Source Connector SDK

Most scraping code starts as a pile of fetch() calls and regexes and turns into an incident three months later -- a redirect that leaks past an allowlist into a private IP, a price parser that turns 'from $9.99' into a confident $9.99, a debug log that quietly persists a session cookie. This SDK is the connector layer built to make those failure modes structurally hard to reach, extracted so you do not have to re-discover them.

WHAT SHIPS: a SourceConnector contract (implement only the capabilities your source actually has); a DefaultFetcher that cannot be constructed without an explicit policy admission -- exact-host allowlisting, SSRF refusal including the IPv6 transition ranges that embed a private IPv4 address, a body-size cap enforced while streaming, and a fixed truthful User-Agent with zero stealth, fingerprint or proxy surface (enforced by a test that scans the whole source tree for evasion-shaped identifiers); a JSON-LD / embedded-JSON / static-HTML extraction ladder; a price parser that returns integer minor units or null with a stated reason, never a guessed float; evidence records that are scoped, then redacted, then hashed, in that fixed order, so the checksum always covers the redacted bytes; token bucket, backoff with jitter, circuit breaker, kill switch and idempotency keys; and field-coverage drift detection.

You bring your own policy/admission store (a working reference implementation ships in the tests) and your own serving layer if you do not want the built-in fetcher. Ships as compiled ESM plus .d.ts, zero runtime dependencies. Illustrative hosts in comments and tests are RFC 2606 example domains.

VERIFIED: 205 tests, measured by running the suite.

DELIVERY: signed download of a hash-verified tarball, immediately on purchase. Permissive licence: unlimited products, unlimited clients, unlimited seats, no attribution, perpetual and irrevocable. One restriction, do not republish the source as source.