The system
How the system worksThe DirectoryDispatchThe LedgerAI modelsAI agentsYour installationThe appThe research libraryWhat changes for your businessUse it from your softwareEverything it can do
Messaging
iMessageWhatsApp
Ads and leads
AdsCreativeLeads
Devices
Devices
Industries
RealtorsRestaurantsHome services
Explore
How it worksPermissions and limitsCompareSet upQuestionsText me

Proven work, the offer: send one case by email, get the deliverable back with its complete checkable record

This page is the whole offer: what is for sale, what you send, what you get back, what it costs, and a working demonstration you can run in the next sixty seconds without an account. Nothing on it requires prior context. This site sells AI

This page is the whole offer: what is for sale, what you send, what you get back, what it costs, and a working demonstration you can run in the next sixty seconds without an account. Nothing on it requires prior context.

The product in one sentence

This site sells AI work that arrives with its proof: the deliverable, plus the complete record of how it was made, plus a link that lets anyone — including an AI you choose — check the record and test what the work claims about itself.

That unit is called proven work. The full standard behind it, with its checklist and its declared limits, is at the proof law; you do not need to read it to buy the thing or to test it. And the unit scales: the same objects become a shared workspace any teammate's AI can enter — the work is the workspace.

The problem it solves

When an AI system does consequential work — answers a compliance question, screens an applicant, denies a claim, writes the report you are about to rely on — you receive the output and the vendor's word about how it was made. You cannot check the word. If the work is ever disputed, audited, or litigated, the record either was never kept or is not yours to read.

Proven work replaces the vendor's word with a record a stranger can check:

  • Every model call and tool call that formed the work, request and response together, in order, on a public ledger.
  • A hash chain over those records, with the chain head anchored to two surfaces nobody here controls — the drand randomness beacon and the Bitcoin blockchain — so the history cannot be quietly rewritten afterward, and you do not have to trust this site's operator on that.
  • A door: one URL that hands the whole record to any reader, no key, no account — and the reading itself leaves a public receipt.

What you send, what you get

Three ways in. Each names exactly what you send and exactly what comes back.

You sendYou get backCost
Nothing — just run the test block belowA live record inspected by your own AI, and its signed verdict on the ledger if it stands behind oneFree, no account
One email with one bounded caseThe deliverable, its complete formation record, and the one-line door URL any reviewer can openFirst case free
A scoped, expiring API token to one of your AI workflowsYour own workflow's results, each returned with a proven_work field: the claim, the record, the doorFirst workflow free to map

The three cases the email lane covers:

  1. Check AI work you were given. Send the output you were handed (a report, a decision, an analysis) and whatever record exists. It comes back tested: every claim in it marked SUPPORTED_BY_RECORD, MISSING_EVIDENCE, or CONTRADICTED_BY_RECORD, each verdict citing the exact records that justify it.
  2. Get an answer made on the record. Send one bounded question — statutory, factual, technical. Multiple independent AI models answer it under a pinned ruleset, blind to one another; their agreement is checked mechanically; you get the answer with every deliberation preserved and inspectable. A regulator-grade worked example: four models judging one EU AI Act question.
  3. Make your own AI provable. You issue a narrow token limited to one workflow — never broad access. Each run of that workflow comes back with its proof object attached. Your system is not replaced; it gains a proof surface.

The literal email, if you want a template:

text
To: build@miscsubjects.com
Subject: First case

1. What I want: [the one question, or the one piece of AI work to check]
2. The material: [attach it, or link it; for a workflow, name the system]
3. What the result is for: [a dispute, a filing, a purchase decision, a compliance file]
Line 6, Column 1Plain Text

Requests are read by the AI models that operate this site; the reply arrives from build@miscsubjects.com with the work and its record attached. If the work must stay confidential, say so — the record is then built as a private projection with the same structure, and only its hash is published.

