Host SHA-256 primitive sha256_hex_of_text beside the HMAC seam (d0_store_operation_wall, before PR 4a)

[gunbai-bot]

Host SHA-256 primitive for the d0_store_operation_wall chain (proud-deer-538, option A). It lands before #12537 (PR 4a), which consumes it.

Why. The fabric store door digests the approved intent on every protected write, checking that the SHA-256 of the presented intent is the approved revision. The only substrate SHA-256, the pure extdeps.crypto.sha2 fold, costs about 200k interpreted eval steps per 64-octet block, measured below. That put #12537's admission witnesses at 533k–910k steps against the 72.3k new-witness budget, and it is the door's per-write cost too.

What. sha256_hex_of_text(text) returns the lowercase-hex SHA-256 of the text's UTF-8 bytes. It is wired exactly like hmac_sha256_hex, and is the same RustCrypto sha2 crate the HMAC seam already uses:

• src/v1/04_method.dag builtin signature, plus the regenerated v1_compiler_infer_method.rs (the stage0 --required-regen candidate; only this file drifted);
• the interpreter arm and sha256_hex_of_text_digest, plus the regenerated v1_interpreter_dispatch_generated.rs;
• the gunbc.v1_interpreter_primitive_surface authored arm;
• std.primitives sha256_hex_of_text_contract and both rosters;
• the gunbc.primitive_egress evidence row.

Seed growth. gunbc.host_crypto_digest_seed_growth enumerates both host digest seams: sha256_hex_of_text_digest, and hmac_sha256_hex_tag, which had no receipt. It is registered in gunbc.seed_growth_admission. Retirement trigger: an emitted rt_function_registry bridge body admitted by the #12389 gate, so that the digests run in emitted code.

v1 admission. This serves v2's fabric store door, which is the purpose test of gunbc.v1_maintenance_standing v1_seed_standing. That standing records refusals, not per-change admissions, so the admission is stated here and in the seed-growth receipt.

Evidence.

• On the floor: test.claim.sha256_host_known_answer_witness. Nine host-vs-known-answer claims, at most 7 eval steps each, all PASS locally:
  • empty, abc and the NIST 448-bit text, whose values test.claim.sha256_fips_witness_test (grandfathered) asserts of the pure fold, so host == pure holds transitively through the shared value;
  • the padding boundaries 55, 56, 63, 64, 65 and a 130-octet three-block text, with known answers from an independent implementation (Python hashlib). On these the pure fold is not asserted on the floor.
• Local receipt (not gated): the direct comparison, host == pure fold == known answer, on all nine vectors. It was run with claim_batch built from this branch, and all nine PASS. Its eval steps were: empty 236,231; abc 205,849; 55 → 206,192; 56 → 391,980; 63 → 392,807; 64 → 392,661; 65 → 393,392; 448-bit 396,340; 130-octet 582,121. That is the per-block cost that motivated this change.

🤖 Generated with Claude Code

[gunbai-bot]

Addressing review 72974 in b583a71. Both placements are fixed. sha256_hex_of_text_contract and its comment now sit above the HMAC "Issuance:" block, which is back on hmac_sha256_hex_contract. sha256_hex_of_text_digest sits above the /// doc, which is back on hmac_sha256_hex_tag.

— sent from lively-dove-256

[gunbai-bot]

Addressing review 73004 in 402742c. All three findings are fixed.

1. Citations. The unresolvable extdeps.crypto.hash sha256_hex_of_text is gone. The interpreter comment and the egress row now cite std.primitives sha256_hex_of_text_contract as the home, and extdeps.crypto.sha2 sha256_hex as the pure fold sharing its known answers.
2. Consumer. The frontier and its trigger are now on sha256_hex_of_text_contract. The consumer is gunbc.fabric_store_operation_admission intent_digest in Fabric store per-operation admission fold (d0_store_operation_wall PR 4a) #12537, which already calls it on its branch. The trigger is that change landing. If the door digests some other way, the primitive is deleted with it.
3. Arm placement. The SHA-256 arm now sits above the HMAC issuance comment, under its own one-line comment (total, no refusal).

— sent from lively-dove-256

[gunbai-bot]

Folded into the integration branch integration/2026-09-30 (operator ruling 2026-09-30); this head is merged there as-is. — sent from proud-deer-538