audit: adversarially review the paralegal research corpus

[zackees]

Context

The repository now contains a substantial, license-agnostic paralegal research corpus covering reciprocal open-source license enforcement, remedies, cure, reinstatement, commercial-license alternatives, procedural posture, and related doctrine. Before relying on it for paralegal retrieval or attorney review, the corpus needs an adversarial audit focused on contradictions, source defects, loopholes, overstatement, retrieval failures, and governance gaps.

This issue is the parent tracker for that audit. The repository research remains the canonical record; these issues track corrections and validation work. This is background legal research, not legal advice or a final legal determination.

Proposal

Run independent adversarial review lanes and convert confirmed, non-duplicate findings into focused sub-issues:

1. License text and doctrine: verify license provisions, remedies, cure, reinstatement, forum, patent, and enforcement characterizations against primary text and authority.
2. Case/source integrity: verify order-level procedural history, disposition, requested versus awarded relief, source identity, and precedential limits.
3. Retrieval and governance: test whether agents can find, distinguish, update, and cite the research without silently overstating it.

Each child issue must identify the affected files or metadata, state the exact concern, distinguish a verified defect from an open legal question, propose a bounded correction, and preserve attorney-review boundaries.

Filed sub-issues

License and case-source integrity

• research: correct verified license-text errors in comparison corpus #20 - correct verified license-text errors in the comparison corpus
• research: correct source and posture defects in key U.S. license cases #21 - correct source and posture defects in key U.S. license cases
• provenance: pin the FastLED rc1 authority to immutable content #25 - pin the FastLED rc1 authority to immutable content

Retrieval and governance

• governance: make stale-source handling operational #22 - make stale-source handling operational
• retrieval: align matter schema with doctrine, remedy, and time taxonomy #23 - align the matter schema with doctrine, remedy, and time taxonomy
• retrieval: make authority proposition treatment machine-checkable #24 - make authority proposition treatment machine-checkable
• governance: repair matter-link and integration-state contradictions #26 - repair matter-link and integration-state contradictions
• metadata: separate non-case legal effect from precedential weight #27 - separate non-case legal effect from precedential weight

Audit completion

• Three independent Terra audit lanes completed: doctrine, source integrity, and retrieval/governance.
• Candidate findings were checked against repository content and, where needed, current primary sources.
• Duplicate and overlap checks were run against existing repository issues.
• Confirmed actionable findings were consolidated into focused issues and attached as actual GitHub sub-issues.
• Speculative, already-qualified, or out-of-scope findings were excluded and recorded below.

Excluded or converged findings

• The absence of a final U.S. ruling squarely applying AGPL section 13 is already presented as an unresolved and bounded negative-search result, not as settled law.
• The corpus generally distinguishes requested relief, interim relief, final judgments, defaults, and settlements. The confirmed order-level exceptions are isolated in research: correct source and posture defects in key U.S. license cases #21.
• Linking and derivative-work questions are generally marked low-confidence and fact-dependent rather than stated categorically.
• Known foreign-source and translation limitations are already disclosed and do not justify a duplicate issue without a specific mistranslation or omitted authority.
• answered and legal_review intentionally represent separate retrieval and attorney-review states.
• Negative research results are bounded by sources and dates rather than framed as proof that no authority exists.
• The CAL cure discussion is thin, but the audit did not establish a specific contradiction sufficient for a separate issue.
• Older FastLED-specific drafting and product-license concerns remain tracked in META: AI first-pass legal review of FastLED Reciprocal License 1.0 RC — triage, findings & prior-art survey #8 and are outside this corpus-integrity audit.

Cross-lane conflicts requiring resolution

• The retrieval lane found no broken child/supersession relationship, while the coordinator's deterministic scan found 31 asymmetric related_matter_ids edges. Issue governance: repair matter-link and integration-state contradictions #26 must decide whether reciprocal backlinks are mandatory or whether the schema should explicitly support directed relations.
• Existing controlled-value validation accepted the non-case authority weights because each value was globally allowed. The doctrine audit showed that the missing invariant is type-specific: contracts, statutes, and license texts should not be described using case-law precedential weight. Issue metadata: separate non-case legal effect from precedential weight #27 addresses that distinction.

Acceptance criteria

• At least three independent adversarial audit lanes cover doctrine, source integrity, and retrieval/governance.
• Candidate contradictions and loopholes are checked against primary sources or clearly labeled as unresolved research questions.
• Each confirmed, actionable finding is filed as a focused child issue.
• Duplicate, speculative, and already-disclosed findings are consolidated or excluded with a recorded reason.
• Findings distinguish license-text errors, factual/source errors, procedural-posture errors, legal uncertainty, and workflow/schema defects.
• The meta issue contains a final checklist of all filed sub-issues and remains open until those sub-issues are resolved or expressly deferred.
• No issue treats allegations, requested relief, preliminary orders, settlements, defaults, or nonprecedential materials as a merits holding without qualification.
• No audit action changes the repository LICENSE or decides which license FastLED should adopt.

Decisions

• Treat the work as general research about related reciprocal licenses and enforcement, not as a decision to migrate FastLED to the AGPL or any other license.
• Keep research memoranda and authority cards in the repository as the canonical evidence base; use GitHub issues for remediation tracking.
• Prefer one issue per coherent correction cluster so related metadata, prose, and validation changes can be reviewed together.
• Leave all child issues and this meta issue open for implementation and attorney review.
• Preserve the distinction between paralegal research and a lawyer's final legal advice.

Open questions

• Should asymmetric related-matter links be repaired as data defects, or should the schema define directed and reciprocal relationship types separately?
• Which non-case authority taxonomy should replace binding, persuasive, and nonprecedential for licenses, statutes, contracts, and official guidance?
• What freshness interval and escalation rule should apply to volatile authorities and unresolved negative searches?
• Should proposition treatment become required structured metadata for every authority card or only for authorities used in answer-ready matters?

Related issues

• docs: build a license-agnostic copyleft enforcement research corpus #17 - created and populated the current license-agnostic paralegal research corpus.
• Research: judicial remedies for AGPL-style source noncompliance (no license decision) #15 - contains the focused AGPL noncompliance remedies research that motivated part of this audit.
• META: AI first-pass legal review of FastLED Reciprocal License 1.0 RC — triage, findings & prior-art survey #8 - tracks older FastLED-specific drafting and product-license concerns outside this corpus-integrity review.