Use Finity and share a receipt.
Frame one real decision with the seven prompts. Report the boundary, next move, observed result, and what you revised. A null or failure is welcome.
Participation [B] · useful disagreement welcome
Use it. Test it.
Make it answerable.
Start as a person, not a member: use Finity and share what happened, attack a public claim at its strongest point, or contribute bounded evidence. Each route opens a public GitHub form. This static site submits nothing itself.
No adherence requiredUse one tool, report one consequence, or leave.
Disagreement is productiveA precise contradiction is more valuable than generic assent.
No epistemic purchaseMoney, status, model count, and affiliation cannot raise a claim's tier.
Three human routes · public by default
Bring a receipt.
Or bring a better objection.
These links open prefilled forms in the public repository. Do not include secrets, private data, medical information, or anything you cannot publish. A public issue is a custody item. Filing alone does not establish independent review, validation, or a tier upgrade; submitted material is assessed at its own provenance and limits.
Attack a public claim.
Quote the strongest version, identify the failure, offer the best rival, and name the evidence that would decide between them.
Contribute bounded evidence.
Bring one source, dataset, replication, native-domain correction, or discriminator. State its custody, limits, and what it cannot establish.
Participation boundary
Participate without enrolling.
This page offers public ways to test, criticize, or contribute evidence. It creates no membership, funding relationship, assent obligation, or transfer of authority. You may use a method, submit a public issue, or leave.
No study, funding programme, grant scheme, compute service, or private-data intake is open.
Optional technical route
Read the research protocol.
Keep custody.
Choose an open research socket. This is a public planning contract, not an active research programme or compute service. It does not accept payments, credentials, private data, or live model jobs.
Never paste or send a personal API key.
Credentials remain with their owner or in a restricted server-side project. Any future compute service would require bounded credits or project grants—not reusable secrets in a browser form. This release offers no such service.
Provider guidance: OpenAI key safety ↗ · Google Gemini key safety ↗
Loading contribution modes…
Compute envelope
Any future run needs a hard edge.
If a future contribution route is opened, purpose, cost, tools, data, publication, expiry and revocation must be explicit before execution. No such route is open in this release.
Who and why
Principal, declared purpose, question identifier, permitted providers and attribution.
- No donor truth authority
- No hidden secondary purpose
- No application outside scope
What may be spent
Maximum currency spend, tokens, runs, rate, models and expiry.
- Hard server-side limits
- Visible cost receipt
- Automatic stop at exhaustion
What may be seen
Allowed sources, data classification, tools, retention and deletion route.
- No secrets in prompts or logs
- No arbitrary public tool execution
- No undisclosed training use
What must be published
Prompt and model provenance, outputs, disagreements, failures, costs and review status.
- Nulls remain visible
- Sponsor cannot suppress results
- Revocation stops future runs
Candidate packet
A model answer is raw intake.
Every useful answer must compile into the same inspectable packet before it enters the public laboratory.
candidate-packet.txt
nativeSourceAndDomain: nativeDefinitions: candidateSocket: claimAndTier: formalMap: assumptions: strongestRivals: noPlacementRival: nonCircularDiscriminator: killCriterion: survivorIfKilled: dataCodeAndCitationCustody: modelProviderAndVersion: promptAndRunDate: computeCost: independentReviewRoute: