weather-play

---
name: weather-play
description: Produces one dated two-branch operating plan, written before the forecast arrives: what the venue does in a heatwave and what it does in a washout, with numbered triggers tied to named Met Office and UK Health Security Agency alerts, the cold chain and staff welfare controls each branch needs, the money attached to each move, and the person who decides. Use in the fortnight before a summer or bank holiday peak, or the first time a hot spell or a storm costs you a service you had staffed for.
---

# The plan for a heatwave and for a washout, written before either

You give this your site layout, your staffing pattern for the peak, your covers by area, and your last two years of takings for the same period. You get back one dated file with two branches: a hot weather plan and a wet weather plan, each with numbered triggers stated as named alert levels or measured thresholds, each move costed, and a named decider with a decision time. It will not forecast the weather. It makes sure that whichever way it goes, nobody is improvising at nine in the morning with a full rota already booked.

## What it does

1. **Write down what you are protecting, in order, before you write a single trigger.** The order is people, then stock, then the day's takings, and it has to be written down because under pressure it inverts. Staff welfare comes first, the cold chain second because a failure there is both a safety issue and a cash one, and revenue third. Then list what is exposed at your site: which areas have no shade, which fridges sit next to a hot line, which seating is unusable in rain, how many covers each area holds, and what proportion of a peak day's takings sits outside. A venue taking forty per cent of a bank holiday outside runs a different risk from one taking five per cent, and the plans that follow should not look the same.

2. **Take the forecast from a named source on terms you can actually use, and write the source into the plan.** The Met Office Weather DataHub is described as "the single point of access for all Met Office Public Task weather data". It offers site specific spot data as GeoJSON for a given latitude and longitude at hourly, three hourly and daily resolution, a land observations API giving recent historical weather from ground based instruments, and National Severe Weather Warning Service warnings in Atom format. The support pages state that "The Met Office has chosen to support the free provision of data up to a usage limit for some APIs", that an API key is provided after subscribing, and that use carries an attribution requirement of "Powered by Met Office data". Record which product you are on, the key holder and the attribution wording, because a feed nobody owns fails silently.

3. **Learn what the warning colours actually mean before you write triggers against them.** The Met Office combines "a combination of both the impact the weather may have and the likelihood of those impacts occurring". Amber means "There is an increased likelihood of impacts from severe weather, which could potentially disrupt your plans", and you "should think about changing your plans and taking action to protect yourself and your property". Red means "Dangerous weather is expected and, if you haven't already done so, you should take action now to keep yourself and others safe". Yellow is the one that gets misread: the Met Office says some yellow warnings cover low level disruption while others mean "the weather could bring much more severe impacts to the majority of people but the certainty of those impacts occurring is much lower", and that you should "read the content of yellow warnings to determine which weather situation is being covered". So a trigger cannot say "on a yellow warning". It has to say which yellow.

4. **Register for the health alerts as well, because they arrive earlier and they are about people rather than weather.** The Weather-Health Alerting System is run by the UK Health Security Agency in partnership with the Met Office, and GOV.UK describes it as providing "an early warning when adverse temperatures are likely to impact on the health and wellbeing of the population". Heat-Health Alerts run "from 1 June to 30 September" and Cold-Health Alerts "from 1 November to 31 March", with extraordinary alerts possible outside those windows. Alerts go by email to those who have registered, free, through the GOV.UK page. Put a named person on the distribution list rather than a shared inbox nobody opens, and a second for when the first is away, which is guaranteed to be during the heatwave.

5. **Build the hot weather branch around the two things that actually break: the cold chain and the people.** On stock, Schedule 4 of the Food Safety and Hygiene (England) Regulations 2013 makes it an offence to keep food that supports pathogen growth above 8 degrees Celsius, with a tolerance for holding above that "for service or on display for sale for a period of less than four hours". A fridge that holds 4 degrees in March is not the same fridge in a 33 degree kitchen with the door opening every ninety seconds, so the hot branch names the extra checks, who does them and when, and what happens to a load that fails one. On people, the Health and Safety Executive states that "There is no maximum temperature for workplaces" and that employers must assess temperature related risks and put controls in place.

