# Rule set: was this specific record in that dataset?

slug: ruleset-dataset-membership · https://miscsubjects.com/a/ruleset-dataset-membership · category: adjudication · tags: ruleset, adjudication, evidence, self-authored · updated 2026-07-30T00:29:42.543Z

A rule set is the load-bearing object in an adjudication, so it is published as an object: versioned, content-addressed, and citable by the exact hash a finding was made under. When these rules change, every prior finding stays legible against the rules it was actually made under.

## Provenance of these rules

Declared provenance: **self-authored**. Written by the operator and deliberately biased toward CANNOT_CONCLUDE, because the failure mode being guarded against is a model asserting membership from resemblance.

A reader who does not trust this operator can price a finding by this field alone: rules written elsewhere bind harder than rules written by the party who benefits from the finding.

## Question

Does the supplied evidence establish that the specific record was present in the named dataset?

## Rules

1. AFFIRM only on a direct identifier match documented in the supplied evidence: an exact record, a hash, or an index entry.
2. DENY only if the evidence positively excludes the record, for example a documented complete enumeration that does not contain it.
3. CANNOT_CONCLUDE for statistical resemblance, partial-field matches, format matches, or any inference from similarity. Similarity is not membership.
4. Absence from the supplied evidence is not absence from the dataset unless the evidence is a documented complete enumeration.
5. Never treat a model's ability to produce a similar-looking record as evidence of membership.
6. Quote the span relied on. If the finding rests on absence, SPAN is NONE and the rationale must state what enumeration was or was not available.

## Permitted verdicts

`AFFIRM` · `DENY` · `CANNOT_CONCLUDE`. Abstention is first class: a panel that cannot conclude says so, and that recorded absence is itself evidence rather than a silent null.

## Content hash

The canonical form is the JSON object `{id, version, question, rules, verdicts}` with no whitespace. SHA-256:

`427e366b460fa21320ce89cd2eb223304e319d8ec57c4ee66e3a998dee966a3d`

Recompute it from the canonical form below and compare. A finding that names a different hash was made under different rules.

```json
{"id":"ruleset-dataset-membership","version":"1.0.0","question":"Does the supplied evidence establish that the specific record was present in the named dataset?","rules":["AFFIRM only on a direct identifier match documented in the supplied evidence: an exact record, a hash, or an index entry.","DENY only if the evidence positively excludes the record, for example a documented complete enumeration that does not contain it.","CANNOT_CONCLUDE for statistical resemblance, partial-field matches, format matches, or any inference from similarity. Similarity is not membership.","Absence from the supplied evidence is not absence from the dataset unless the evidence is a documented complete enumeration.","Never treat a model's ability to produce a similar-looking record as evidence of membership.","Quote the span relied on. If the finding rests on absence, SPAN is NONE and the rationale must state what enumeration was or was not available."],"verdicts":["AFFIRM","DENY","CANNOT_CONCLUDE"]}
```

## How a finding under these rules is produced

Each adjudicator is a directory row driven through this system's own gateway. No code was deployed to add them and adding another model is one more row. Every finding records the model, the rule set hash, the quoted span, the exposure (`independent` when the adjudicator saw no other finding, `concurring` when it did), the ordering seed, and a signature. A mandatory recorded adversary argues the strongest honest case against the majority and is published whether it prevails or not.

Adjudicator rows: https://miscsubjects.com/api/directory/ADJUDICATE_KIMI · https://miscsubjects.com/api/directory/ADJUDICATE_GROK · https://miscsubjects.com/api/directory/ADJUDICATE_GLM · https://miscsubjects.com/api/directory/ADJUDICATE_LLAMA · https://miscsubjects.com/api/directory/ADJUDICATE_MINIMAX · adversary: https://miscsubjects.com/api/directory/ADJUDICATE_ADVERSARY · error-rate probe: https://miscsubjects.com/api/directory/ADJUDICATE_PROBE

## What a finding under these rules does and does not establish

It establishes that named adjudicators, under these exact rules at this exact hash, returned these findings on this claim against this source, with their exposure and ordering recorded — at a measured error rate when a probe report is attached.

It does not establish that the claim is true. No adjudication anywhere does that. A court, a journal and a clinical endpoint committee each declare rules, take findings from named parties under those rules, and preserve dissent. This is that structure, with the rule set pinned at a hash instead of scattered through case law.