chore(engine): regenerate dist/soul.c from merged main — the build input was 4 days behind the source #133

2026-08-07T16:16:56Z

tim.lingo commented

2026-08-07 16:16:56 +00:00

Merging #109/#114/#115/#127/#130 landed the source on main. It did not change what the build compiles. CI compiles the committed dist/soul.c, last regenerated 2026-08-03.

Measured on origin/main before this commit:

string	count	meaning
`conv_hist_key`	0	the per-session history key
`agentic_safety_screen`	0	the #129 crisis-escalation fix
`plain_chat_envelope`	0	#109

A CI or GKE build off main today would have shipped an engine with none of the five merged fixes, including the P0 safety fix, while main's source read as correct. That gap is the whole of #111.

After regeneration: conv_hist_key 9, agentic_safety_screen 3, plain_chat_envelope 5. 1,164,103 bytes, 1226 inlined bodies (gate wants ≥1200). Compiles clean — 903,096 bytes, 0 errors. Boots on a throwaway port with a throwaway HOME and answers /health.

Rung: RUNS. Not in a DMG. Production on :7770 never touched.

Two things this does not fix:

Nothing in the tree regenerates this file. Only a human running the recipe. It lags in batches, never per-change, and it will drift again. neuron-ui#209 (the CI automation) is the real fix.
The shipped round-9.1 brain still matches no committed source — built from a branch with an amalgam deliberately not committed (8b06bbe). Strings in that binary (plain_chat_envelope, __working__, max_tokens":16384) were absent from every committed state. This makes main reproducible going forward; it does not retroactively fix the artifact already handed out.

Refs #111

🤖 Generated with Claude Code

Merging #109/#114/#115/#127/#130 landed the **source** on main. It did not change what the build **compiles**. CI compiles the committed `dist/soul.c`, last regenerated 2026-08-03. Measured on origin/main before this commit: | string | count | meaning | |---|---|---| | `conv_hist_key` | **0** | the per-session history key | | `agentic_safety_screen` | **0** | the #129 crisis-escalation fix | | `plain_chat_envelope` | **0** | #109 | A CI or GKE build off main today would have shipped an engine with **none of the five merged fixes**, including the P0 safety fix, while main's source read as correct. That gap is the whole of #111. After regeneration: `conv_hist_key` 9, `agentic_safety_screen` 3, `plain_chat_envelope` 5. 1,164,103 bytes, 1226 inlined bodies (gate wants ≥1200). Compiles clean — 903,096 bytes, 0 errors. Boots on a throwaway port with a throwaway HOME and answers `/health`. **Rung: RUNS.** Not in a DMG. Production on :7770 never touched. **Two things this does not fix:** 1. **Nothing in the tree regenerates this file.** Only a human running the recipe. It lags in batches, never per-change, and it will drift again. neuron-ui#209 (the CI automation) is the real fix. 2. **The shipped round-9.1 brain still matches no committed source** — built from a branch with an amalgam deliberately not committed (`8b06bbe`). Strings in that binary (`plain_chat_envelope`, `__working__`, `max_tokens":16384`) were absent from every committed state. This makes main reproducible going forward; it does not retroactively fix the artifact already handed out. Refs #111 🤖 Generated with [Claude Code](https://claude.com/claude-code)

tim.lingo added 1 commit 2026-08-07 16:16:58 +00:00

chore(engine): regenerate dist/soul.c from merged main — the build input was 4 days behind the source

Neuron Soul CI / build (pull_request) Failing after 14m33s

Details

Neuron Soul CI / deploy (pull_request) Has been skipped

Details

17e53456b5

Merging #109, #114, #115, #127 and #130 landed the SOURCE on main. It did not
change what the build actually compiles. CI compiles the committed
dist/soul.c, and that file was last regenerated 2026-08-03 (e610a41). Measured
against origin/main before this commit:

  conv_hist_key          0     (the per-session history key — #129's fix)
  agentic_safety_screen  0     (the crisis-escalation fix itself)
  plain_chat_envelope    0     (#109)
  size                   1,147,662 bytes

So a CI or GKE build off main today would have shipped an engine with none of
the five merged fixes in it, including the P0 safety fix, while main's source
read as correct. That gap is the whole of issue #111.

After regeneration from merged main:

  conv_hist_key          9
  agentic_safety_screen  3
  plain_chat_envelope    5
  size                   1,164,103 bytes, 1226 inlined bodies (gate wants >= 1200)

Compiles clean: 903,096 bytes, 13 warnings, 0 errors. Boots on a throwaway
port with a throwaway HOME and answers /health — RUNS, not just BUILT.
Production on :7770 was never touched; no live store was read or written.

Recipe: gen-soul-amalgam.sh (elc --target=c in a scratch tree with the .elh
headers removed, which is what forces full inlining) then cc-brain.sh.

TWO THINGS THIS DOES NOT FIX, both worth their own work:

1. Nothing in the tree regenerates this file. It is only ever updated by a
   human running the recipe and committing, and history shows it lags in
   batches, never per-change. It will drift again. neuron-ui#209 tracks the
   automation and is the actual fix.

2. The shipped round-9.1 brain still matches no committed source: it was built
   from fix/liveness-engine-91 with a regenerated amalgam that was deliberately
   not committed (8b06bbe). Strings present in that binary — plain_chat_envelope,
   __working__, max_tokens":16384 — were absent from every committed state.
   This commit makes main reproducible going forward; it does not make the
   already-shipped artifact reproducible retroactively.

Refs #111

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>