6. **Write the hot weather controls as specific moves, using the list the regulator actually published.** In a press release of 19 June 2026 the Health and Safety Executive said that "Employers must assess risks to the health and safety of their workers by law, including risks from extreme weather such as heatwaves", that heat "is classed as a hazard and should be treated like other hazards", and that employers should discuss changes with workers. The controls it names include adequate ventilation and shade, enough breaks for workers to cool down, blinds or reflective film, moving workstations out of direct sunlight, free drinking water, relaxed dress codes and flexible patterns so work happens during cooler times. Turn each into a line with a cost and an owner: who buys the parasols, who moves the prep to six in the morning, who covers the extra breaks.

7. **Write the wet weather branch as pre-committed moves with money attached, not as a hope that people will come inside.** Outside covers do not move indoors, they cancel, so the wet branch is about recovering some of the margin rather than all of it. Name the moves in advance: which staff are stood down and how much notice their contract requires, which deliveries are cut and by when the supplier must be told, which menu lines are pulled because they were ordered for the terrace, whether outside furniture has to come in and who does it, and what goes out to the booking list and in whose name. Each line carries a cash figure. A wet weather plan without numbers is a list of good intentions and it will lose to the instinct to keep everyone on and hope.

8. **Set every trigger as a number or a named alert, with a decision time and a single decider.** A trigger reads like this: if a Met Office amber warning for rain covering our postcode is live at 08:00, or the site specific forecast shows more than a stated millimetre total between 12:00 and 20:00, the named manager stands down two of the six on the terrace rota by 08:30 and tells the supplier by 09:00. Two things make it work. The time, because a decision available all day gets made at four in the afternoon when the wages are committed. And the single name, because a trigger owned by the team is owned by nobody. Write the fallback too, for when the person is unreachable and for when the forecast changes after the decision point.

9. **Build the baseline that lets you argue about this next year, and date the file.** The Met Office publishes free UK and regional climate series covering maximum, minimum and mean temperature, sunshine, rainfall, rain days of 1.0mm or more and days of air frost, as monthly, seasonal and annual values for the UK, the four nations and regions including the Midlands and East Anglia, under Crown copyright. Pull the values for your region for the peak month for the last several years and put them next to your own takings. That is the only way to test the weather excuse honestly. Some poor weeks really were the weather and some were not, and a venue that cannot tell keeps fixing the wrong thing. Date the file and each source.

## Then it checks

1. The file states in order what is being protected, and lists the exposed areas with covers and the proportion of peak takings that sits outside.
2. Every trigger names a specific Met Office or UK Health Security Agency alert level with the hazard it covers, or a measured threshold with its unit, and none reads simply "a yellow warning" or "if it looks like rain".
3. Every trigger carries a decision time, a single named decider, a named fallback, and what happens if the forecast changes after the decision point.
4. Every move in both branches carries a cash figure and an owner, and none reads "as needed" or "reduce staff".
5. The hot branch names the extra cold chain checks with times and a named checker, and states what happens to a load that fails one, rather than asserting the fridges will cope.
6. The forecast source is named with its product, key holder and attribution wording, the health alert registration names two people, 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 message to the booking list about a closed terrace goes out under the owner's name and goes to them first.
- Never produce a weather forecast, never restate one as though it were yours, and never publish one to customers. Point at the Met Office forecast and let the customer read it.
- Never state a typical rainfall effect on trade, a sector average for outdoor covers or any figure claiming what a degree of temperature does to sales. No such figure exists for one independent venue, and a plan built on one is planning somebody else's site.
- Refuse to write a trigger that tells staff to continue outdoor service during a live red warning, however it is framed. The Met Office advice on a red warning is to take action to keep yourself and others safe, and a plan that overrides a regulator's safety advice for a day's takings is not one this skill will write.
- Never treat a temperature target as a substitute for a risk assessment. There is no maximum workplace temperature in law, so the control is the assessment and the measures, not a number on a thermometer.
- Never write a trigger without a decision time. A decision that can be made all day is made too late.
- This output is a working document prepared for the owner to check against their own site, rota and contracts, and for their health and safety adviser and environmental health officer to check where it touches staff welfare or food safety. It applies published guidance to an operating plan; it is not a risk assessment and it does not certify that the venue is safe in any weather.

