PL-307 — Claims, Epistemic Status, Verification & Promotion Specification
Version: 0.1
Status: LOCKED
Project: Loom Research System
Effective: 2026-09-08
1. Purpose
PL-307 defines the state model governing Loom claims and establishes the distinction between:
-
workflow status;
-
epistemic status;
-
verification level; and
-
promotion stage.
These dimensions must not be collapsed into a single claim-status field.
PL-307 also defines:
-
the relationship between evidence, observations, claims, and diagnostics;
-
contradiction handling;
-
promotion eligibility;
-
demotion and correction;
-
validator requirements;
-
diagnostic constraints; and
-
provenance requirements.
Schema compliance is necessary but not sufficient for research credibility.
A valid object is not necessarily a true object.
A claim becomes credible only through the underlying evidence, verification, provenance, review, and correction processes defined elsewhere in the Loom system.
2. Core Epistemic Rule
AI may propose. Evidence must establish. Humans may promote.
AI-generated material is never authoritative merely because it was generated by a model.
The Loom system must preserve the distinction between:
proposal → evidence → assessment → verification → governance decision
No AI agent may independently promote its own output to the permanent Loom research record.
3. The Four Dimensions
Every claim MUST maintain four separate dimensions.
| Dimension | Question answered | Examples |
| -------------------- | ------------------------------------------------------------- | ------------------------------------------------------------- |
| status | Where is the object in the workflow? | proposed, under_review, completed |
| epistemic_status | What does the available evidence currently allow Loom to say? | unknown, contested, supported, known |
| verification_level | How deeply has the claim been checked? | none, source_backed, cross_verified, primary_verified |
| promotion_stage | Where does the claim sit in Loom's governed research record? | candidate, supported, canonical |
These fields answer different questions and must not substitute for one another.
4. Workflow Status
status describes operational progress only.
Allowed values:
proposed
The claim has been created but has not completed verification.
under_review
Verification, contradiction analysis, or human review is in progress.
completed
The applicable verification and contradiction procedures have been completed.
completed does not imply that the claim is supported.
rejected
The claim has been determined unsuitable for further advancement.
deprecated
The claim was previously used but is no longer considered valid or current.
A deprecated claim MUST reference the applicable correction, replacement, or supersession record.
5. Epistemic Status
epistemic_status describes the substantive evidentiary condition of the claim.
Allowed values:
proposed
A human or machine has suggested the claim.
No evidentiary conclusion has yet been reached.
unknown
Available research does not permit Loom to determine whether the claim is supported or contradicted.
Unknown is a legitimate research result.
unresolved
Relevant evidence exists, but the available evidence is inadequate, conflicting, incomplete, or otherwise insufficient to resolve the claim.
contested
Credible evidence supports the claim while credible evidence also challenges it.
The contradiction MUST be recorded.
supported
The available evidence supports the claim under the applicable Loom requirements.
Limitations may remain.
Supported does not mean absolute proof.
known
A deliberately restricted status indicating that high-quality evidence establishes the claim under Loom's current standards.
known MUST NOT be used as an ordinary synonym for supported.
Unless otherwise specified by a domain-specific methodology, known requires:
-
no unresolved contradiction;
-
primary verification where an applicable primary source exists;
-
human review;
-
explicit scope limitations.
If these requirements cannot be satisfied, the claim MUST NOT be assigned known.
6. Verification Level
Verification level describes the depth of checking performed.
For claims, allowed values are:
none
source_backed
cross_verified
primary_verified
none
No substantive verification has occurred.
source_backed
At least one retrieved source contains evidence that appears to support the claim.
A URL discovered through search is insufficient.
cross_verified
The claim has been examined against multiple relevant and independent sources.
Independence MUST be evaluated rather than inferred from source count.
primary_verified
The relevant primary documentation has been examined where an applicable primary source exists.
If no applicable primary source exists, the verification record MUST explicitly record:
primary_applicability: not_applicable
primary_verified MUST NOT be claimed where applicable primary documentation has not been examined.
For artifacts, additional verification levels may include:
editorially_verified
published
These describe artifact-level review and publication, not merely claim verification.
7. Source Retrieval Requirement
A source is considered retrieved only when Loom has obtained usable documentary content.
Examples include:
-
a local document;
-
an API response;
-
an archive snapshot;
-
OCR text derived from a retrieved document;
-
a verified documentary excerpt.
