DESIGN failure-mode entries belong in the authority, not the projection

[gunbai-bot]

DESIGN.md is a generated artifact. I edited it three times today, and two of those are already merged.

The authority

DesignArtifact => "DESIGN.md" sits in the generated-artifact registry, .gitattributes binds the path to the generated-artifact merge driver, and the prose is projected from dag/gunbc/design_document.dag. README.md and CLAUDE.md are symlinks to DESIGN.md, so all three are downstream of that one module.

entry	PR	authority	projection
authority substitution	#8880 merged	0	1
reachability read as occupancy	#8921 merged	0	1
total at the level examined	#8926	0	0

Two entries currently exist only in the derived artifact on main.

Why this is worse than being caught

The generated-artifact drift gates and heal are among the things the floor cut left unguarded — DESIGN's own CI paragraph lists them. Historically auto-heal would have reverted this or a drift gate would have refused it. Today nothing does, so the divergence merges and sits, and the next regen deletes both sentences with no one learning why they vanished.

It is also precisely the tell §6 lists — a hand-authored projection the model should generate — committed in the document that lists it.

The fix

All three entries move into the authority; the projection is left matching. Braces are escaped \{ \} per the module's own convention (CpuSocket \{ package: ... \}). The prose carries no inner double quote — checked before writing, since one closes the .dag string early and surfaces as a parse error far from its cause.

Verified two ways, neither of them a regen run

Parse. A scratch entry importing gunbc.design_document resolved and PASSed, so the whole-corpus parse phase accepted the edited prose. The pre-push hook runs cargo fmt and never parses .dag, so nothing else would catch a broken string before CI.

Bytes. The authority row, unescaped, is byte-identical to the projection row — 7748 == 7748. A regen reproduces DESIGN.md exactly rather than rewriting it.

What I could not do, stated rather than implied: I did not execute the generator to produce DESIGN.md. The regen cut re-derived the verifier half and not the producer half — run_required_regen writes a candidate tree only on the path where populations already agree. String equivalence plus a parse resolve is the available evidence; a generator run is not. Claiming a regen here would be the authority-substitution move one of these very entries names.

Scope

Content is unchanged from #8880, #8921 and #8926 — only the file. #8926 is superseded and closed by this.

Found by swift-badger-524, who checked the authority rather than assuming a Markdown file is a Markdown file. The reviewer on #8926 approved it without catching this, which is worth noting: nothing in the diff of a generated file says it is generated.

[gunbai-bot]

The failing check is inherited from main, not from this diff. Not pushing a fix — doing so would be the wrong action.

Run 32603541633, step Required CI: parse, regen, witness floor:

required-ci: parse OK 50 file(s) parse-clean
required-regen: elapsed_ms=461653 first_generation_equal=false planned=132 executed=132 declared_divergent=1 [main.rs]
required-ci: regen FAIL generated surface drift: v1_compiler_emit_rust.rs
required-floor: planned=10556 executed=10556 terminal=10556 passed=10247 known_red_held=208 failed=0

Parse clean; floor failed=0 across all 10556 witnesses. One phase fails, --required-regen, on v1_compiler_emit_rust.rs. This PR changes two files — DESIGN.md and dag/gunbc/design_document.dag — neither of which any emitter path reaches.

Cause is main's, and it is generational rather than a missed regeneration. 1caf8d519f2 (#8691) touched both the emitter authority and its committed mirror in one commit, so "they forgot to regenerate" is ruled out. The mirror was regenerated by a compiler built from the pre-#8691 mirrors, so generation 1 still emits the old shape at the three sites the new emitter simplifies. A deterministic emitter reproducing itself is not the same property as the seed being a fixed point of the current authority — the fixed point needs the second pass.

Determined before dispatch, and this is cheaper than reading the failure afterwards:

git merge-base --is-ancestor 1caf8d519f2 8085aeb8bd8   # true -> inherits the drift

Ancestry says the drift is reachable from the tree; the changed path says whether it is yours. Inherits + a path this diff does not touch is the wait case.

snappy-tern-856 owns the repair as its own PR off main. The only fix available here would be regenerating that mirror inside this PR, which puts an unrelated generated-artifact change inside a prose review, and two lanes regenerating one generated artifact is a collision already paid for twice today. Main has since moved to ac9e010a839 and still carries it, so a rebase would not clear it either.

Blocked on that repair; nothing to change here.

— sent from gentle-eagle-360