Add primary-source analysis conformance profile

This commit is contained in:
Matthew Raymer
2026-10-10 03:10:57 +00:00
parent 4594791b0b
commit c90de83c0b
2 changed files with 268 additions and 1 deletions
+8 -1
View File
@@ -18,8 +18,15 @@ governed standards change under the then-current standards.
Profiles do not authorize their own adoption or promotion. Repository governance and the
then-current standards govern acceptance, revision, deprecation, and replacement.
The first adopted profile is:
The adopted profile is:
- `devidence-acquisition-representation-v0.2.md` — acquisition, representation identity,
integrity comparison, compound/container materialization, bounded representation-adequacy
assertions, reproducible provenance, and associated failure semantics.
The candidate profile is:
- `devidence-primary-source-analysis-interpretive-derivation-v0.1.md` — direct primary-source
confirmation, raw/normalized distinction, bounded corpus-negative claims, reference
resolution, interpretive derivation, operation-taxonomy versioning, authority bindings,
disagreement preservation, and supersession.
@@ -0,0 +1,260 @@
# DEVIDENCE Primary-Source Analysis and Interpretive Derivation Conformance Profile
## Version 0.1
Status: CANDIDATE EXTERNAL CONFORMANCE PROFILE
Authority scope: implementations explicitly claiming this profile and version after adoption
Base Layer authority: NONE
This document is a candidate external conformance profile. It does not amend DEVIDENCE Base
Layer meanings and does not become part of Base Layer conformance merely by being present in
the standards repository.
Normative keywords such as **SHALL**, **SHALL NOT**, **SHOULD**, and **MAY** describe the
requirements proposed for an implementation that explicitly claims conformance with this
profile and version after adoption. Until adoption, this document is non-normative.
## Purpose
This profile defines testable behavior for research systems that verify source-grounded claims,
perform bounded corpus searches, resolve references, and derive interpretive observations from
preserved primary-source material.
The profile uses existing DEVIDENCE Base Layer concepts:
- resource;
- representation;
- acquisition;
- observation;
- assertion or claim;
- evidence;
- derivation;
- provenance;
- coverage;
- change and supersession.
It does not introduce a universal truth authority, domain canon, confidence scale, or
interpretive taxonomy.
## Primary-source confirmation
When a material claim is directly testable against an accessible primary representation, a
conforming implementation SHALL either:
1. acquire and examine an identified primary representation; or
2. record why primary-source confirmation was not available through the attempted path.
A secondary source MAY identify a useful claim, locator, citation, or search target, but the
secondary report SHALL NOT be represented as direct confirmation of a primary text when the
primary representation was not examined.
The confirmation record SHALL preserve, to the provenance ceiling supported by the source:
- resource identity;
- representation identity;
- edition, witness, corpus, revision, or other bounded source identity when applicable;
- acquisition route and tool/version;
- observed or retrieved time as applicable;
- exact locator or source span when available;
- raw or preserved representation identity;
- cryptographic digest when available or required;
- normalization and transformation history used for analysis;
- the observation actually made;
- the claim ceiling describing what the observation does not establish.
## Raw, normalized, and interpreted forms
Exact preserved material, normalized analytical text, translations, transcriptions, extracted
features, and interpretive conclusions SHALL remain distinguishable.
A conforming implementation SHALL NOT silently replace an exact source representation with a
normalized, translated, corrected, reconstructed, or interpreted form.
When normalization or transformation is used for searching or analysis, derivation provenance
SHALL identify:
- input representation;
- transformation or normalization method;
- software or model identity and version where applicable;
- configuration material necessary to reproduce the transformation;
- resulting representation or analytical object;
- relevant digests or stable identities.
## Bounded corpus-negative claims
A statement that a term, pattern, feature, or record is absent from a corpus is a bounded
analytical observation, not a universal absence claim.
For a corpus-negative result, a conforming implementation SHALL preserve:
- exact corpus identity;
- exact revision, release, commit, tree, snapshot, or equivalent bounded version;
- declared scan scope;
- inclusion and exclusion rules;
- file or member inventory sufficient to audit the scan;
- integrity identities for the scanned material when practical;
- search pattern or query;
- character normalization and case-handling rules;
- search tool and version or reproducible search procedure;
- search result;
- whether zero matches were expected or merely observed;
- coverage or completeness limits known to the implementation.
A zero-hit result SHALL NOT be promoted to a claim about manuscripts, editions, witnesses,
repositories, or corpora outside the bounded scan.
Acquisition failure, unreadable material, excluded files, unsupported encodings, and
materialization failure SHALL NOT be converted into source absence.
## Reference resolution
A citation, footnote, bibliography entry, embedded reference, or secondary-source claim that
points to another source MAY generate a separate acquisition or verification target.
A conforming implementation SHALL preserve the distinction between:
- the fact that source A cites or refers to source B;
- the representation of source B actually acquired;
- the observation made from source B;
- an analyst's inference relating A and B.