The following do NOT constitute retrieval:
-
a search result;
-
an unvisited URL;
-
an AI-generated citation;
-
a reference to a document that Loom cannot access;
-
a secondary description of an unavailable document.
A source that has only been discovered MUST NOT independently serve as evidence.
8. Source Independence
Source independence means independence from the subject of the claim and from other purportedly corroborating sources.
A source may be primary without being independent.
For example, a government agency reporting on its own program may constitute a primary source while also representing an interested institutional perspective.
Similarly, ten news articles reproducing the same wire report do not constitute ten independent sources.
Loom therefore evaluates:
-
provenance;
-
institutional relationship;
-
shared underlying source;
-
citation relationships;
-
syndication;
-
derivative reporting; and
-
substantive independence.
Source count MUST NOT be treated as equivalent to evidentiary independence.
9. Evidence Object
An EVD object represents documentary evidence extracted from a retrieved source.
Evidence MUST preserve sufficient provenance to permit a reviewer to identify its origin.
At minimum:
id: EVD-0001
type: evidence
source_id: SRC-0001
extracted_content: "..."
location: "page 14"
retrieved_at: 2026-09-08
document_hash: "sha256:..."
Evidence is not itself a claim.
The evidence record should represent what the source contains before Loom interprets what that information means.
10. Observation Object
Observation is an explicit research object.
It is a preliminary interpretation derived from evidence.
Observation is not a promotion stage.
The distinction is:
Evidence → Observation → Claim
An observation may state:
The available records indicate that warnings were present before the intervention.
The observation is an interpretation requiring further formulation and testing.
The claim may subsequently state:
State officials received documented warnings concerning water quality before the emergency declaration.
The claim then enters Loom's verification system.
This prevents Observation from being simultaneously:
-
an informal note;
-
a research object; and
-
a governance stage.
11. Promotion Stages
Promotion stages describe governance decisions regarding inclusion in the Loom research record.
The stages are:
Candidate → Supported → Canonical
candidate
The claim has sufficient basis to warrant continued research or inclusion as a provisional research object.
supported
The claim has satisfied the applicable evidence and verification requirements.
canonical
The claim has been explicitly admitted to the permanent Loom research record.
Promotion is a governance action.
Meeting verification conditions creates eligibility for promotion. It does not automatically promote an object.
The validator MUST therefore return:
ELIGIBLE_FOR_PROMOTION
rather than:
PROMOTED
when the necessary conditions are met but a governance decision has not occurred.
12. Promotion Eligibility
The basic relationship is:
Epistemic assessment
+
Verification requirements
+
Provenance completeness
+
Applicable review
↓
PROMOTION ELIGIBILITY
↓
Human governance decision
↓
Promotion
A claim MUST NOT become canonical solely because an automated validator determines that its evidence requirements have been met.
13. Promotion Matrix
| Epistemic status | Possible stage | Constraint |
| ---------------- | ------------------------------------------------- | -------------------------------------------- |
| proposed | candidate | Provisional only |
| unknown | candidate | Cannot advance as supported knowledge |
| unresolved | candidate | Must remain explicitly incomplete |
| contested | candidate, potentially supported | Contradiction must be documented |
| supported | candidate, supported, potentially canonical | Requires applicable governance |
| known | supported, potentially canonical | Restricted; requires additional requirements |
This matrix establishes eligibility, not automatic promotion.
14. Contradiction Handling
Contradiction MUST be represented structurally.
A claim MUST support:
contradictory_evidence:
- EVD-0011
contradiction_status: unresolved
contradiction_note: "..."
contradiction_resolution: null
Allowed values:
none
documented
unresolved
resolved
none
No credible contradictory evidence has been identified.
documented
Contradictory evidence has been identified and recorded.
unresolved
A substantive contradiction remains unresolved.
resolved
The contradiction has been evaluated and a documented resolution has been reached.
documented MUST NOT be used to conceal unresolved disagreement.
15. Contradiction Constraints
If contradictory evidence exists:
-
epistemic_status: knownis invalid when the contradiction remains unresolved. -
promotion_stage: canonicalis invalid when the contradiction remains unresolved. -
contradiction_statusMUST accurately reflect the state of the conflict. -
Any resolution MUST identify the reasoning and evidence supporting the resolution.
A claim may remain supported despite contradictory evidence only where the contradiction has been explicitly assessed and does not overturn the claim under the applicable methodology.
