cellar-check

---
name: cellar-check
description: Works out how many saleable pints each container should give you, measures how many it actually gave, splits the gap across the named causes with the cost of each, and ranks the fixes by pints recovered a week. Use when draught gross profit has slipped, when the yield feels wrong but nobody can prove it, or before you sign a new supply agreement.
---

# Find the yield you are pouring down the drain

You give this your delivery notes, your till data for draught, your line lengths and bore, your cleaning schedule, your glassware and your dispense setup. You get back one dated file: the theoretical yield of every container you buy, the yield you actually achieved over a measured window, the gap split across line pull-through, ullage, froth, quality dumps and giveaway with a cost against each, the safety and equipment checks that belong in the same visit, and a ranked list of fixes with pints recovered a week next to each. It measures, it does not estimate, and it says which numbers it could not obtain.

## What it does

1. **Compute the theoretical yield of every container before you look at a single sales figure.** One imperial pint is 0.56826125 litres, so a 50 litre keg holds 87.99 pints, a 30 litre keg holds 52.79, a nine gallon firkin holds 72 and an eleven gallon barrel holds 88. Build one row per container size you actually buy, taken from the delivery note and not from the brand's marketing, because 30 litre and 11 gallon are not the same container and the difference is nearly three pints. This number is the denominator for everything that follows, and getting it from memory is how a yield investigation produces a confident wrong answer.

2. **Measure achieved yield over a fixed window rather than over a feeling.** Choose a window of whole weeks, count opening and closing cellar stock in containers and part containers by weight, add the deliveries from the notes, and take the draught sales from the till in pints by product. Achieved yield per container is sales plus closing minus opening minus deliveries, expressed per container emptied. Record the window's start and end dates and who counted. A window that starts mid-keg and ends mid-keg with nobody weighing the part container is not a measurement, and the result will be whatever the guess was.

3. **Split the gap across five named causes and refuse to leave a residual unnamed.** Line pull-through at each clean. Ullage and sediment left in the container. Froth poured over the line on a busy service. Quality dumps, meaning the first pints after a clean and anything pulled through on a line change. Giveaway, meaning the liquid that goes into the glass above what was sold. Every pint of the gap belongs in one of those five or in a sixth row called not yet identified, and that sixth row is the finding, not a rounding allowance.

4. **Compute the line loss instead of arguing about it.** The volume of a line is its length multiplied by the cross-sectional area of its bore, and a cleaning cycle pulls that through at least twice, once to fill with cleaner and once to fill back with beer. For a 20 metre run at 9.5mm internal bore the arithmetic is pi multiplied by 0.00475 squared multiplied by 20, which is 1.42 litres, about 2.5 pints per fill. Multiply by lines, by fills per clean and by cleans per year and the figure stops being an opinion. Measure the actual bore rather than assuming it, because a python carrying several different bores will not match a single calculation.

5. **Check the glassware and settle the head question with evidence.** Record for every draught glass whether it is a brim measure, which must be filled to the top, or a lined glass, which has space above the line for a head. Aberdeenshire Council's Trading Standards service, publishing on 11 July 2024, cited the British Beer and Pub Association standard that "a pint should contain a minimum of 95% liquid and 5% head", and told drinkers "If you don't want up to 5% to be head, you can ask for a top up". Both halves matter commercially. Serving brim glasses with a head is giving away liquid and serving lined glasses filled to the brim is giving away more, and the file records which glass is on which line.

6. **Test the short measure exposure, because it is a criminal offence and not a service complaint.** Section 28 of the Weights and Measures Act 1985 provides that any person who "in selling or purporting to sell any goods by weight or other measurement or by number, delivers or causes to be delivered to the buyer... a lesser quantity than that purported to be sold, or... a lesser quantity than corresponds with the price charged, shall be guilty of an offence". Pull a sample of pints across different bar staff and different times of day, weigh them, record the liquid volume against the stated quantity, and write the result down with the date and the name of the person who poured. A yield problem and a short measure problem look identical in the stock figures and only one of them is an offence.

