Repository navigation
Call for independently authored LLM Errata adapters and validator #5
Description
Call for independent implementations
LLM Errata needs independently authored adapters and a separately produced
validator before it can claim interoperability.
Repository: https://github.com/thomaswillner/llm-errata
Current verdict: NOT_PROD_READY.
Requested work
Implement one adapter from the published schemas, vectors, and conformance
requirements without copying implementation code from prototype/.
Candidate substrates include:
- file-backed Markdown or knowledge-base memory;
- vector memory;
- graph memory;
- an agent framework's durable memory store; and
- an intentionally opaque or incomplete importer.
The implementation must preserve correction, supersession, and erasure as
different operations; quarantine affected state before repair; run negative,
positive, and preservation checks; and report partial, unknown, or failed
without converting missing evidence into success.
Independence and evidence
An implementation report must name its authors, repository and commit, supported
specification version, dependencies, test commands, unsupported behavior, and
licence. Shared conformance vectors are expected. Shared reference-adapter code
disqualifies the implementation as independent evidence.
The current personal-use licence requires written permission to implement the
specification. Open a GitHub issue naming the research or standards scope,
repository, organization, requested version, and intended publication. A
permission grant does not imply technical endorsement.
Activity
thomaswillner commentedon Aug 12, 2026
The provider-neutral semantic interface and deterministic conformance fixtures are now published for review in draft PR #8 at commit
35052868cde8b6f16fc6c20b4136ae0277f9a00e:#8
Requested evidence is unchanged: two independently authored adapters plus a separately produced validator, working from published schemas/vectors rather than copying
prototype/. Current personal-use licence still requires a written scoped permission grant before implementing the specification. Open here with author/repository/organization, research or standards purpose, requested version, and intended publication.This call does not claim interoperability or production readiness. Phase 2 remains incomplete; G2 remains
BLOCKED; overall verdict remainsNOT_PROD_READY.thomaswillner commentedon Aug 12, 2026
@joshuaswarren — inviting Remnic as a candidate for one independently authored adapter after LLM Errata's Phase 2 surface is complete.
Why this fit is specific: Remnic already publishes provenance, human-gated corrections, stale-belief suppression, a negative correction benchmark, propagation across agents, and receipts. LLM Errata asks whether an independent importer can consume an authenticated post-export correction, quarantine known descendants before rebuild, run negative/positive/preservation checks, and attest incomplete coverage rather than grading only its own store.
Call and current draft: #8
Boundaries:
prototype/.NOT_PROD_READY.If interested, reply with intended repository, implementation scope, and whether Remnic would be independently operated for a later synthetic-data experiment.
thomaswillner commentedon Aug 12, 2026
@alphaonedev — inviting ai-memory-mcp for a separate role: independently produced validator, or one independently operated system in the later experiment—not reviewer and maker of the same evidence.
Why this fit is specific: ai-memory-mcp already exposes SQLite/PostgreSQL backends, memory lineage, signed events/checkpoints, federation/replay controls, audit verification, lifecycle hooks, and explicit evidence gates. That makes it a strong candidate to test whether LLM Errata's receipts and unknown/partial coverage semantics survive a heterogeneous, federated substrate.
Call and draft surface: #8
Boundaries:
NOT_PROD_READY.If interested, please choose one role—validator or operated system—and name the repository/version, supported stores, opaque surfaces, and publication constraints.
thomaswillner commentedon Aug 12, 2026
Final Phase 2 target available for independent implementations
Internal Phase 2 is complete and frozen for external implementation work:
50e895fbfec544b16c94caa07bf2d1f4049a42e29547aec8328b601489dda067c6e62f287229b2b24a413dac2c9e7be98e429804The surface now includes durable quarantine checkpoints before CLI repair, provider-neutral semantic probes, owner-key rotation and rotated-key refusal, sequence-conflict and invalid-target cases, content-free erasure evidence, and mutation coverage for every signed receipt field.
Seeking:
prototype/; andCurrent personal-use licence still requires a written scoped permission grant before implementation. Reply with proposed repository, author or organization, selected role, requested version, intended research or standards purpose, supported stores, and conflicts. Permission is authorization, not technical evidence or endorsement.
Internal tests, CI, invitations, and repository-owner work do not satisfy G4. No independent adapter or validator has been accepted yet. Repository remains
NOT_PROD_READY.thomaswillner commentedon Aug 13, 2026
Corrected implementation target; case-by-case permission removed
The earlier target and permission language in this thread are obsolete. The stable implementation surface is:
08b95263c9ed700c43aea0b285696956cc23e87803abc492319b875a7d528e0e8de05714bc5a7219b42031fc7c3f42cff1f0bf14Independent commercial and non-commercial implementations of the specification now have an irrevocable, worldwide, royalty-free grant without case-by-case permission. Implementing products or services must credit LLM Errata, Thomas Willner, and the repository; reference code under
prototype/,scripts/, andtests/remains personal-use.@joshuaswarren — Remnic's provenance, correction, stale-belief suppression, propagation, receipts, and negative benchmark make it a strong candidate for one independently authored adapter. Would you kindly review the call and say whether a clean-room Remnic adapter is feasible? “Not feasible,” with the blocking reason, is valuable evidence.
@alphaonedev — ai-memory-mcp's lineage, signed checkpoints, replay controls, multiple stores, and evidence gates make it a strong candidate for the separately authored validator role. Would you kindly review that role and say whether independent receipt validation against the published schemas and vectors is feasible? If validator ownership is unsuitable, a referral is welcome.
No implementation is requested without its author's own decision. Please choose only one evidence role per author or organization and disclose conflicts. Invitations and repository-owner work do not satisfy G4. Current verdict remains
NOT_PROD_READY; G4 remainsBLOCKED.thomaswillner commentedon Aug 13, 2026
@pcarleton — I’m inviting a narrow conformance challenge because you are the leading contributor to the Model Context Protocol conformance framework, including revision-pinned requirements and explicit expected-failure handling.
Plain version: LLM Errata is a human-owned, substantially AI-assisted experimental protocol for correcting AI memory after one system exported it and another copied or derived from it. It remains
NOT_PROD_READY.The challenge: please try to break the independent-validator design. Can the checked-in schemas, vectors, mutation corpus, and semantic fixtures grade two clean-room adapters without importing hidden reference behavior, accepting vacuous passes, or allowing an expected-failure baseline to normalize a protocol violation? A counterexample, minimal validator architecture, or recommendation to narrow, redesign, or retire the G4 contract is more valuable than approval.
Implementation target: commit
ac4468faf73c2cc7949dd29b2a2a151f5bd23116, canonical surface digest7e0d6c88c1ca3a87743ac70ba2a3dfea0b350d112d2d3c59a3c6cbb537568f12, and independent-implementation contract.Please disclose relevant relationships or conflicts. This invitation and any interested-party feedback are not independent evidence; only separately authored artifacts and qualifying validation can affect G4. No endorsement, partnership, certification, or commercial commitment is implied. Any product or commercial implementation must provide accessible credit to LLM Errata and Thomas Willner under the repository terms.