16. Claim Schema
Minimum claim representation:
id: CLM-0001
type: claim
statement: "..."
request_id: REQ-0001
status: under_review
epistemic_status: contested
verification_level: cross_verified
promotion_stage: candidate
related_evidence:
- EVD-0001
- EVD-0004
contradictory_evidence:
- EVD-0011
contradiction_status: unresolved
contradiction_note: "State timeline provides an alternative explanation."
contradiction_resolution: null
scope:
population: null
geography: "Flint, Michigan"
timeframe: null
exclusions: []
provenance:
proposed_by: "human"
proposed_at: "2026-09-08T15:32:00Z"
public: false
created: "2026-09-08"
updated: "2026-09-08"
17. Scope Integrity
Every substantive claim MUST have an identifiable scope.
Scope MAY include:
-
population;
-
geography;
-
timeframe;
-
institution;
-
event;
-
measurement;
-
exclusions;
-
applicable conditions.
A claim MUST NOT become broader during the transition:
Evidence → Observation → Claim → Diagnostic
without an explicit evidentiary basis for the expansion.
A diagnostic MUST NOT imply a broader causal or institutional conclusion than its supporting claims permit.
18. Provenance and Actors
Claims MUST preserve the identity or role of actors involved in their creation and review.
At minimum, provenance should distinguish:
proposed_by
extracted_by
verified_by
reviewed_by
promoted_by
Actors may be:
human
system
agent
model
Where applicable, machine provenance SHOULD include:
-
system or agent identifier;
-
model identifier;
-
model version;
-
prompt or instruction reference;
-
timestamp;
-
input object IDs.
The presence of machine provenance does not confer evidentiary authority.
19. Verification Events
Verification history MUST be append-only.
A verification event should record:
event_id: EVT-0001
claim_id: CLM-0001
actor: human
action: contradiction_review
timestamp: "2026-09-08T15:32:00Z"
inputs:
- EVD-0011
result: unresolved
reason: "Contradictory timeline remains materially relevant."
Existing verification events MUST NOT be silently rewritten when the claim changes.
Corrections and subsequent assessments MUST create new events.
20. Diagnostics
A DIAG object represents an analytical interpretation using Loom's ontology.
A diagnostic is not itself evidence.
A diagnostic may use claims but cannot upgrade them.
Hard constraint
A diagnostic's epistemic status cannot be stronger than the weakest material supporting claim.
Therefore, if a diagnostic depends upon a claim that is:
unknown
unresolved
contested
the diagnostic MUST NOT be represented as unqualified supported or known.
The diagnostic MUST preserve the uncertainty of its underlying claims.
Diagnostics MUST NOT:
-
invent ontology nodes silently;
-
suppress contradictory evidence;
-
diagnose from a single weak source where stronger evidence is required;
-
manufacture causal certainty;
-
promote their own conclusions;
-
exceed the scope of their supporting claims.
If no existing Loom concept adequately describes the evidence, the diagnostic SHOULD record that limitation rather than silently creating a new category.
21. Invalid-State Validation
The Loom validator MUST identify epistemically inconsistent objects.
Examples include:
Invalid: unresolved contradiction + known
epistemic_status: known
contradiction_status: unresolved
Result:
INVALID_STATE
RULE: EPISTEMIC-004
SEVERITY: ERROR
REMEDIATION: Resolve contradiction or downgrade epistemic status.
Invalid: source-backed without retrieved evidence
verification_level: source_backed
when no retrieved source or mapped evidence exists.
Invalid: primary verification without applicable primary evidence
verification_level: primary_verified
without qualifying primary-source evidence.
Invalid: canonical unresolved claim
promotion_stage: canonical
epistemic_status: unresolved
Invalid: diagnostic exceeds supporting claim
DIAG status > weakest supporting CLM status
The validator MUST reject the object or mark it ineligible.
22. Validator Rule Structure
Every validation error MUST include:
-
rule code;
-
severity;
-
affected object;
-
reason;
-
required remediation;
-
whether override is permitted.
Example:
rule_code: EPISTEMIC-004
severity: error
object_id: CLM-0003
reason: "Claim declares known while contradictory evidence remains unresolved."
remediation: "Resolve contradiction or downgrade epistemic status."
override_allowed: true
23. Override and Correction
Validation rules may encounter legitimate exceptional circumstances.
An override MUST NOT silently bypass a rule.
Where an override is permitted, it MUST record:
override:
rule_code: EPISTEMIC-004
authorized_by: "..."
reason: "..."
timestamp: "..."
review_reference: "REV-..."
A correction object MUST be created where the override changes the public or permanent research record.