7. **Decide the takeaway question before somebody decants a keg, because it is a duty question.** HMRC's draught relief guidance states that "Serving a drink from a draught container for consumption on your premises does not count as repackaging", which is the reassurance. The warning is next to it: repackaging draught products "means transferring them to other non-draught containers. For example, decanting from a keg into bottles", full duty has been paid on the product only in defined circumstances, and "If you do not have approval, or the full rate of Alcohol Duty has not been paid, you may have to pay a penalty". Record whether the premises sells takeaway draught at all and, if so, what the supplier has confirmed in writing about the duty paid on that stock.

8. **Do the safety and equipment checks in the same visit, because the largest cellar loss is not beer.** The Health and Safety Executive records that carbon dioxide "is a colourless and odourless gas and, because of this, people are unable to see it or smell it at elevated concentrations", that rising concentrations "can cause headaches, dizziness, confusion and loss of consciousness", and that "fatalities from asphyxiation have occurred when, at high concentrations, it has entered confined spaces such as tanks, sumps or cellars and displaced Oxygen". It gives the workplace exposure limits as 5000 parts per million over eight hours and 15000 parts per million over fifteen minutes. Record the gas storage, the alarm if there is one and when it was last tested. Then record every measuring instrument on the dispense, because the March 2026 bulletin on intoxicating liquor measuring equipment states that such equipment "must bear an approved stamp before it can be used for trade".

9. **Rank the fixes by pints recovered a week, put a cost and an owner on each, and date the file.** Three groups: fix this week, fix this quarter, and monitor with a date. Every item carries the pints a week it recovers, computed from steps 3 to 5, the cost of doing it, and the name of the person responsible. In the footer, the measurement window, the date every figure was taken, and an explicit list of the numbers that could not be obtained. A cellar report without a window and without the not-yet-identified row is a report that will be quoted for two years after its figures stopped being true.

## Then it checks

1. Every container size has a theoretical yield computed from the delivery note's stated volume, and the conversion used is stated in the file.
2. The measurement window is a whole number of weeks with named start and end dates, part containers were weighed at both ends, and the person who counted is named.
3. Every pint of the gap is allocated to one of the five named causes or to a not-yet-identified row, and the six rows sum to the gap with no unexplained remainder.
4. The line loss figure names the measured length and the measured bore for each run, and no run is costed from an assumed bore.
5. Every draught glass in use is recorded as brim or lined and matched to the line it serves, and the short measure sample names the date, the time, the pourer and the measured volume of each pint pulled.
6. The gas storage, the alarm and its last test date are recorded, every measuring instrument is recorded with its stamp or as unverified, and the footer lists the figures that could not be obtained.

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 finding that names a member of staff who poured a short pint goes to the owner and nowhere else.
- Never estimate a yield loss that could have been measured. Pull-through is arithmetic, ullage is a weighing, and giveaway is a sample of real pints. An estimate entered as a measurement is how a cellar report exonerates the wrong cause.
- Never state an industry standard wastage percentage, a typical cellar loss, a normal draught gross profit or a benchmark yield. No such figure exists for a single independent UK venue, and an owner who reaches theirs will stop looking while the leak continues.
- Never report a short measure finding as a yield finding or the reverse. They have the same signature in the stock figures, different causes and completely different consequences, and the file names which one it has evidence for.
- Refuse to recommend serving a brim glass short of the brim, topping up only on request, or any other practice whose effect is to deliver less liquid than was sold. Yield is recovered by fixing lines, cleaning cycles, glassware and dispense, not by pouring less than the customer paid for.
- Never enter a cellar to take a measurement on the strength of this document. Gas risk is a matter for the owner's own risk assessment and their competent person, and a reading taken by somebody who should not have been down there is not worth having.
- This output is a working document prepared for the owner to check against their own delivery notes, till data and equipment, and for their accountant, their brewer or supplier, their Trading Standards service and their health and safety adviser to check where it touches duty, measures or gas safety. It applies published rules and the owner's own measurements. It is not tax advice, it is not legal advice, it is not a risk assessment and it does not certify that anything is safe or compliant.

