Import a public correction issue
The public `/research/contribute` journey now exposes both options together: assemble a correction envelope entirely on-device, or open the structured public correction Issue Form when every source and statement is safe for public disclosure. The off-site form link is browser-proven without navigation and is tracked only as an aggregate form-open signal; an open never counts as a returned or accepted correction.
After exporting a GitHub `Research correction` Issue Form as JSON, a facilitator can create a non-overwriting, privacy-safe correction envelope without manually transcribing its fields:
pnpm research:correction-issue-import -- --input=<github-issue-export.json> --output=<private-correction-import.json>
The importer requires the public correction labels, exact target fields, attributable HTTPS evidence, AI disclosure, and all checked contribution declarations. It discards account, author, and assignee metadata and retains the bounded public body for deterministic replay. The result is only a proposal for independent review; it cannot accept or publish the correction or establish contributor qualification, novelty, authentication, title, legal conclusions, external impact, revenue, or income.
The resulting import artifact can be used directly as `correctionEnvelopePath` in `config/research-correction-propagation.json` together with its raw file SHA-256. `pnpm research:correction-propagation` replays the retained issue and rejects a modified wrapper before deriving a distinct, unpublished packet candidate; no envelope extraction or field copying is required.
Route a mixed issue batch
Facilitators can import up to 100 exported expert-review and correction issues together:
pnpm research:github-intake-batch -- --input=<github-issues.json> --output-directory=<new-private-routing-directory>
The offline router requires each issue to match exactly one supported label contract, rejects duplicate and ambiguous issues, applies the corresponding privacy-safe importer, and writes individual non-overwriting artifacts plus a replayable manifest. It retains no account values and performs no GitHub calls, outreach, qualification, correction acceptance, publication, or payment action.
Every routed item includes its deterministic artifact filename and an exact next command. Correction imports use `pnpm research:correction-import-propagate` with the finalized packet, imported correction, and both independently retained raw-file SHA-256 receipts. The command rejects receipt drift and overwrite attempts, fully replays the import, and writes one unpublished propagation artifact without requiring configuration edits.
Purpose
Meta Museum accepts reproducible source discoveries, corrections, identity
challenges, alternative hypotheses, workflow comparisons, and independent
reviews. The public starting point is `/research/contribute`; it routes a
visitor to a bounded task without requiring an account or collecting a lead.
Participation is not validation. A visit, download, AI critique, self-review,
or returned file does not establish expertise, novelty, impact, demand, or
acceptance. Only a qualifying evidence assembly evaluated against the current
release can affect A+ readiness.
Choose a pathway
- Build or challenge a packet. Use `/research/commons` to state a question,
identity fingerprint, claim, contradiction, alternative hypothesis, and
source plan. The result stays on the device and includes a reproducibility
digest.
- Reproduce or correct a frozen investigation. Use
`/research/rosenberg/validate` to inspect the exact dossier revision and
download a pseudonymous independent-review envelope.
- Compare research workflows. Qualified researchers can use
`/research/rosenberg/researcher-study` for a counterbalanced conventional
source and Meta Museum comparison. The tool records time and citations but
cannot certify the participant or score the result.
- Propose a bounded correction. Follow the
correction and attribution policy(correction-and-attribution-policy.md).
A material correction counts only after an accountable human incorporates
it into a distinct, versioned, limitations-bound packet.
The device-local builder on `/research/contribute` creates and downloads a
schema-valid, digest-bound correction envelope without an account, upload,
email address, or artwork file. The researcher retains custody and chooses
whether and how to share it.
- Prepare a no-account independent review. The indexed contribution page
assembles an integrity-signed review envelope entirely on-device. It binds a
reproduction result, three public sources, corrections, alternative
hypotheses, a provisional novelty assessment, limitations, and AI disclosure
to one packet digest. It asks for no name, email, affiliation, artwork upload,
identity evidence, or qualification file. Its public JSON Schema is
`/schemas/research-expert-review-envelope/v1`. The envelope is a review
proposal only; separate facilitator verification is mandatory before it can
count as qualified, independent, externally validated, or A+ evidence.
If the reviewer voluntarily returns the unchanged file through an approved
channel, an operator can replay-verify it without transcription using
`pnpm research:expert-review-envelope-import -- --input=<returned-envelope.json> --output=<new-private-candidate.json>`.
The candidate still requires a distinct facilitator decision matching the
exact envelope receipt and candidate artifact SHA-256; import alone
establishes no qualification or review, and a decision cannot be reused.
The validation and comparison tools are intentionally excluded from search
indexing because they are controlled study instruments. The contribution hub
is indexable and explains their purpose before a participant enters one.
Minimum evidence
source evidence.
limitation.
custody policy.
facilitator.
- Bind the response to the exact frozen revision or packet digest.
- Inspect and cite the source URLs independently; do not copy an AI summary as
- Separate observation, inference, disagreement, negative search, and
- Record meaningful corrections or reproduction failures explicitly.
- Preserve the original downloaded file and its SHA-256 under the approved
- Disclose AI assistance, conflicts, and independence to the accountable
Privacy and custody
Use the generated pseudonymous participant code in evidence files. Do not put
names, email addresses, affiliations, private collection data, credentials, or
sensitive artwork information in a downloaded packet. If qualification or
attribution is required, the facilitator keeps the identity mapping, consent,
qualification, conflict disclosure, and retention decision separately under
approved private custody.
The web tools do not upload responses. Returning a file through a separately
approved channel may expose the sender's identity to its operator; that channel
and its privacy terms are outside the evidence file and require explicit human
choice.
The structured GitHub issue is an optional public return channel, not the only
way to create or retain a review. An organically arriving reviewer can download
the envelope without any account and later choose an operator-approved private
channel. Meta Museum does not infer identity, qualification, independence, or
consent from a download or a pseudonymous code.
If a visitor has explicitly granted analytics consent, choosing one of the four
contribution pathways emits an aggregate route-level GA4 event. The event does
not contain a participant code, contact information, question, artwork data, or
response content. A pathway open measures navigation only and never counts as a
qualified review, correction, reproduction, scholarly result, or income.
Review and decision boundary
Agents may plan queries, retrieve public sources, extract candidate evidence,
check citations, flag identity collisions, detect contradictions, and assemble
draft reviewer packets. Agents may not establish novelty, authenticate an
object, settle identity/title/legal questions, certify participant expertise,
incorporate a scholarly correction, publish, contact anyone, or activate
commercial terms.
See the expert review guide(expert-review-guide.md),
independent validation program(independent-validation-program.md), and
Research Commons methodology(methodology.md) for the evaluator, custody, and
publication boundaries.