Overrides are governance events, not validator failures to be hidden.
24. Demotion
Promotion is reversible.
A claim may move:
Candidate → Supported → Canonical
and later:
Canonical → Contested
Canonical → Revised
Canonical → Deprecated
Canonical → Rejected
A demotion MUST preserve:
-
previous state;
-
new state;
-
triggering evidence;
-
reason;
-
reviewer;
-
timestamp;
-
correction reference.
Canonical means admitted to the permanent research record.
It does not mean permanently immune from revision.
25. Claim State Examples
Supported
status: completed
epistemic_status: supported
verification_level: primary_verified
promotion_stage: supported
contradictory_evidence: []
contradiction_status: none
Interpretation:
The verification process is complete, primary documentation was applicable and examined, and the evidence currently supports the claim.
Contested
status: completed
epistemic_status: contested
verification_level: cross_verified
promotion_stage: candidate
contradictory_evidence:
- EVD-0011
contradiction_status: unresolved
Interpretation:
The claim has meaningful supporting evidence but also meaningful contradictory evidence. It remains provisional.
Unknown
status: completed
epistemic_status: unknown
verification_level: cross_verified
promotion_stage: candidate
Interpretation:
Loom searched the available evidence but cannot responsibly determine the answer.
Unknown is an acceptable result.
26. Schema Compliance Is Not Research Credibility
PL-307 explicitly rejects the proposition that a formally valid object is necessarily a credible research finding.
Credibility additionally requires:
-
real retrieved sources;
-
authentic evidence extraction;
-
appropriate source evaluation;
-
meaningful contradiction search;
-
accurate provenance;
-
human review where required;
-
transparent limitations;
-
correction mechanisms; and
-
adversarial testing.
The schema prevents certain forms of epistemic failure.
It does not prove that the underlying research is correct.
27. Machine Enforcement Requirement
The rules in this specification MUST eventually be represented in machine-readable schemas and validation tests.
Markdown documentation alone does not constitute enforcement.
Minimum implementation requirements:
schemas/
claim.schema.yaml
evidence.schema.yaml
observation.schema.yaml
diagnostic.schema.yaml
verification.schema.yaml
provenance-event.schema.yaml
and:
tests/
fixtures/
adversarial/
results/
At minimum, the test suite MUST demonstrate that the validator rejects:
-
known+ unresolved contradiction; -
canonical+ unresolved epistemic status; -
source_backedwithout retrieved evidence; -
primary_verifiedwithout applicable primary evidence; -
diagnostic status stronger than supporting claims.
28. Adversarial Testing Is a Release Requirement
Adversarial testing is not a future final phase.
Every implementation of PL-307 MUST be tested against known failure conditions.
Tests MUST ask:
-
Can an unretrieved source become evidence?
-
Can derivative sources be falsely counted as independent?
-
Can a contradictory source disappear from the final representation?
-
Can an unsupported claim become canonical?
-
Can a diagnostic exceed its evidence?
-
Can the system manufacture certainty from insufficient evidence?
-
Can a correction be made without destroying historical provenance?
A release that fails a critical epistemic test MUST NOT be labeled as satisfying PL-307.
29. Relationship to Other Loom Objects
The intended architecture is:
REQUEST
↓
SOURCE
↓
EVIDENCE
↓
OBSERVATION
↓
CLAIM
↓
VERIFICATION
↓
PROMOTION
↓
ARTIFACT
Diagnostics operate on the verified research layer:
CLAIMS + EVIDENCE
↓
DIAGNOSTIC
The promotion ladder is independent of the evidentiary pipeline:
Candidate → Supported → Canonical
This separation prevents the research object lifecycle from being confused with the evidentiary status of a claim.
30. Normative Summary
Loom MUST preserve the following distinctions:
Workflow is not truth.
Verification depth is not epistemic status.
Epistemic status is not promotion.
Primary source is not automatically independent.
Source count is not corroboration.
Contradiction documentation is not contradiction resolution.
Schema validity is not research credibility.
AI agreement is not evidence.
Canonical is not infallible.
And above all:
AI may propose. Evidence must establish. Humans may promote.
Status
PL-307 v0.1 is locked as the governing claim-state specification for the Loom Research System.
Future revisions MUST be versioned.
Changes to:
-
state definitions;
-
promotion rules;
-
validation rules;
-
contradiction handling;
-
provenance requirements; or
-
diagnostic constraints
constitute a specification change and MUST NOT be silently introduced through implementation.