PROOF SELFTEST 8/8 · witnessed in Claude 2026-10-11 · Edition 1.
THE SEAL 98 of 100 · LEGENDARY · WITNESSED · Trigger 5 · Machinery 5 · Law 5 · Portability 5 · Proof 5
IT CAN BE WRONG an AI reading a text file: it can be wrong and can invent a number, a rule, a name or a date; check what matters against your own records and the source; the risk is yours. THE-CLAUSES.md rides in the packet and is part of the licence.
THE PUNCH · SHA-256 OF SKILL.MD · E7EBB48A4AB9723A2CBCCC6E
CITE THIS CARDTHE COMPUTE BUDGET CARD, Edition 1. Lund Studio LLC, Westminster, Colorado. https://lundr.tech/?view=compute-budget-card. Punch E7EBB48A4AB9723A2CBCCC6E. Retrieved 2026-10-11.
What it is.
Energy and compute per AI task, reported and capped. sheet writes a blank figures sheet (watt-hours per 1,000 tokens for each model, watts for each GPU, a facility overhead, and where the figures came from and when) for the owner to fill; report adds up a usage log of tokens or GPU-seconds by task, day or model in watt-hours; cap holds each task and each day to a budget: OK, WARN from the warning line, OVER past the cap, UNKNOWN where a figure is missing; estimate works out a planned job before it runs. The card never invents a provider's figure: a model with no figure stays unknown, and a sheet that does not say where its figures came from is refused. It waits on providers reporting energy per call; until then the totals are arithmetic on the owner's estimates.
It carries a machine that shows its working: sheet — the blank figures sheet; report — Wh per task, day or model; cap — OK, WARN, OVER or UNKNOWN; estimate — a planned job, worked out — 8 checks in its selftest, every command on the record. It will never invent a provider's figure; never fill in a figure that is missing; never call an estimate a measurement.
Six laws, written into the card.
They are in the card itself, so the AI follows them.
THE OWNER'S FIGURESEvery figure comes from a source the owner names, with its date; a sheet without them is refused.
UNKNOWN STAYS UNKNOWNA model or GPU with no figure is reported as unknown and never guessed.
PLAIN SUMSTokens × Wh per 1,000 ÷ 1,000, or watts × seconds ÷ 3,600, times the overhead: watt-hours, with the working shown.
WARNED BEFORE THE CAPWARN from the warning line and OVER past the cap; UNKNOWN is never OK.
PER TASK AND PER DAYBoth are reported, and both can be capped.
ARITHMETIC, NOT A METERThe totals are as good as the figures, and every run names where the figures came from.
What rides in the card.
In the card’s own voice, with a machine that shows its working.
THE GUIDESreferences/HORIZON-TWO.md · THE-COMPUTE-BUDGET.md
THE MACHINEcomputebudget.py · four commandssheet — the blank figures sheet; report — Wh per task, day or model; cap — OK, WARN, OVER or UNKNOWN; estimate — a planned job, worked out. Every command prints its working and refuses a bad input; 8 checks in its selftest, every command on the record.
What it does once your AI has it.
sheet The blank figures sheet
report Wh per task, day or model
cap OK, WARN, OVER or UNKNOWN
estimate A planned job, worked out
On the record Five usage lines on the owner's figures (overhead 1.1) come to 429.1 Wh: the fine-tune 385 Wh, contract review 41.6 Wh, the inbox summary 2.48 Wh, and image tags unknown because the sheet has no figure for its model; against 400 Wh a task and 450 Wh a day the fine-tune and 10 Oct (415.2 Wh) are WARN and the unknowns read UNKNOWN, never OK. A planned job of 1,000 calls is 1.155 kWh, OVER a task's budget; a sheet with no source is refused.
What it will never do.
It will never invent a provider's figure; never fill in a figure that is missing; never call an estimate a measurement.