## Built from
- Met Office, "Met Office Weather DataHub", https://www.metoffice.gov.uk/services/data/met-office-weather-datahub, no publication date shown on the page, read 16 September 2026: the single point of access description and the site specific spot, land observations and warnings products in step 2.
- Met Office, "Weather DataHub: Support and FAQs", https://datahub.metoffice.gov.uk/support/faqs, no publication date shown on the page, read 16 September 2026: the free provision up to a usage limit, the API key on subscription and the "Powered by Met Office data" attribution in step 2.
- Met Office, "Weather warnings guide", https://weather.metoffice.gov.uk/guides/warnings, no publication date shown on the page, read 16 September 2026: the impact and likelihood matrix and the verbatim meanings of yellow, amber and red behind step 3 and the refusal in the Rules.
- GOV.UK, "Weather-Health Alerting System", https://www.gov.uk/guidance/weather-health-alerting-system, published 1 June 2023, last updated 22 April 2026, read 16 September 2026: the UKHSA and Met Office partnership, the 1 June to 30 September and 1 November to 31 March seasons and the email registration in step 4.
- The Food Safety and Hygiene (England) Regulations 2013, Schedule 4, https://www.legislation.gov.uk/uksi/2013/2996/schedule/4/made, read 16 September 2026: the 8 degrees Celsius offence and the four hour service tolerance behind the cold chain checks in step 5.
- Health and Safety Executive, "Temperature in the workplace: What the law says", https://www.hse.gov.uk/temperature/employer/the-law.htm, no publication date shown on the page, read 16 September 2026: no maximum workplace temperature, the 16 and 13 degree minima and the risk assessment duty in step 5.
- Health and Safety Executive, "Risks to workers from extreme heat must be managed", https://press.hse.gov.uk/2026/06/19/risks-to-workers-from-extreme-heat-must-be-managed/, published 19 June 2026, read 16 September 2026: heat as a hazard to be assessed, and the named control measures that become the costed lines in step 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: the free variables, regions and monthly resolution used for the baseline in step 9.

# weather-play

## You are given
The site layout of one UK hospitality business as a list of areas, each with its name, whether it is `indoor`, `covered outdoor` or `open outdoor`, its cover count, whether it has shade, and the proportion of a peak day's takings it carries. The staffing pattern for the peak, per area, with roles, hours and hourly cost. The refrigeration list with, per unit, its location, what it holds, its normal holding temperature and whether it sits near a heat source. The last two years of takings for the same calendar period, by day, net of VAT. The Met Office regional climate series already downloaded for the venue's region as a CSV. The two branch plans already written by Claude and approved by the owner, each as a list with the move in the owner's own words, its cash effect, its owner and its notice period. The trigger set already decided, each with its alert level or measured threshold, its unit, its decision time, its named decider and its named fallback. The forecast product, its key holder and the attribution wording. Every judgement about what the venue will do in each branch has already been made. You do not invent a move, a trigger or a threshold.

