Own bounded static acquisition end to end

[forkwright]

Finding

Zetesis still does not own an enforced end-to-end static-acquisition mechanism. PR #78 landed the primitive safety floor on main at c922b43b909cf43179b2bab6730ca40e9f7edf0b, but the public Crawler boundary remains implemented outside the crate. Redirect validation, connect-to-validated-address behavior, streaming/decompression/extraction limits, and evidence-envelope construction are therefore still prose obligations for each consumer.

Primitive floor already landed

PR #78 now provides and gates:

• immutable, externally non-forgeable ValidatedTarget values;
• a separate non-serializable, non-cloneable local-target authorization capability instead of caller-controlled allow_local_targets data;
• private PageContent fields whose constructors and deserializer share URL/content-type/body/text limits, without claiming post-allocation validation is streaming protection;
• compile-fail doctests in the required hybrid gate.

Those invariants are no longer acceptance work for this issue and must not regress.

Remaining evidence

• crates/sylloge/src/crawler.rs is still a public trait implemented outside the crate; its redirect and connect-time requirements are prose.
• No crate-owned StaticAcquirer currently owns private no-auto-redirect transport, per-hop validation, validated-address connection, or bounded wire/decompression/extraction.
• No versioned evidence envelope currently binds the hop chain, exact limit profile, raw/extracted digests, extractor identity/version, and deterministic schema-versioned fingerprint.
• The required malicious-response and positive anonymous-public-GET fixtures do not yet exist.
• Paid-provider authorization issue Make paid-research authorization atomic, identity-bound, and fleet-scoped #47 remains unrelated: anonymous static HTTP GET needs network authority and resource bounds, not paid-research budget approval.

Why this matters

Provider-controlled URLs and responses cross an SSRF and resource-exhaustion boundary. Leaving redirect handling, DNS pinning, decompression, extraction, and evidence identity to each consumer guarantees divergent implementations and makes a caller appear safe after only the first hop. Dioptron and Aletheia should consume one Zetesis producer rather than reproduce this state machine.

Remaining correction

Land one concrete, consumer-neutral StaticAcquirer whose public operation owns the entire state transition:

• keep its one-hop transport private with automatic redirects disabled;
• validate the initial target and every redirect before each hop, connect only to the validated address set, preserve the URL host for HTTP Host/TLS SNI, and impose a finite redirect limit;
• stream and separately bound wire bytes, decompressed bytes, and extracted text; reject unsupported or ambiguous encodings and stop at the first exceeded limit;
• return a versioned evidence envelope containing requested/final target, validated hop chain, fetch timestamp, response metadata, exact limit profile, raw-content digest, extracted-content digest, extractor identity/version, and a deterministic schema-versioned fingerprint;
• execute malicious-response fixtures for retargeting, local redirects, DNS-answer changes, chunked overruns, decompression bombs, oversized extraction, and envelope/fingerprint stability in required CI;
• prove with a positive fixture that anonymous GET to a public HTTP(S) target succeeds without paid-provider authorization or ledger, while local/private targets remain denied without the separate authority;
• keep anonymous static GET independent of Make paid-research authorization atomic, identity-bound, and fleet-scoped #47 and expose the landed producer to Dioptron and Aletheia only through thin adapters.

Done when: the concrete acquirer—not a marker trait or empty shell—enforces this full path in crate-owned code, its positive anonymous-GET and negative security fixtures run in required CI, and consumers can persist the returned evidence envelope verbatim.