It waits on providers reporting the energy of each call. Until then every figure is the owner's estimate, and the totals are arithmetic on it, not a measurement. It reads the usage log it is given, and work not logged is not counted. Energy only: for cost, THE TOKEN COST CARD. It is an AI reading a text file: it can be wrong, so check what matters against your own records and the source.
What’s in the buy.
CARD.md — the whole card: who it is, its laws, when it says no, what it carries · SKILL.md — the door your AI reads first · START-HERE.md — the first five minutes, for you
HORIZON-TWO.md + THE-COMPUTE-BUDGET.md — the guides in references/, in the card's own voice
scripts/computebudget.py — the machine: sheet · report · cap · estimate, standard library only, selftest 8/8
WITNESS.md + SELFTEST.md — every command on the record, as it printed · MANIFEST.md, every file hashed · AUDIT.md, the twelve checks
MY-LEDGER.md — the card's memory, on your own machine · TIN.md · LICENSE-NOTE.md · THE-CLAUSES.md · HOSTS.md
card.svg + card-machine.svg + card.json · card.html, the card as one offline file · the doors for ChatGPT, Gemini, Cursor, Claude Code and MCP hosts (rules/, system/, mcp/)
PROOF · SELFTEST 8/8 · 4 COMMANDS ON THE RECORD · WITNESSED IN CLAUDE 2026-10-11 · EDITION 1 · PUNCH E7EBB48A4AB9723A2CBCCC6E
ON EVERY DOOR · A CARD FOR YOUR AI, A FILE FOR YOUR COMPUTER
In Claude: turn on code execution under Settings › Capabilities, then Customize › Skills › + › Create skill, and upload the zip as it is — START-HERE.md in the box walks you through it, and tells you how to keep the ledger. In ChatGPT, Gemini and the rest: a Project, a GPT or a Gem with the card’s files attached and the short door in the box pasted as its instructions. The words run anywhere an AI reads Markdown; the machines run wherever the host runs Python. Witnessed in Claude only. The host marks →
Take it.
Nº 1173 PROPOSED — LAUNCHING SOON
Sold as is under the term printed. A card instructs an AI you operate; results depend on your model, your files and your judgement. Not medical, legal, tax or financial advice, never the person’s boss, and not a licence to practise. The terms →
One payment. The file is yours to keep and keeps working at its edition; new editions issued in the year after you buy reach you by email as replacement files. Tax is added at checkout where it applies. Nothing auto-renews.
A TERM, NOT A SUBSCRIPTION: NOTHING RENEWS ITSELF AND NOBODY IS CHARGED AGAIN. WHEN A TERM ENDS THE FILES STAY ON YOUR COMPUTER AND STAY READABLE — FREEZE, NOT LOSS.
THE MACHINE · IN THE OPEN · FROM THE WITNESSED RUN OF 2026-10-11
computebudget.py, every command showing its working.
A small Python program, standard library only. Your AI runs it; you can too. Every block below is pasted from the card’s own WITNESS.md, as printed.
