docs(rfc): amend adversarial harness with the DST execution layer

[azimafroozeh]

What & why

This PR amends RFC-032 in place: a new §13 records the deterministic execution layer beneath the harness, and nothing in §§1-12 is restructured.

• It answers the simulation RFC's unresolved question 1: docs(rfc): add RFC 0037 deterministic simulation harness #507 amends this RFC in place, through this companion.
• §3.3 gains its stronger sibling, split by execution layer: normalized replay stays the contract on the real substrate, while instruments in the simulated world get the ID-inclusive (scenario, seed) contract for free.
• Two of §4's instruments now have working twins in the simulated world: the durable-write-cut sweep's crash-state enumeration, and the fuzz generator's shape shared with the operation sampler.
• Two §6 entries are corrected with evidence: the libc half of the Antithesis rejection is a ~60-line interposition in use today, and the never-evaluated lane (contribute the missing seams upstream to Lance) is now open, first ask the injectable clock, evidence recorded in RFC-031 §11.
• §4.1's no-replay disclaimer gains its checked upgrade path: the strict-replay meta-test pattern.

Local verification

• Markdown-only diff to docs/rfcs/0032-adversarial-correctness-harness.md; cargo gates not run: no code touched
• bash scripts/check-agents-md.sh: 77 links, 72 docs OK

Greptile Summary

This documentation-only PR amends RFC-032 to describe how its adversarial instruments relate to the deterministic simulation execution layer.

• Defines separate normalized-replay and ID-inclusive determinism contracts for real and simulated substrates.
• Describes simulated twins for crash-cut enumeration and operation-sequence generation.
• Revises the earlier evaluation of entropy interposition and upstream Lance seams.
• Adds strict replay as an acceptance pattern for any future record/replay arbiter.

Confidence Score: 4/5

The PR should not merge until the nonexistent RFC-031 §11 citation is corrected; the transient review-state wording is otherwise non-blocking.

The amendment presents a specific measurement as recorded in RFC-031 §11, but RFC-031 ends at §10 and contains no such evidence, leaving a central design rationale unverifiable.

Files Needing Attention: docs/rfcs/0032-adversarial-correctness-harness.md

Important Files Changed

Filename	Overview
docs/rfcs/0032-adversarial-correctness-harness.md	Adds the DST amendment, but one evidentiary citation points to a nonexistent RFC-031 section and some rationale is tied to transient review state.

_{Reviews (1): Last reviewed commit: "imp" | Re-trigger Greptile}

Greptile also left 2 inline comments on this PR.

Context used:

• Context used - CLAUDE.md (source)
• Context used - AGENTS.md (source)

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[greptile-apps]

RFC-031 evidence citation is broken

When readers follow this citation to validate the injectable-clock rationale, RFC-031 ends at §10 and contains neither the counting-golden nor byte-wobble evidence, leaving the amendment's stated evidence source unusable.

Context Used: CLAUDE.md (source)

[greptile-apps]

Transient review state in RFC

The amendment explains its durable rationale through “PR #507, in review” and that PR's “unresolved question 1.” Once the review concludes, this wording becomes stale and forces readers to reconstruct historical review state instead of stating the design decision directly.

Context Used: AGENTS.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!