather index. `weather-beside-takings.csv` holds two sets of columns and no derived relationship. - Never write a sector average, an industry gross margin, a typical labour cost percentage or a benchmark wastage rate. No such figure exists for one independent venue and none is to be written into any file. - Never adjust a figure to make two periods comparable. Record the supplied stated difference against the metric it touches and leave both figures as found. - Accrued holiday is computed at 12.07% of paid hours for irregular hours workers only. Where the pattern was not supplied, leave the columns empty and list the person in `gaps.md`. - Never send, publish or share any file, and never connect to a till, payroll or accounting system. These are drafts for the owner and their accountant. - Use British English, GBP, DD Month YYYY dates, 24 hour times, kilograms and degrees Celsius. Money to two decimal places. No em dashes in any file you write, and any supplied text containing one is recorded verbatim and flagged in `gaps.md`. - Every file ends with this line: this is a working document prepared for the owner to check against their own tills, payroll and invoices, and for their accountant to check before it informs any tax, pricing or employment decision. It is not accounting advice and it does not certify that any record meets a legal requirement. ## Return The absolute path of each file written, the row count of each CSV, net revenue excluding tips for both periods with the variance, the total labour cost split between gross pay, employer on-costs and accrued holiday, the cost of sales and the gross margin, the number of loss making trading hours with the five lowest named by date and hour, the value of unsold stock carried forward, the measured waste value and the dates with no waste measurement, the three decisions as supplied with the cash figure or named missing record beside each, the stated differences that could not be applied, and the `gaps.md` item count.
Built from the best public work on this
Sources for peak-debrief
Everything below was opened and read on 16 September 2026. Nothing is cited that could not be loaded.
1. GOV.UK, "Keeping VAT records"
https://www.gov.uk/charge-reclaim-record-vat/keeping-vat-records, no publication date shown on the page, read 16 September 2026.
The reason this page opens the skill is that a debrief is only as good as the records behind it, and a VAT registered venue already holds far better records than it thinks. The page requires a business to keep records of "everything you buy and sell (including zero-rated, reduced and VAT exempt items)", "copies of all invoices you issue", "all invoices you receive (original or electronic copies)", plus self-billing agreements, debit and credit notes, and "any goods you give away or take from stock for your private use".
It also requires a VAT account showing "your total VAT sales", "your total VAT purchases", "the VAT you owe HM Revenue and Customs (HMRC)" and "the VAT you can reclaim from HMRC". Records must be kept for at least six years, extending to ten for the One Stop Shop schemes. Under Making Tax Digital certain records must be kept digitally unless the business is exempt, including the VAT on goods and services supplied and received and the time and value of supply for all transactions.
Step 2 uses this in a very practical way. The commonest error in a peak review is comparing a gross figure from a summary screen against a net figure from last year's accounts, which quietly inflates the current period by a fifth. The VAT account is the reconciliation point that stops it, and it is already sitting there because the law requires it.
Where the skill departs: the page also covers the retail schemes, the flat rate scheme, cash accounting and the specific rules for margin schemes, none of which this skill attempts. A venue on the flat rate scheme has a different relationship between its gross and net figures, and the skill sends that question to the owner's accountant rather than assuming standard VAT accounting.
2. Department for Business and Trade, "Code of practice on fair and transparent distribution of tips", with the Employment (Allocation of Tips) Act 2023
https://www.gov.uk/government/publications/distributing-tips-fairly-statutory-code-of-practice/code-of-practice-on-fair-and-transparent-distribution-of-tips-html-version, published 29 July 2024, and https://www.legislation.gov.uk/ukpga/2023/13/contents, both read 16 September 2026.
The Act supplies the frame, with sections on how and when tips must be dealt with, independent troncs, agency workers, information, enforcement and the code itself. The code supplies the dates and the duties this skill actually uses.
Commencement is explicit: "This code and specified provisions of the legislation come into effect on Tuesday 1 October 2024." The payment deadline is a hard one: "employers must ensure that all tips are distributed to staff, at the latest, by the end of the month following the month in which the tips are paid by customers." There is a policy duty where tipping is more than incidental: "the employer is required to have a written tipping policy when qualifying tips are paid at or are otherwise attributable to a place of business on more than an occasional and exceptional basis." And there is a record duty which is the one this skill depends on: "A tipping record must include detail of all qualifying tips received by the employer at the place of business, and the amount allocated to each worker. This record must be maintained for a period of three years", with workers entitled to make "a written request - limited to one request per worker in a three month period - to view the tipping record of their employer for a period dating back up to 3 years".
