| Internet-Draft | Scientific Data Admissibility Records | November 2026 |
| Watts | Expires 20 May 2027 | [Page] |
Automated research pipelines routinely move scientific data and derived claims from ingestion to publication without a machine-checkable record of what has actually been verified. This document defines a Scientific Data Admissibility Record (SDAR): a small, serializable structure that keeps schema conformance, provenance, numerical verification, empirical evidence, and interpretation as separate layers, each with an explicit status. It specifies a fail-closed aggregation rule so that an unverified or not-yet-run layer can never be silently promoted into an accepted result. The document also defines an XML serialization and processing rules, and is intended as a starting point for discussion rather than a proposed standard.¶
This note is to be removed before publishing as an RFC.¶
This is an individual, personal submission. It has no IETF working group and represents the view of its author only.¶
Discussion of this document is welcome via the author's contact address given in the Authors' Addresses section.¶
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. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Scientific datasets, statistical artifacts, and AI-assisted analyses are now commonly produced by automated pipelines. These pipelines usually report a single outcome, such as "passed", for a result that in fact depends on several distinct kinds of checking: whether the data conforms to its declared schema, where it came from, whether the reported numbers can be recomputed, whether independent evidence supports them, and whether the stated interpretation follows from them.¶
When these kinds of checking are collapsed into one flag, a pipeline can quietly promote a claim: a result that was only schema-valid becomes "verified", or a number that was never recomputed is reported as confirmed. A reader of the final artifact cannot tell which checks were performed, in which environment, or whether a check was run at all.¶
This document defines the Scientific Data Admissibility Record (SDAR) to make that distinction explicit and machine-readable. The design goals are:¶
This document does not define how any check is performed, does not assert that any particular scientific claim is true, and does not define a trust framework for the parties that produce records.¶
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.¶
An SDAR contains one result for each of the following layers.¶
| Layer | Question answered |
|---|---|
| schema | Does the subject conform to its declared schema? |
| provenance | Can the subject's origin and transformations be traced, with inputs bound by digest? |
| numeric | Can the reported numerical values be recomputed from the bound inputs? |
| empirical | Is there independent evidence supporting the values, such as an independent reproduction? |
| interpretation | Does the stated interpretation follow from the verified values, as judged by an identified reviewer? |
A layer result MUST NOT be derived from the result of a different layer. In particular, a "pass" for schema or numeric MUST NOT cause the empirical or interpretation layers to be recorded as "pass".¶
Each layer has exactly one status:¶
A producer MUST NOT record "pass" for a layer unless it can supply the evidence fields required in Section 3.4. The statuses "not-run" and "pending" are first-class values and MUST NOT be presented to consumers as equivalent to "pass".¶
The aggregate state of a record is computed from its required layers as follows:¶
A processor that cannot parse a record, cannot resolve a required digest, or encounters an unknown status value MUST treat the aggregate state as "incomplete" and MUST NOT treat it as "admissible". Which layers are required is a matter of local policy; however, a policy that omits the interpretation layer MUST NOT label the aggregate state in a way that implies interpretive review occurred.¶
A layer with status "pass" MUST carry:¶
A layer with status "pass" in the empirical layer SHOULD identify the party that performed the independent reproduction and SHOULD state whether that party is independent of the producer of the subject.¶
Digests identify the exact bytes a result was computed over. A digest MUST be expressed as an algorithm name and a value; whitespace inside a serialized digest value is not significant and MUST be ignored by processors. Producers SHOULD use a SHA-2 family algorithm [RFC6234]. Where the subject is structured data, producers SHOULD digest a canonical serialization so that equivalent content yields the same digest; for JSON, [RFC8785] is one such canonicalization.¶
If any digest in a record does not match the corresponding artifact at verification time, every layer bound to that digest MUST be treated as "not-run" by the verifier.¶
This section defines an XML serialization of an SDAR. The namespace shown is a placeholder and is not registered; see Section 6.¶
The root element is "sdar". It contains one "subject" element, one "layers" element containing exactly five "layer" elements (one per layer name), and one OPTIONAL "aggregate" element. A "layer" element has a "name" attribute and a "status" attribute, and MAY contain "environment", "produced", "input", and "reviewer" child elements. An "aggregate" element, when present, is informational; consumers MUST recompute the aggregate state from the layers using Section 3.3 and MUST NOT trust a supplied value.¶
The following record describes a dataset whose schema and provenance checks have passed, whose numeric check has failed, and whose remaining checks have not been run. All identifiers, digests, and the namespace are illustrative.¶
<?xml version="1.0" encoding="utf-8"?>
<sdar xmlns="https://example.org/ns/sdar" version="0">
<subject id="dataset-0001" kind="dataset">
<digest alg="sha-256">
00000000000000000000000000000000
00000000000000000000000000000000
</digest>
<title>Example measurement series</title>
</subject>
<layers>
<layer name="schema" status="pass">
<environment>validator rev a1b2c3d</environment>
<produced>2026-11-16T12:00:00Z</produced>
<input alg="sha-256" ref="dataset-0001">
00000000000000000000000000000000
00000000000000000000000000000000
</input>
</layer>
<layer name="provenance" status="pass">
<environment>lineage tool rev 4d5e6f7</environment>
<produced>2026-11-16T12:05:00Z</produced>
<input alg="sha-256" ref="dataset-0001">
00000000000000000000000000000000
00000000000000000000000000000000
</input>
</layer>
<layer name="numeric" status="fail"/>
<layer name="empirical" status="not-run"/>
<layer name="interpretation" status="not-run"/>
</layers>
<aggregate state="rejected"/>
</sdar>
¶
In this example, a processor following Section 3.3 computes the aggregate state "rejected" because the numeric layer has status "fail". Had the numeric layer been "not-run", the aggregate state would have been "incomplete".¶
This document has no IANA actions. The namespace used in the examples is a placeholder, and no registration is requested.¶
An SDAR is only as trustworthy as the party that produced it. This document does not define signing, authentication of producers, or a trust framework; a record that is unsigned can be forged or altered. Deployments that rely on records for decisions SHOULD protect them with an integrity mechanism appropriate to their environment and SHOULD record which party signed each layer result.¶
Digest binding detects substitution of artifacts only if the digest was obtained over an authenticated channel. A producer who controls both a record and the artifact it names can make them consistent while both are wrong.¶
The "admissible" aggregate state describes completeness of checking only. Consumers MUST NOT read it as a statement that the subject is correct or that the interpretation is true. Presenting "admissible" in a user interface without the underlying layer statuses invites exactly the silent promotion this document aims to prevent.¶
Records can contain environment identifiers, reviewer identifiers, and timestamps that may be sensitive. Implementations SHOULD consider whether these fields should be disclosed beyond the parties that need them.¶
General guidance on writing security considerations is given in [RFC3552].¶
The layered separation described here follows the author's ongoing work on evidence-to-consequence governance for research and AI-assisted claims.¶