Evidence Pack Builder

Build a portable, deterministic, checksum-verified ZIP from selected Apify datasets, selected key-value-store records, or bounded inline JSON.

What this does not do

• No source collection.
• No alteration or silent repair of input evidence.
• No signing claim unless a real configured signing key is used. Version 0.1 does not sign.
• No legal or compliance certification.
• No cloud storage outside Apify.
• No executable files inside the package.

Sources and authorization

Select up to 10 Apify datasets and 10 key-value stores through Apify's resource picker. For every selected store, provide exact recordKeys. The Actor uses only the run's limited, user-granted storage permissions; it accepts no Apify token and makes no remote URL requests. Up to 100 inline JSON objects are supported for demos and automated health checks.

Input example

{"inlineItems":[{"id":"demo-1","claim":"Example evidence row","source":"Flintglade fixture"}],"archiveFormat":"zip","includeSourceSchemas":true,"deterministic":true,"maxRows":100}

Optional redactionRules are non-root RFC 6901 JSON Pointers. Applied and missing pointers are recorded; the original value is represented only by its SHA-256 hash.

Outputs

The run KVS receives EVIDENCE_PACK.zip, MANIFEST.json, VERIFICATION.json, and RUN.json. The ZIP contains canonical UTF-8/LF JSON Lines, a readable README, provenance, redaction log, source schema, empty-or-populated error log, and an internal manifest. Every path and checksum is re-read and verified before delivery.

The default dataset receives one result row and one uncharged run summary. A successful result includes the package SHA-256, row/resource/byte counts, verification status, and artifact keys.

Pricing

evidence-package costs $1.00 per verified package. The event is charged exactly once and only after archive path safety, round-trip extraction, manifest equality, and every entry checksum succeeds. Invalid, oversized, inaccessible, partial, corrupt, serialization-failed, verification-failed, or charge-cap-refused requests are not charged. Platform usage is additional.

State and repeated runs

The Actor is stateless. Identical normalized input produces an identical package hash within the pinned runtime. A new run is a new requested package and may charge once after verification.

Failure behavior

No selected item is silently dropped. If any selected source cannot be read or serialized, the package is rejected and uncharged. Empty selected sources return an uncharged terminal result. The result and run summary explain the stable status and error code.

Limits

Default limits are 50,000 rows, 250 MB uncompressed input, and 100 MB ZIP output; hard limits are 100,000 rows and 500 MB uncompressed input. Launch memory is 512-4096 MB (2048 MB default) with an 1800-second timeout.

Source rights and attribution

Users remain responsible for authorization to read and package selected evidence. The Actor records hashed source provenance but makes no ownership, license, legal, or compliance conclusion. See SOURCE_POLICY.md.

Limitations

Version 0.1 supports ZIP only. Byte-identical compression is guaranteed within the pinned Python 3.12 container runtime, not across arbitrary future Python or zlib builds. KVS binary values are preserved as base64 JSON and never executed or recursively unpacked.

Privacy and deletion

Source storage identifiers and record keys are hashed in outputs. Selected evidence is read only for the run and written to that run's Apify storage. No data is sent outside Apify. Delete the run's dataset and KVS to remove outputs. See PRIVACY.md and SOURCE_POLICY.md.

Support: support@flintglade.com