sheet
the blank figures sheet
report
Wh per task, day or model
cap
OK, WARN, OVER or UNKNOWN
estimate
a planned job, worked out
selftest
8 checks that prove the machine; it must print 8/8
sheet · THE BLANK FIGURES SHEET
$ python3 scripts/computebudget.py sheet --out blank.json
a blank figures sheet written to blank.json
every figure starts empty: the owner fills each one from a source they can name, with its date
report · WH PER TASK, DAY OR MODEL
$ python3 scripts/computebudget.py report usage.jsonl --figures figures.json
usage.jsonl · 5 lines · by task
contract-review 102,000 tokens 0 GPU-s 41.6 Wh
fine-tune 0 tokens 1,800 GPU-s 385 Wh
image-tags 10,000 tokens 0 GPU-s energy unknown
inbox-summary 45,000 tokens 0 GPU-s 2.48 Wh
total 429.1 Wh · plus 1 line with no figure, left unknown
no figure in the sheet for: model-new
figures: the owner's meter readings and the provider's published estimate, copied by hand, as of 2026-10-01 · overhead 1.1
report · WH PER TASK, DAY OR MODEL
$ python3 scripts/computebudget.py report usage.jsonl --figures figures.json --by day
usage.jsonl · 5 lines · by day
2026-10-10 113,000 tokens 1,800 GPU-s 415.2 Wh
2026-10-11 44,000 tokens 0 GPU-s 13.9 Wh + 1 line unknown
total 429.1 Wh · plus 1 line with no figure, left unknown
no figure in the sheet for: model-new
figures: the owner's meter readings and the provider's published estimate, copied by hand, as of 2026-10-01 · overhead 1.1
cap · OK, WARN, OVER OR UNKNOWN
$ python3 scripts/computebudget.py cap usage.jsonl --figures figures.json --budget budget.json
each task against 400 Wh (warning at 80%)
OK contract-review 41.6 Wh (10%)
WARN fine-tune 385 Wh (96%)
UNKNOWN image-tags energy unknown: no figure for any line, so the cap cannot be checked
OK inbox-summary 2.48 Wh (1%)
each day against 450 Wh (warning at 80%)
WARN 2026-10-10 415.2 Wh (92%)
UNKNOWN 2026-10-11 13.9 Wh (3%) + 1 line with no figure: the cap cannot be checked
verdict: UNKNOWN: a figure is missing
figures: the owner's meter readings and the provider's published estimate, copied by hand, as of 2026-10-01 · overhead 1.1
estimate · A PLANNED JOB, WORKED OUT
$ python3 scripts/computebudget.py estimate --figures figures.json --model model-large --tokens-in 2000 --tokens-out 500 --calls 1000 --budget budget.json
model-large · 1.16 Wh a call × 1,000 calls = 1.155 kWh
for scale: 1.155 kWh would run a 10 W bulb for 115.5 hours
against 400 Wh a task: OVER
figures: the owner's meter readings and the provider's published estimate, copied by hand, as of 2026-10-01 · overhead 1.1
estimate · A PLANNED JOB, WORKED OUT
$ python3 scripts/computebudget.py estimate --figures figures.json --model model-new --tokens-in 2000 --tokens-out 500
UNKNOWN · there is no figure for model-new in the sheet: the card will not guess one
report · WH PER TASK, DAY OR MODEL
$ python3 scripts/computebudget.py report usage.jsonl --figures nosource.json
refused: the figures sheet must say where its figures came from (source) and when (as_of); the card never supplies a figure
THE NUMBERS, NAMES AND DATES IN THESE RUNS ARE THE STUDIO’S OWN EXAMPLES, INVENTED FOR THE RECORD; NO REAL COMPANY, PERSON OR PROPERTY IS NAMED.
QUESTIONS AT THE TABLE
Asked before the wire.
What does it need from me?
Sheet · the owner's figures, sourced. It asks first for the usage log (task, date, and tokens per model or GPU-seconds per GPU), the figures with their source and date, and the budget per task and per day in watt-hours.
Does it send anything anywhere?
No. The machine is a standard-library Python program that runs on your own computer and sends nothing anywhere; the words run wherever an AI reads Markdown. No server, no account, no key.
What is on the record?
The self-test (8/8) and every command run on real example inputs, pasted into WITNESS.md as they printed; the punch on the face is the SHA-256 of SKILL.md, so a downloaded copy can be checked against this room.
Can it be wrong?
Yes — it is an AI reading a text file. It waits on providers reporting the energy of each call. Until then every figure is the owner's estimate, and the totals are arithmetic on it, not a measurement. It reads the usage log it is given, and work not logged is not counted. Energy only: for cost, THE TOKEN COST CARD.
What will it never do?
It will never invent a provider's figure; never fill in a figure that is missing; never call an estimate a measurement.