That record is why step 3 can subtract tips cleanly instead of guessing. It also explains the rule that forbids netting tips against wages. A venue that quietly treats a service charge as contributing to its wage bill is doing something the code is designed to make visible, and any worker can ask to see the record that shows it.
Where the skill departs: the code sets out fairness factors for allocating tips and the discrimination risks in choosing them, and the Act deals with troncs and enforcement through the employment tribunal. This skill does not design or review a tipping policy and does not judge whether an allocation is fair. It uses the record as a source of two numbers and sends the policy to the owner's HR adviser.
3. The National Minimum Wage Regulations 2015, regulation 59
https://www.legislation.gov.uk/uksi/2015/621/regulation/59, read 16 September 2026.
Regulation 59 requires an employer to keep records sufficient to establish that a worker is being remunerated at at least the minimum wage, in a form that enables "the information kept about a worker in respect of a pay reference period to be produced in a single document", with additional records for unmeasured work under a daily average agreement and for output work.
The retention period is the detail worth getting right, because it changed. The current text reads that records "must be kept by the employer for a period of six years beginning with the day upon which the pay reference period immediately following that to which they relate ends". The version as originally made in 2015 said three years. It was substituted to six by the National Minimum Wage (Amendment) Regulations 2021, S.I. 2021/329, with effect from 1 April 2021, and anyone citing the "as made" text today would state it wrongly.
Step 4 leans on this for a boring but useful reason. A venue reviewing a peak from last summer very often believes it cannot rebuild the labour cost hour by hour because the rota has been thrown away. The payroll records exist, they are required to exist for six years, and the regulation requires them to be producible as a single document per worker per pay reference period. The hour by hour build is almost always possible, and it is the step that finds the loss making sessions.
Where the skill departs: the regulation is about minimum wage compliance and the skill does not audit minimum wage compliance, which is enforced by HMRC with its own penalty regime. It uses the regulation to establish that the records exist and how long they last, and sends any suspicion of underpayment to the owner's payroll provider.
4. GOV.UK, "Holiday pay and entitlement reforms from 1 January 2024"
https://www.gov.uk/government/publications/simplifying-holiday-entitlement-and-holiday-pay-calculations/holiday-pay-and-entitlement-reforms-from-1-january-2024, last updated 1 April 2024, read 16 September 2026.
Used for a single number that a peak review routinely omits. The guidance states that "For leave years beginning on or after 1 April 2024, there is a new accrual method for irregular hour workers and part-year workers in the first year of employment and beyond. Holiday entitlement for these workers will be calculated as 12.07% of actual hours worked in a pay period." It defines an irregular hours worker as one whose paid hours in each pay period are, under the contract, "wholly or mostly variable", and confirms that rolled-up holiday pay is permitted "for irregular hour and part-year workers only" for leave years beginning on or after 1 April 2024.
Peak trading is largely staffed by exactly these workers. A Christmas run using six casuals across four days generates a real holiday liability, and a debrief that costs only the hours paid understates the labour cost of the peak by roughly one twelfth. Over a season that is the difference between a session looking marginal and a session looking loss making, which is a different decision about next year's opening hours.
Where the skill departs: the guidance covers the 52 week reference period, part-year workers, and the interaction with sickness and family leave. The skill applies the 12.07% accrual to irregular hours workers only, and where a worker's pattern was not supplied it leaves the field empty and lists it as a gap rather than deciding, because the definition turns on contract wording this skill does not read.
5. WRAP, "Guardians of Grub: Resources"
https://guardiansofgrub.com/resources/, no publication date shown on the page, read 16 September 2026.
WRAP is the named trade body here and the campaign is its hospitality one. The resources page lists a Starter Guide, a 7-Day Tracking Sheet, a Food Tracking Calculator, bin labels, a Stock Control Sheet, a Plate Waste Toolkit, a Menu Planning Guide and a Food Waste Reduction Action Plan. The 7-Day Tracking Sheet "helps food businesses to easily monitor and measure food thrown away over a 7 day period", and the Food Tracking Calculator "automatically calculates what the food thrown away is costing you and shows what savings your business could make".
Step 5 uses the separation more than the totals. Bin labels split spoilage, preparation and plate waste, and those are three different mistakes with three different owners. Spoilage is a purchasing mistake and belongs to whoever ordered. Preparation waste is a specification and skills mistake and belongs to the kitchen. Plate waste is a portioning and menu mistake and belongs to whoever set the dish. One undifferentiated bin weight tells you none of that, which is why an unmeasured peak produces a debrief that can identify a problem and not a fix.
