Methodology
How this works.
The rules we hold ourselves to: what gets recorded, what gets scored and what deliberately doesn’t, how outcomes are decided, and how to challenge any of it. These are commitments, not descriptions — if we break one, hold us to it.
What gets recorded
Public statements by public figures and organizations, from a public source. We do not record private individuals, and we do not score them under any circumstances.
Every record must quote its source exactly. The wording is checked mechanically against the response the platform itself returned; a quote the source does not contain is refused rather than reviewed. How that check works.
The three tiers
What can be resolved depends on what was said. Forcing everything onto one scale is how accountability projects end up manufacturing false precision.
| Tier | Example | How it resolves |
|---|---|---|
| A — Objective | “Bitcoin will pass $200,000 before 2027.” | From data. No human judgment involved. |
| B — Verifiable | “We will ship this feature by Q3.” | Evidence gathered, a named reviewer decides, reasoning published. |
| C — Contextual | “We will make the country stronger.” | Never scored. Preserved with its source, and that is all. |
Tier C is enforced in code: the system rejects resolution criteria attached to a Tier C record. A number that looks precise but isn’t is worse than no number.
The words we use for outcomes
Verdicts describe evidence, never character. We do not publish “lied,” “broken promise,” or an honesty score for a person.
- Correct / Incorrect — a Tier A prediction measured against data.
- Fulfilled / Partially fulfilled — a commitment met, or met in part.
- No evidence of fulfillment as of [date] — we looked and found none. This is a statement about the evidence, not an accusation.
- Superseded — circumstances changed the question.
- Unresolvable — the criteria cannot settle it.
Disputes
Anyone can dispute a record or a resolution. The person a record is about gets a priority lane — being able to answer on your own record is the fairest part of this design.
| Stage | Commitment |
|---|---|
| Acknowledgement | Within 72 hours of filing |
| Resolution or status | Within 14 days |
| Publication | The outcome is appended to the record’s history and anchored — whether it goes against us or not |
| Re-dispute | Permitted with new evidence |
Disputes must cite something specific: a misquote, a wrong date, a misattributed source, a resolution the evidence doesn’t support. Disagreeing with what someone said is not grounds — the record is that they said it.
That a dispute exists is public the moment it is filed. The claim text is published when it is reviewed, not on filing: an open submission box whose contents are instantly permanent and anchored would be a way to publish abuse through us.
Corrections
We will get things wrong. When we do, the original record is not edited — the correction is appended, the original stays visible, and both are anchored. You can always see what we said before, what changed, and why.
A project that quietly fixed its mistakes would be asking you to trust that there weren’t any. The permanent trail is the point.
Mirror the archive
The complete archive is downloadable at /api/v1/export, including a digest of every record it contains. That digest is committed to the blockchain in every anchor batch.
If you keep a copy and a record later disappears, your copy and the anchored digest will disagree with ours — provably, without needing our cooperation. Third parties holding copies is the only real answer to “you could quietly drop the records you don’t like.”
We cannot make anyone do this, and as far as we know nobody has yet. Until someone independent holds a copy, the guarantee is a promise rather than a proof. If you are a journalist, researcher or archivist reading this: mirroring costs you one HTTP request and turns our word into something you can check.
Not built yet
Stated plainly so nobody assumes otherwise: automatic settlement covers only a small set of allowlisted metrics, and every other record carries a locked test that a person has to read. There is no named external reviewer pool, and dispute resolution is handled by hand. Records are anchored to Base mainnet, and every record states which network its batch went to.