## Built from
- The Weights and Measures Act 1985, section 28, https://www.legislation.gov.uk/ukpga/1985/72/section/28, read 16 September 2026: the short weight offence in its two limbs, which is step 6 and the rule separating a yield finding from a short measure finding.
- The Weights and Measures (Intoxicating Liquor) Order 1988, article 2, https://www.legislation.gov.uk/uksi/1988/2039/article/2, the revised version as it now stands, read 16 September 2026: the permitted draught quantities and the requirement to serve in a capacity measure unless the four metered-dispense conditions are met, which is what step 5 checks the glassware against.
- Aberdeenshire Council Trading Standards, "Don't lose your head, warn Trading Standards", https://www.aberdeenshire.gov.uk/news/2024/jul/don-t-lose-your-head-warn-trading-standards, published 11 July 2024, read 16 September 2026: the British Beer and Pub Association standard of a minimum 95% liquid and 5% head, and the customer's ability to ask for a top up, both quoted in step 5.
- HM Revenue and Customs, "Check if you can pay less Alcohol Duty on draught alcoholic products", https://www.gov.uk/guidance/check-if-you-can-pay-less-alcohol-duty-on-draught-products, published 29 June 2023 and last updated 1 February 2026, read 16 September 2026: the statement that serving from a draught container on the premises is not repackaging, the definition of repackaging, and the penalty warning, all in step 7.
- Health and Safety Executive, "General hazards of Carbon Dioxide", https://www.hse.gov.uk/coshh/basics/carbondioxide.htm, the page shows no publication date and records that it was updated on 20 May 2026, read 16 September 2026: the colourless and odourless description, the symptoms of rising concentration, the cellar asphyxiation warning and the two workplace exposure limits, all in step 8.
- Office for Product Safety and Standards, "WM 1008: Intoxicating liquor measuring equipment, version 1, March 2026", https://www.gov.uk/government/publications/weights-and-measures-bulletins/wm-1008-intoxicating-liquor-measuring-equipment-version-1-march-2026, published March 2026 and updated 12 March 2026, read 16 September 2026: the requirement that measuring equipment bear an approved stamp before it can be used for trade, in step 8.

# cellar-check

## You are given
Every delivery note for draught products across the window, showing product, container size in litres or gallons as printed, number of containers and delivery date. The till export of draught sales for the same window, in pints by product. The opening and closing cellar stock counts, with every part container recorded by weight, plus the tare weight of the empty container and the density figure the owner uses. The window as named start and end dates, the name of the person who counted, and confirmation that it is a whole number of weeks. The dispense layout: every line by reference, its product, its length in metres, its internal bore in millimetres, and whether that bore was measured or assumed. The cleaning schedule: cleans per line per year, fills per clean, and the dates of the last four cleans. The glassware in use per line, each recorded as brim or lined with its nominal capacity. The short measure sample: for each pint pulled, the date, the time, the pourer, the line, the stated quantity and the measured liquid volume in millilitres. The gas storage, the alarm if one exists and the date it was last tested. The measuring equipment inventory with any stamp reference. Whether the premises sells takeaway draught, and any written supplier confirmation about the duty paid on that stock. The judgements already made by Claude: the allocation of each portion of the gap to a named cause, and the ranking group for each proposed fix.