Where the skill departs: WRAP publishes sector-level cost figures and this skill quotes none of them at an owner. It also refuses to estimate waste that was not measured at the time, which is a real limitation and is stated as one. A venue that did not track its peak gets a gap in the file and a reason to track the next one, not a plausible number that will be treated as fact in twelve months.
6. Met Office, "UK and regional series"
https://www.metoffice.gov.uk/research/climate/maps-and-data/uk-and-regional-series, no publication date shown on the page, read 16 September 2026.
Free historic climate data published under Crown copyright, covering "Max temp Min temp Mean temp Sunshine Rainfall Rain days ≥1.0mm Days of air frost" as "monthly, seasonal and annual values", downloadable "in rank or year order", for the UK, the four nations and regions including "England E & NE", "England NW/Wales N", "Midlands" and "East Anglia".
Step 6 exists because the weather is the most used and least tested explanation in hospitality. Putting the published figures beside the takings settles it in five minutes, in either direction. A wet August that was genuinely the wettest in years is worth knowing, because it stops an owner rebuilding a menu that was never the problem. An ordinary August is worth knowing for the opposite reason.
Where the skill departs: the skill forbids any causal claim and forbids computing a correlation. A regional series covers several counties and a handful of monthly points cannot support a coefficient, but a coefficient once written down will be quoted afterwards as though it could. Two columns side by side, read by someone who knows the site, is both more honest and more useful.
7. GOV.UK, "UK bank holidays"
https://www.gov.uk/bank-holidays, last updated 24 November 2025, read 16 September 2026.
Cited for one job in step 1: making the like for like comparison honest. Easter moves, and the page shows it, with Good Friday on 3 April and Easter Monday on 6 April in 2026, and Good Friday on 26 March with Easter Monday on 29 March in 2027. It also states the substitution rule: "if a bank holiday is on a weekend, a 'substitute' weekday becomes a bank holiday, normally the following Monday", which is why Christmas 2027 falls on Monday 27 and Tuesday 28 December.
Comparing the first weekend of April this year with the first weekend of April last year is not comparing Easter with Easter, and the difference in a venue that trades on holiday footfall can be most of the variance being analysed. Checking the dates takes thirty seconds and removes an error that otherwise contaminates every number downstream.
Where the skill departs: Scotland and Northern Ireland have different lists and the skill uses whichever nation applies, saying which. It also does not attempt to normalise or weight for the calendar shift. It records the difference as a stated difference and leaves both figures as found, because an adjusted figure is a figure somebody has to trust rather than check.
Best public prompt we found for this job
The best public artefact for this job is the `metrics-review` skill in Anthropic's `knowledge-work-plugins`, raw source at https://raw.githubusercontent.com/anthropics/knowledge-work-plugins/main/product-management/skills/metrics-review/SKILL.md. The repository has 24,123 stars, read from api.github.com on 16 September 2026. At around eighteen thousand bytes it is one of the larger skills in the repository, and three of its instructions are good enough to take.
The first is its insistence on asking for context before analysing anything:
Any known events that might explain changes (launches, outages, marketing campaigns, seasonality)?
That is the stated differences list in step 1, gathered before the numbers rather than invented afterwards to explain them.
The second is its scorecard shape, a table of current value, previous value, change, target and status, which became `period-summary.csv` in the Codex output. A metric with a stated comparison beside it is checkable. A metric in a sentence is an assertion.
The third is its separation of what happened from why it likely happened, and its requirement to say "Whether this is a one-time event or a sustained trend". A single wet bank holiday is not a trend and treating it as one is how a venue stops opening on a day that is usually its best.
What we did not copy is the framing and the output. `metrics-review` is written for a product team with an analytics tool, a North Star metric and an L1 and L2 metric hierarchy. An independent venue has a till, a payroll export and a pile of invoices, and asking it to define a North Star metric before it can review a bank holiday is a way of ensuring the review never happens. So the sources in this skill are the records the business is legally required to hold already, which means the data always exists.
Two deliberate removals. The skill asks for "Bright Spots" and "Areas of Concern" sections, which in practice produce a long list read once. This skill produces exactly three decisions with cash figures against them, because three get implemented. And `metrics-review` ends with recommended actions and an overall health assessment. This one refuses to state whether the season was a success. That judgement involves the owner's cash position, their staff, their lease and how tired they are, and a document that supplies a verdict quietly replaces the thinking it was built to support.
Want this running in your business?
I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.
Foxera