| Internet-Draft | Scientific Data Record | November 2026 |
| Watts | Expires 20 May 2027 | [Page] |
This document specifies a JSON record format for scientific data and the claims derived from it. Each record binds a claim to an immutable source digest, a claim class, an evidence tier, a mandatory falsifier, a set of typed check results, and a provenance trail. Processing is fail-closed: a missing, unknown, or not-run check blocks admissibility. The format proves custody and unchanged bytes. It does not certify that a claim is true.¶
This note is to be removed before publishing as an RFC.¶
This is an individual, unreviewed design proposal. No experimental result, priority claim, or peer-review outcome is asserted. The example in Section 6 uses synthetic values and placeholder digests.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 20 May 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document.¶
Scientific results are increasingly produced and summarized by automated systems. A summary can be repeated across documents until it appears independently supported, even though only one underlying source exists. This document defines a record that keeps the claim, its source, its evidence tier, its falsifier, and the results of the checks run against it in a single machine-verifiable unit.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
A Record is a JSON object [RFC8259]. Table 1 lists its members. Unknown members MUST be ignored for processing but MUST be included when computing the record digest (Section 4).¶
| Member | Type | Requirement | Meaning |
|---|---|---|---|
| schema_version | string | REQUIRED | Format version. This document defines "1". |
| record_id | string | REQUIRED | Identifier unique within the issuing archive. |
| source | object | REQUIRED | Source identifier and digest. |
| claim | object | REQUIRED | Claim text and claim class. |
| evidence_tier | string | REQUIRED | One of the tiers in Section 3.3. |
| falsifier | string | REQUIRED | Non-empty falsifier statement. |
| checks | array | REQUIRED | Zero or more check results. May be empty. |
| provenance | array | REQUIRED | Entity, activity, and agent references. |
| state | string | REQUIRED | One of the states in Section 3.7. |
| created_at | string | REQUIRED | Timestamp in the format of [RFC3339]. |
| record_digest | object | REQUIRED | Digest of the record itself (Section 4). |
The "source" object has a REQUIRED string member "source_id" and a REQUIRED object member "digest" with string members "alg" and "value". This document defines the algorithm identifier "sha-256", with "value" being the lowercase hexadecimal encoding of the digest of the source bytes. A source object MAY carry an OPTIONAL "description" string.¶
The "claim" object has a REQUIRED string member "text" and a REQUIRED string member "class". The class MUST be one of:¶
A record's class describes what kind of statement the claim is. It does not describe how likely the claim is to be true.¶
The "evidence_tier" member MUST be one of "T1" (empirical), "T2" (derived from identified inputs), "T3" (rigorous conjecture), or "T4" (speculative). The scale is local to this format and is not a clinical or regulatory grading.¶
Each element of "checks" is an object with REQUIRED string members "name", "authority_class", and "result". "result" MUST be "PASS", "FAIL", or "NOT_RUN". A check object MAY carry "tool" and "run_id" strings. A check that was specified but not executed MUST be recorded as "NOT_RUN" and MUST NOT be omitted or recorded as "PASS".¶
Each element of "provenance" is an object with REQUIRED string members "role" and "ref". "role" MUST be "entity", "activity", or "agent", following the separation used in [PROV-O]. "ref" is an identifier or URI for the referenced item.¶
The "state" member MUST be "ADMISSIBLE", "BLOCKED", or "HISTORICAL". "HISTORICAL" marks a record retained for lineage that is no longer active. The processing rule for "ADMISSIBLE" is in Section 5.2.¶
The record digest is computed as follows.¶
Reordering JSON members MUST NOT change the digest. A verifier MUST recompute the digest and MUST reject the record on mismatch. A digest identifies bytes. It says nothing about whether the described claim is correct.¶
Processing is fail-closed: an unknown, missing, or not-run condition blocks a state transition and is never treated as a positive finding.¶
A processor MUST reject a record that:¶
A record MAY be in state "ADMISSIBLE" only if every check that the local policy requires has result "PASS" and at least one check with "authority_class" "admissibility" has result "PASS". A required check that is absent, "NOT_RUN", or "FAIL" MUST yield state "BLOCKED". Repetition of the same assertion across records that share a source does not count as independent support.¶
Archives SHOULD store records as an append-only sequence of events and derive the current state as a projection of that sequence. A correction is a new record or event that supersedes the earlier one, which is then marked "HISTORICAL". An archive MUST NOT silently edit a stored record.¶
The following record uses synthetic values. The digest values are placeholder strings, not computed digests, so the example is illustrative and would fail the digest verification of Section 4.¶
{
"schema_version": "1",
"record_id": "urn:example:record:0001",
"source": {
"source_id": "urn:example:source:values-v1",
"digest": {
"alg": "sha-256",
"value": "<placeholder: 64 hex characters>"
},
"description": "Three synthetic numbers: 1, 2, 3."
},
"claim": {
"text": "The arithmetic mean of the source values is 2.",
"class": "DERIVED"
},
"evidence_tier": "T2",
"falsifier": "A recomputed mean differs from 2.",
"checks": [
{
"name": "schema-validation",
"authority_class": "structural",
"result": "PASS"
},
{
"name": "recompute-mean",
"authority_class": "admissibility",
"result": "NOT_RUN"
}
],
"provenance": [
{"role": "entity", "ref": "urn:example:source:values-v1"},
{"role": "agent", "ref": "urn:example:agent:author"}
],
"state": "BLOCKED",
"created_at": "2026-11-16T00:00:00Z",
"record_digest": {
"alg": "sha-256",
"value": "<placeholder: 64 hex characters>"
}
}
¶
The record is in state "BLOCKED" because the required admissibility check was not run, even though the structural check passed.¶
The format provides custody and tamper evidence, not truth. A validly digested record can describe a false claim, and a reader who treats a well-formed record as authoritative is exposed to a false aura of correctness. Implementations SHOULD display the boundary between integrity and admissibility wherever records are rendered.¶
Canonicalization differences can cause digest mismatches or, if implemented incorrectly, collisions between distinct records. Implementations MUST use a conforming [RFC8785] serializer. Digest algorithm agility is handled by the "alg" member. Future versions can add algorithms, and verifiers MUST reject algorithms they do not support rather than skipping verification.¶
Records can be replayed, reordered, or selectively omitted. Archives SHOULD use an append-only event log with its own integrity protection. This document does not define signatures; deployments that need signer attribution can layer them over the canonical bytes.¶
A diagnostic result that is promoted to a state change, or a passing check of the wrong authority class that is accepted as satisfying another, would let an unverified claim become "ADMISSIBLE". Processors MUST enforce Section 3.5 and Section 5.2.¶
Records can name authors, participants, or third-party material in "provenance", "description", or claim text. Authors SHOULD omit personal data that is not needed to verify the claim and SHOULD confirm rights and consent before publishing a source or its digest. A digest of low-entropy personal data can be reversed by guessing and SHOULD NOT be published for such data.¶
This document has no IANA actions.¶
None.¶