Public API Reproducibility Snapshotter

Fetch bounded public or user-authorized JSON API pages with GET-only requests, preserve request/response evidence, canonicalize the payload, record pagination and cache behavior, and compare later snapshots deterministically.

What this does not do

• It does not crawl arbitrary websites or follow unconfigured links.
• It does not send POST, PUT, PATCH, or DELETE requests.
• It does not automate a browser.
• It does not discover or bypass credentials.
• It does not claim that an observed schema is a provider's complete contract.
• It does not execute links or code found in responses.

Inputs and cost

Provide 1 to 100 explicit HTTPS JSON URLs. Pagination is declared per request and is capped at 20 pages. Optional headers are secret fields and are allowlisted. volatileJsonPointers removes only explicitly named fields from the canonical comparison; the output manifest records every removal attempt.

The default example is one Flintglade-controlled fixture page. One successfully normalized page, including a valid 304 reconstruction, is one api-page-snapshot event at $0.05. Ten normalized pages are up to $0.50 in Actor events, plus any platform usage shown by Apify. A blocked URL, invalid pagination, authentication failure, rate-limit exhaustion, transient failure, invalid JSON, or oversized response is not charged. The Actor stops scheduling new pages when the run charge cap is reached. The synthetic apify-default-dataset-item event is disabled; summaries and failure rows are uncharged.

Sources and authorization

The Actor fetches only explicit user URLs that satisfy the HTTPS, DNS, redirect, body-size, and content-type gates. A user must have authorization to request any protected endpoint and must supply credentials through masked Apify input. No automatic HTML fallback exists. Source metadata is evidence from the provider, not a universal truth claim.

Output

Each normalized page is a result envelope containing the redacted request, response status and safe headers, redirect chain, raw hash, canonical payload, observational schema, pagination manifest, cache metadata, and the AP-001 schema-drift section. The raw response is retained only when retainRaw is true. Large artifacts belong in run storage rather than a dataset cell.

The run ends with one uncharged run-summary row containing requested, normalized, reused, failed, and charged counts. Failure states distinguish input rejection, source outage, rate limit, invalid response, terminal 404/410, and internal invariant failures. A contacted 404/410 is a chargeable terminal negative page result because the evidence is the requested outcome.

Persistent monitoring requires a named key-value store

One-off runs are stateless. Persistent comparison uses the user-selected baselineStoreId; the Actor never silently falls back to stateless mode when a requested store is unavailable. A baseline is committed only after Apify accepts the charged result. A 304 reuses a validated prior canonical payload; an outage cannot erase or advance a baseline.

Data handling and limitations

Apify run storage and the user-selected Apify key-value store receive the output. Flintglade does not receive payment-card data and no external database, advertising tracker, or hidden analytics is embedded. Delete run data through Apify. Raw source bytes are not retained by default. Query values, credentials, cookies, signed URLs, and authorization headers are redacted from output and logs. The Actor is not a source-of-truth certification service.

Support

Support: support@flintglade.com. See CHANGELOG.md for release changes and SOURCE_POLICY.md for the current source contract and maintenance probe.