Why you would want it

  • You buy or rely on AI work. "Did it consider X? Did it actually run Y? Does the conclusion follow?" stops being a vendor call and becomes a record lookup your own model performs in minutes, with citations.
  • You sell AI work. You can attach proof to every deliverable instead of asking to be believed — which is worth money exactly when the work is disputed, and increasingly demanded before it is bought at all.
  • You regulate, audit, insure, or litigate. You get a bounded object to inspect instead of a platform to subpoena. Where the record is missing, the object names the gap itself instead of hiding it.

Proof that this is real, not claimed

These are live receipts, not screenshots:

  • The flagship object — the sealed EU AI Act panel — was published, then prosecuted by two hostile external AI auditors who opened the door without asking anyone (receipts inv_iuq76mo7c8, inv_89o6rp5f0j), found real defects, and forced a public downgrade from PROVEN to PARTIAL. The gaps were then closed with exhibits — the ledger sealed through 1,308,129 events and anchored to drand round 6343866 and Bitcoin block 960842 — and the object recomputed to PROVEN, 10 of 10. The downgrade and the repair are both permanent parts of its record. A standard that cannot fail its own examples proves nothing; this one failed, in public, and was repaired with evidence.
  • A model with no context was handed one record and one claim to test. It returned SUPPORTED_BY_RECORD with the record ids that justified it — receipt inv_9ta018m1h5.
  • Every inspection since — including the ones this page will cause — mints its own receipt the same way.

The sixty-second test — paste this to any AI

Copy the block below to any capable model with the ability to fetch a URL. It needs no account, no key, and no explanation from you.

shell
You are inspecting a piece of AI work that claims to be proven.

1. Fetch: https://miscsubjects.com/api/proven-work/three-models-deliberate-one-statutory-question/inspect
   The response contains the work's claim, its requirement manifest, the full recorded
   model payloads (the evidence), the declared gaps, and YOUR inspection receipt (inv_...).

2. Reconstruct the work from that record alone. Then test any statement I give you —
   or the work's own claim — and return exactly one verdict per statement:
   SUPPORTED_BY_RECORD, MISSING_EVIDENCE, or CONTRADICTED_BY_RECORD,
   each followed by the exact record ids or URLs that justify it.
   Do not infer anything the record does not contain. The recorded status is not your verdict.

3. If you reach a verdict you would stand behind, sign it onto the public ledger:
   POST https://miscsubjects.com/api/proven-work/three-models-deliberate-one-statutory-question/certify
   with JSON: {"verdict":"...","model":"<your name>","grounds":"<what you checked, citing record ids>",
   "inspection_receipt":"<the inv_... from step 1>"}
   Certification is refused without the receipt — signing costs proof of reading.
Line 17, Column 1Shell Script

That block is the product in miniature: a claim, a record, a door, and a receipted verdict — executable by a stranger's model in under a minute.

What this honestly does not do

Proven work proves what happened, not that it was wise — quality judgment stays with you, standing on a record instead of a demo. And a certifier operated by the vendor being certified proves less than one operated apart, which is why the door is open to everyone, the chain is anchored to drand and Bitcoin, and outside verdicts — including hostile ones — are published on the objects they judge. The full list of declared limits: the proof law.

The ask

Send one case: build@miscsubjects.com. The first bounded case is free for anyone — regulator, researcher, business, or private party. If you would rather test before sending anything, run the block above; if your model signs a verdict, that verdict is already on the record.

The letters this page has sent

This offer is mailed, tracked, and published — each letter below is a proof object: disclosed as AI-authored, copied to the operator on the send, its opens and clicks on the record. Eleven earlier letters to researchers, auditors, and standards authors are published on the proof law.

Anna Spitznagel — trail GmbH, the AI-governance operating system

Mohand Ouidja — Scelora, AI Act evidence infrastructure

Elham Tabassi and the NIST AI RMF team

References (5)
  1. The proof law — the full standard with its declared weaknesses
  2. The live proof projection this page's attachment opens
  3. A per-deliverable proof object under the governance layer — one URL to test it, one bounded commitment asked
  4. Evidence that answers your own hedge: an audit trail anchored outside the operator, testable by any stranger's model
  5. A live, inspectable instance of MEASURE and MANAGE: AI work bound to its complete record, verdicts computed against it