[InauguralPhysicist]

Closes #1943.

Evidence (the Codex Cloud draft's report, reviewed):

Replay-refusal deduplication

replay_blocks(fn, boundary) is now the single implementation. Its callers pass
these inputs:

• subprocess/concurrency: recv, try_recv, recv_timeout, exec_capture,
  proc_read_buf;
• database: db_connect, db_query_value, db_execute, db_query_json;
• store: store_open, store_close, store_put, store_get, store_delete,
  store_query, store_count, store_update, store_collections, store_drop.

A comparison script constructed the old per-helper message and the new shared
message for all 19 inputs and asserted byte equality. It reported
old/new replay messages: byte-identical for all 19 inputs; there was no
disagreement. git grep -n 'not replayable under EIGS_REPLAY' -- src now finds
only the shared format string (plus unrelated inline special-case sites), not
per-boundary helper copies.

Filesystem tape-kind deduplication

The nine literal file_io_error inputs are ls, mkdir, read_bytes,
read_bytes_buf, read_text, remove_file, rename, write_bytes, and
write_text. A comparison script evaluated the removed predicate for every
input and compared it with membership in k_tape_kinds; all nine decisions
were true and identical, with no disagreement. file_io_error_message now
uses trace_core_kind_declared, whose answer comes directly from that table.

tools/tape_kinds_check.sh now also treats literal file_io_error calls as
tape producers. The unchanged tree was green:

tape-kinds: taped=52 declared=52 missing=0 stale=0
tape-kinds: PASS

Planting file_io_error("unlisted_plant", ...) produced the required red:

MISSING: 'unlisted_plant' records an N value but has no k_tape_kinds row
tape-kinds: taped=53 declared=52 missing=1 stale=0
tape-kinds: FAIL

After restoring the source the same check returned to PASS with 52 taped and
52 declared names. No tools/selftests.txt row pins this gate's count.

Checks

• make -j2: passed.
• bash tests/test_replay.sh: 39 passed, 0 failed.
• (cd src && EIGS=../build/release/eigenscript bash ../tests/test_replay_boundary_exit.sh): 15 passed, 0 failed.
• bash tools/tape_kinds_check.sh: passed.
• bash tools/workflow_yaml_check.sh: passed after installing PyYAML.
• bash tools/ci_tier_check.sh: passed after installing PyYAML.
• bash tools/precheck.sh: 14 gates passed, one portability audit skipped;
  four orchestration failures were environmental on the first run: PyYAML was
  initially absent (subsequently installed and both affected workflow gates
  passed independently), and this checkout has no origin remote/ref, so the
  changed-file selftest and changelog merge-base checks cannot resolve
  origin/main.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-24T19:24:09.159950Z	ea07aa1	PR opened
🔒 Security Review	✅ Completed 2026-10-24T19:28:54.087792Z	ea07aa1	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[codspeed]

Merging this PR will not alter performance

✅ 13 untouched benchmarks

---

_{Comparing replay-blocks-one-impl-1943 (e82419b) with main (a8e09cd)}