Truth, interpretation, or evidentiary support SHALL NOT be inherited transitively merely
because a reference exists.
When a referenced source is unavailable, the unresolved reference SHALL remain visible rather
than being silently treated as confirmed.
## Interpretive derivation
When an implementation records how a source applies, explains, extends, harmonizes, narrows,
or otherwise interprets an earlier rule, text, or source, the result SHALL be represented as a
derivation or analytical observation rather than as part of the underlying source text.
An interpretive-derivation record SHALL preserve, when applicable:
- identity of the governing or antecedent source material being interpreted;
- exact source span or locator for that basis;
- identity of the interpreting source;
- exact source span or locator for the interpretive act or resulting ruling;
- the case, problem, question, or circumstance being addressed;
- the resulting ruling, application, explanation, or conclusion as observed in the source;
- any rationale explicitly stated by the interpreting source;
- the analytical classification of the interpretive operation;
- the deriver or analyst identity;
- derivation method and version;
- confidence or uncertainty where legitimately supplied;
- alternative plausible classifications or unresolved ambiguity;
- a claim ceiling separating observation from reconstruction.
The record SHALL distinguish at least these three layers:
1. **source observation** — what the relevant representation actually says or contains;
2. **interpretive operation observation** — what movement from antecedent material to applied
ruling can be directly supported by the interpreting source;
3. **analytical reconstruction** — the researcher's classification or explanation of that
interpretive movement.
A conforming implementation SHALL NOT represent an analytical reconstruction as though the
ancient or primary source explicitly stated that classification.
## Operation taxonomies
A domain profile or local research contract MAY classify interpretive operations with labels
such as analogy, reciprocal application, category extension, procedural concretization,
harmonization, exception, priority resolution, grammatical inference, or protective
implementation.
Such labels are analytical vocabulary, not Base Layer meanings.
The implementation SHALL:
- define the vocabulary it uses;
- version that vocabulary;
- preserve the evidence supporting each assigned label;
- permit an operation to remain UNKNOWN or multiply classified;
- preserve later reclassification without rewriting the original record.
## Governing-source or authority bindings
A domain-specific research contract MAY declare that one source collection, rule set, canon,
standard, or authority governs the interpretation of other writings.
DEVIDENCE itself does not choose that authority.
When such a binding exists, a conforming implementation SHOULD preserve:
- identity of the domain contract establishing the authority relationship;
- identity of the governing source or rule set;
- the scope in which the authority relationship applies;
- version or revision of the domain contract;
- the evidentiary or methodological role assigned to subordinate or interpretive sources.
The existence of a governing-source binding SHALL NOT cause derived interpretation records to
be merged into the governing source representation.
## Disagreement and alternative derivations
Different analysts, models, communities, editions, or source witnesses may support different
interpretive derivations.
A conforming implementation SHALL preserve materially distinct derivations separately when
their provenance, source basis, operation classification, or conclusion differs.
Disagreement SHALL NOT be flattened into a synthetic consensus record unless that synthesis is
itself represented as a separately attributable derivation.
## Supersession and correction
Later source acquisition, improved transcription, corrected normalization, new witnesses, or
revised interpretation MAY supersede an earlier analytical result.
Supersession SHALL preserve:
- the earlier observation or derivation;
- the later observation or derivation;
- the relationship between them;
- the reason for change;
- the evidence responsible for the change where available.
Correction SHALL NOT erase the historical analytical record.
## Research-state mutation boundary
Acquiring, preserving, indexing, searching, validating, or interpreting a source SHALL NOT by
itself mutate unrelated claims, conclusions, coverage records, queues, or publication state.
If a research system uses a validated observation to update another research object, that
update SHOULD occur through a separately attributable and auditable derivation or workflow
step.
## Minimum conformance vectors
An implementation claiming this profile SHOULD be tested against at least:
- direct primary-source confirmation of a positive textual claim;
- direct primary-source confirmation that contradicts a secondary report;
- primary source unavailable through the attempted route;
- exact-locator extraction from a preserved representation;
- raw-to-normalized derivation with reproducible provenance;
- bounded corpus search with positive matches;
- bounded corpus search with a valid zero-hit result;
- corpus search with unreadable or excluded material producing bounded or UNKNOWN coverage;
- secondary citation that resolves to a separately acquired primary source;
- unresolved reference that remains unresolved;
- interpretive derivation with an explicitly stated source rationale;
- interpretive derivation whose operation classification is analytical rather than explicit;
- two plausible operation classifications preserved without forced collapse;
- disagreement between independent derivations;
- later correction or supersession preserving the prior record;
- domain authority binding that remains separate from DEVIDENCE Base Layer semantics.
## Authority boundary
This candidate profile does not amend DEVIDENCE Base Layer meanings, declare any source
canonical or authoritative, establish truth from preservation, or make an interpretive
classification binding merely because it was recorded.
If adopted, its authority applies only to implementations explicitly claiming this profile and
version. Promotion of any distinction from this profile into the Base Layer requires a
separate governed standards change.