## Produce
Write into a `./weather-play-output/` folder:

1. `exposure.csv` with these columns in this order: `area_name`, `area_type`, `covers`, `has_shade`, `percent_of_peak_takings`, `takings_at_risk_gbp`, `staff_scheduled`, `staff_cost_gbp`, `weather_sensitive`. `area_type` is exactly one of `indoor`, `covered outdoor`, `open outdoor`. `weather_sensitive` is `yes` for every `open outdoor` row and for any `covered outdoor` row with `has_shade` of `no`. `percent_of_peak_takings` across all rows must total 100 and any shortfall or excess is listed in `gaps.md`.
2. `triggers.csv` with columns: `trigger_ref`, `branch`, `alert_or_threshold`, `source`, `hazard_covered`, `measured_value`, `unit`, `decision_time`, `decider_name`, `fallback_name`, `action_ref`, `what_if_forecast_changes`, `trigger_is_specific`. `trigger_ref` is T001 upward. `branch` is exactly one of `hot`, `wet`. `source` is exactly one of `Met Office warning`, `Met Office site specific forecast`, `UKHSA Weather-Health Alert`, `own measurement`. `trigger_is_specific` is `no` where the alert is a Met Office colour with no named hazard attached, and every such row is listed in `gaps.md`.
3. `moves.csv` with columns: `action_ref`, `branch`, `move_verbatim`, `owner_name`, `notice_period_hours`, `cash_effect_gbp`, `cash_effect_direction`, `latest_time_to_act`, `depends_on_trigger_ref`. `cash_effect_direction` is exactly one of `saving`, `cost`, `lost revenue`. `move_verbatim` is copied exactly from the owner and never reworded. Any row with an empty `cash_effect_gbp` or an empty `owner_name` is listed in `gaps.md`.
4. `cold-chain-checks.csv` with columns: `unit_id`, `location`, `holds`, `normal_temperature_c`, `near_heat_source`, `check_times`, `checker_name`, `action_if_above_8c`, `four_hour_tolerance_applies`. One row per refrigeration unit. `action_if_above_8c` is copied verbatim from the supplied plan. Any row with no `checker_name` or an empty `action_if_above_8c` is listed in `gaps.md`.
5. `weather-and-takings.csv` with columns: `year`, `month`, `mean_temp_c`, `max_temp_c`, `rainfall_mm`, `rain_days_1mm_or_more`, `sunshine_hours`, `takings_net_gbp`, `takings_change_on_prior_year_percent`. One row per month of the supplied series where takings also exist. Rows with a climate value but no takings, or the reverse, are written with the missing field empty and listed in `gaps.md`.
6. `branch-summary.md` with, for each of the two branches, the total cash effect in GBP, the count of moves, the count of triggers, the earliest decision time, and the takings at risk from the weather sensitive areas. Figures only, copied from the CSVs. No commentary and no recommendation.
7. `feed-register.md` recording the forecast product name, the key holder, the attribution wording exactly as supplied, the two named people on the Weather-Health Alert distribution, and the date the registration was confirmed. Written empty apart from a note in `gaps.md` where any of these were not supplied.
8. `gaps.md` as a numbered list of: any trigger with no decision time, no decider or no fallback; any trigger marked `trigger_is_specific` of `no`; any move with no cash figure or no owner; any refrigeration unit with no checker or no action on failure; any area whose takings percentage is missing; a takings percentage total that is not 100; any month with climate data but no takings; a missing attribution wording; fewer than two named people on the alert distribution; and any figure that could not be computed because an input was missing.

## Rules
- Codex records, computes and checks. It never writes a move, never sets a trigger, never chooses a threshold, never rewords the owner's plan, and never produces or restates a weather forecast.
- Never write a trigger action that continues outdoor service during a red warning. Where such a move is supplied, record it verbatim, mark it in `gaps.md` as contradicting the published safety advice, and do not compute a cash benefit for it.
- Never infer a cash effect for a move that has none supplied. Leave the field empty and list it in `gaps.md`. A move with no number is not a move with a number of zero.
- Never write a sector average, an industry figure for weather sensitivity, or any claim about what a degree of temperature or a millimetre of rain does to trade. No such figure exists for one independent venue and none is to be written into any file.
- Climate values are copied from the supplied Met Office series unchanged. Never interpolate a missing month, never average across regions, and never adjust a value.
- Never call a weather API, never sign up for a service and never publish anything to customers. These files are drafts for the owner.
- Correlation between a climate column and a takings column is reported as the two columns side by side and nothing more. Never state that weather caused a change in takings.
- Use British English, GBP, DD Month YYYY dates, 24 hour times, degrees Celsius and millimetres. 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 site, rota and contracts before it is relied on, and for their health and safety adviser and environmental health officer to check where it touches staff welfare or food safety. It is not a risk assessment and it does not certify that the venue is safe in any weather.

## Return
The absolute path of each file written, the row count of each CSV, the number of triggers in each branch and how many are marked not specific, the total cash effect of each branch in GBP split into savings, costs and lost revenue, the takings at risk from weather sensitive areas in GBP and as a percentage of a peak day, the earliest decision time in each branch, the refrigeration units with no named checker, the months where climate data and takings could be paired, the forecast product and attribution wording as supplied, the two names on the alert distribution, and the `gaps.md` item count.