Reports, contributions and bonus calls
Preserve source evidence, validate new records and track earned usage.
Report an observed conflict
report_observation_issue / POST /v1/reports records a discrepancy and its original evidence without altering the corpus:
{
"observation_id": "RETURNED_OBSERVATION_ID",
"category": "stage_misattribution",
"summary": "The quoted source assigns this pressure to drying, not the reaction.",
"evidence_quote": "VERBATIM_SOURCE_WORDING",
"idempotency_key": "record-pressure-stage-v1"
}Categories are field_conflict, stage_misattribution, missing_source, structure_mismatch, other. Use an actual observation ID and a quote from the inspected source. The example placeholders must be replaced; they are not valid evidence. Unknown values or an empty search alone do not establish an error.
The service captures a trusted original snapshot and semantic digest. Reviewers can accept only supported condition overrides backed by the current source. Changed records require revalidation. Accepted corrections appear immediately in retrieval with original conditions and correction provenance; raw evidence is retained.
Contribute a source-backed reaction
submit_reaction_contribution / POST /v1/contributions accepts original reaction_smiles, source and idempotency_key. Source requires public HTTPS url, record_id, a quote of 20–4000 characters and a supported sharing-rights declaration: CC0-1.0, CC-BY-4.0, public-domain, or actual permission-granted. Include license_url when available; do not invent a license.
Optional conditions use Celsius, bar, agent SMILES and a source solvent label. Omit unknown conditions or preserve explicit nulls. The original submission is immutable. Automatic validation may attempt coefficients and SynRBL, but stores original and algorithmic candidates separately.
Reward milestones
| Milestone | Actual bonus calls |
|---|---|
| Eligible, novel, automatically valid submission | Up to 1 receipt call, subject to daily anti-abuse limits |
| Accepted after fresh chemistry checks plus source, license and quality review | 10 more calls, once |
| Duplicate, identity equation, unsupported imputation or invalid/unverified input | No automatic entitlement |
Receipt rewards are bounded per verified user and workspace. Changing a request key or scaling every stoichiometric coefficient does not make an old transformation new. Reward units are calls, not money. Read rewards from the actual response; never promise a reward based on a submitted request.
Review and durable storage
Use get_submission_status({kind,submission_id}) in MCP or workspace-scoped REST lists/details to check progress. States distinguish pending/needs_review, accepted, rejected and duplicate. Reports and status/guide reads are non-billable.
Only designated verified first-party staff can approve or retract a review. External connectors cannot approve themselves. Accepted community records join exact shared search and observation retrieval. A retraction removes the effective record/override while preserving history and reward audit.
Original snapshots, append-only events, correction versions, contribution data and the bonus ledger live separately from immutable corpus releases. Staff exports support reconciliation and training preparation; a database backup preserves complete history. An export is not permission to grant credits or bypass review.