## Produce
Write into a `./cellar-check-output/` folder:

1. `yield.csv` with these columns in this order: `product`, `container_size_as_printed`, `container_size_litres`, `conversion_used`, `theoretical_pints_per_container`, `window_start`, `window_end`, `weeks_in_window`, `opening_stock_pints`, `deliveries_pints`, `closing_stock_pints`, `sales_pints`, `containers_emptied`, `achieved_pints_per_container`, `gap_pints_per_container`, `gap_pints_total`, `part_containers_weighed`. `conversion_used` states the litres per pint figure applied, as supplied, and `theoretical_pints_per_container` is the litres divided by it, to two decimal places. `part_containers_weighed` is `yes` or `no`, and any `no` is listed in `gaps.md`.
2. `gap-allocation.csv` with columns: `product`, `cause`, `pints`, `basis`, `measured_or_supplied`, `cost_per_pint`, `cost`. `cause` is exactly one of `line pull-through`, `ullage and sediment`, `froth over the line`, `quality dumps`, `giveaway`, `not yet identified`. `measured_or_supplied` is exactly `measured` or `supplied by the owner`. For every product the six `pints` values must sum exactly to `gap_pints_total` in `yield.csv`, and any product where they do not is listed in `gaps.md`.
3. `lines.csv` with columns: `line_ref`, `product`, `length_m`, `bore_mm`, `bore_measured_or_assumed`, `line_volume_litres`, `line_volume_pints`, `fills_per_clean`, `cleans_per_year`, `pints_per_year`, `last_four_clean_dates`. `line_volume_litres` is pi multiplied by the square of half the bore in metres, by the length in metres, by 1000. Every `assumed` row is listed in `gaps.md`.
4. `glassware-and-measures.csv` with columns: `line_ref`, `glass_type`, `nominal_capacity`, `stamped`, `sample_date`, `sample_time`, `pourer`, `stated_quantity_ml`, `measured_liquid_ml`, `shortfall_ml`. `glass_type` is exactly `brim` or `lined`. One row per sampled pint, plus one row per line with the sample columns empty where no pint was sampled, and every such line is listed in `gaps.md`.
5. `safety-and-equipment.csv` with columns: `item_ref`, `item_type`, `location`, `detail`, `last_tested_or_stamped`, `status`. `item_type` is exactly one of `gas storage`, `gas alarm`, `ventilation`, `optic`, `thimble measure`, `flowmeter`, `wine dispenser`, `stamped glass`, `self service dispense`, `other`. `status` is exactly one of `recorded`, `not found`, `not checked`. Anything other than `recorded` is listed in `gaps.md`.
6. `fixes.csv` with columns: `fix_ref`, `what_to_fix`, `cause_addressed`, `pints_recovered_per_week`, `basis_for_that_figure`, `cost_to_do_it`, `owner`, `group`. `group` is exactly one of `fix this week`, `fix this quarter`, `monitor`, and is supplied rather than decided. Sorted by `pints_recovered_per_week` descending within group.
7. `gaps.md` is a numbered list of: every window that is not a whole number of weeks; every part container not weighed; every product whose six allocation rows do not sum to the gap; every line with an assumed bore; every line with no sampled pint; every sampled pint with a shortfall; every safety or equipment item not recorded; every fix with no pints figure or no owner; and every figure the owner could not obtain, named individually.

## Rules
- Codex computes, measures against supplied readings, records and checks. It never allocates a cause, never decides a ranking group, never names a member of staff as responsible for a shortfall, and never fills a missing reading.
- Never estimate a yield loss. Where a reading was not taken the cell is empty, `measured_or_supplied` says `supplied by the owner` only where the owner supplied a number, and the missing reading goes into `gaps.md`.
- Never balance the gap allocation by adjusting a cause. Where the six rows do not sum to the gap, the difference goes into the `not yet identified` row, no other row is touched, and the product is listed in `gaps.md`.
- Never write an industry standard wastage percentage, a typical cellar loss, a normal draught gross profit or a benchmark yield. No such figure exists for a single independent UK venue and none is to appear in any file.
- Never combine a short measure finding with a yield finding in the same row or the same total. They live in different files and carry different consequences.
- Never recommend any change to how a drink is poured, how full a glass is filled, or when a top up is offered. The fixes file ranks what the owner supplied and nothing else.
- Never write a gas reading, a risk rating or a safety conclusion. The safety file records what was observed, who observed it and when.
- Never contact a supplier, a brewer, HMRC, Trading Standards or a health and safety adviser, and never submit anything anywhere.
- All money is in GBP and rounded to the penny, volumes are in litres or millilitres as the column names say, pints are recorded to two decimal places, and every computed figure shows the inputs it came from in the same row.
- Use British English, GBP and DD Month YYYY dates. No em dash characters 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 delivery notes, till data and equipment, and for their accountant, supplier, Trading Standards service or health and safety adviser to check. It applies supplied rules to supplied measurements and does not certify that anything is safe or compliant.

## Return
The absolute path of each file written, the row count of each CSV, the measurement window with its start and end dates and the number of weeks, the theoretical and achieved pints per container for every product with the gap in pints and in GBP, the six-way allocation for each product and the size of its not-yet-identified row, the computed line volume for every run with the measured or assumed bore named, the total pints a year lost to cleaning, every sampled pint with a shortfall named with its date, time and line, the count of safety and equipment items by status with each one not recorded named, the fixes ranked by pints recovered a week within their supplied groups, the list of figures that could not be obtained, and the `gaps.md` item count.