Molecular Evidence Pack Builder

Build one bounded, cross-source molecular evidence dossier for an explicitly declared protein or enzyme family, with source-level provenance and a tamper-detecting portable ZIP.

What this does not do

• No de novo drug discovery.
• No medicinal chemistry recommendation.
• No clinical inference.
• No unrestricted all-database query.
• No nested calls to paid Actors.
• No inclusion of incomparable bioactivity as if pooled.
• No public listing before one-family scientific validation and operator approval.

Sources and authorization

Version 0.1 accepts only public identifiers and makes typed, bounded GET requests to official UniProt, PubChem, RCSB PDB, AlphaFold DB, and ChEMBL endpoints listed in SOURCE_POLICY.md. It does not accept arbitrary URLs, credentials, private datasets, or source tokens. It does not fall back to scraping, search engines, paid child Actors, or a broader source when an API fails.

Each fact carries an official request URL with query values redacted, a JSON evidence path, the raw-response SHA-256, and a versioned source contract. Source values are preserved alongside normalized values. New provider fields are listed as unmapped instead of being silently dropped.

Input

Declare one family with 1–25 exact UniProt accessions and explicit inclusion/exclusion rules. Supply 0–500 exact PubChem CIDs, ChEMBL molecule IDs, InChIKeys, or SMILES. The Actor performs identifier resolution only; it does not discover related compounds.

{
  "family": {
    "name": "Flintglade demo family",
    "targetIdentifiers": [{"uniprotId": "P69905"}],
    "inclusionRules": ["exact target identifier"],
    "exclusionRules": ["ambiguous target mapping"]
  },
  "compounds": [{"type": "pubchemCid", "value": "2519"}],
  "includeStructures": true,
  "includeBioactivity": false,
  "includeCoordinates": false,
  "maxActivities": 100
}

For bioactivity, every target needs a declared and source-consistent chemblTargetId, and every compound needs an exact ChEMBL molecule ID. bioactivityFilters explicitly names admitted assay types, quantity types, relation handling, units, organisms, and the fixed declared_only target scope. Up to 50,000 normalized rows are admitted. Relation operators and raw decimal strings are never overwritten.

Output

A successful default-dataset result gives family-level counts, conflicts/refusals, source coverage, archive SHA-256 and size, verification status, artifact references, and the molecular-dossier billing decision. The final row is an uncharged run-summary with source retries, bytes, elapsed time, artifact references, accepted event count, and stop reason.

The default key-value store receives:

• MOLECULAR_EVIDENCE_PACK.zip — deterministic portable dossier;
• MANIFEST.json — byte length, SHA-256, schema version, and generation rule for each artifact;
• VERIFICATION.json — archive round-trip and entry-hash checks.

The ZIP contains canonical JSON, safe CSV tables, deterministic Markdown, source map, limitations, schema, checksums, and its internal manifest. CSV display cells beginning with =, +, -, or @ are prefixed to prevent spreadsheet formula execution; raw JSON remains unchanged. Archive paths are generated, normalized, and checked for traversal, aliases, symlinks, executable extensions, decompression hazards, and seeded tamper.

Pricing and charge limits

The launch event is exactly molecular-dossier at $1.00 per verified dossier. One run can emit at most one such event.

• 1 verified dossier = up to $1.00 in Actor event charges, plus platform usage shown by Apify.
• Scope rejection, target/compound ambiguity, required-source outage, rights failure, archive verification failure, internal failure, and charge-cap rejection = $0 in molecular-dossier events.
• The default dataset summary and all failure rows are uncharged.
• Optional sections may be absent only when the input disabled them; a requested required section cannot silently disappear into a paid result.

Before source work, the runtime asks Apify how many molecular-dossier events remain under the user’s cap. If fewer than one can be accepted, it emits an uncharged charge_limit_reached result and summary without contacting a provider. After a verified result is pushed with the event name, it checks charged_count and event_charge_limit_reached; it never assumes the event was accepted. apify-default-dataset-item must remain disabled, and there is no Stripe or second payment path.

The $1.00 launch price follows the approved build packet and is sensible for the controlled default. Seven evidence-heavy local Docker runs at 4 GiB/1 CPU completed 7/7 with 3.611-second p50, 4.056-second nearest-rank p95, 98.625 MiB peak RSS, and zero aborts. At the documented $0.20/CU and 20% Store commission assumptions, the conservative p95 estimate leaves $0.799099 creator net after commission and configured-memory compute. This is candidate-equivalent local evidence, not hosted performance; preserve $1.00 unless hosted evidence justifies an operator-approved change. See PUBLICATION_EVIDENCE.md.

Stateless behavior, partial failures, and negative results

This Actor is stateless and writes no persistent baseline. A provider outage or malformed response that prevents a requested section produces an uncharged source_error; no partial scientific table is presented as complete. Ambiguous identity produces an uncharged, inspectable refusal. Experimental and predicted structures are always separate classes. Censored or incompatible ChEMBL rows retain relation, unit, assay, target, and organism, receive explicit comparability labels, and are never pooled.

Resource limits

• 25 targets and 500 supplied compounds maximum;
• five RCSB entries plus one bounded AlphaFold response per target;
• 50,000 admitted activity rows maximum, paged 200 at a time;
• 5 MB JSON response and depth 40 per request;
• 100 MB verified archive maximum;
• 1 GB minimum, 4 GB default, 8 GB maximum memory;
• 1,800-second hosted timeout.

The implementation keeps only one raw source page at a time and stores normalized rows rather than retaining raw response bodies. Identical recorded evidence produces identical archive bytes; observation time, run ID, retries, and platform metadata are excluded from dossier canonical content.

Data handling and privacy

Only declared public identifiers, family labels/rules, and source evidence are processed. Outputs remain in the run’s Apify dataset and key-value store. Flintglade receives no hidden copy, telemetry, advertising data, payment-card data, or customer secret. See PRIVACY.md for retention and deletion details.

Limitations and publication boundary

This dossier is source evidence, not universal truth. It does not assess therapeutic relevance, efficacy, safety, target validity, or a “best” structure/compound. Provider rights can vary at record level; provenance and attribution must travel with reused content.

The Actor is fully prepared for candidate upload based on all locally closable software, scientific-review, usefulness, security, reproducibility, and candidate-equivalent Docker gates. Paid demand and external researcher testimony are business validation rather than software-readiness blockers and are not claimed. Upload-time platform checks and operator approval remain in RELEASE_CHECKLIST.md. Do not upload, promote, or publish from this packet.

Support: support@flintglade.com. See SUPPORT.md and CHANGELOG.md.