ROOT-1: consume the approved meaning/externalization/quality-floor copy into the DESIGN authorities

[gunbai-bot]

Supervised DESIGN authority cut per the ROOT-N serial program (operator-ratified; source copy approved on #9769 at bbe0d9d and merged as 2603a63).

Approval target: head 5e7e6bb2efd5b2c90449398350e75c329e2eb7b5 — contains main@b41d56484f (which contains bfa440b66d / #9778 and 22ab698d9f). The earlier heads e599eed112 / 67a1fddc6d are superseded by the mainline-integration cycle: additive roster conflict with #9741's surface_shorthand_preempts_resolved_identity resolved theirs-first in source order (one splice defect — an unclosed row — was caught by the regen REFUSING to parse, and repaired).

Exactly five paths vs main:

• dag/gunbc/design_document.dag — P1 (§3 meaning fork), P2 (§4b quality floor), P3 (§5 externalization), verbatim from the approved copy
• dag/gunbc/recurring_failure_mode.dag — meaning_fork and externalized_degradation rows + roster tail (evidence empty per the module's standing note)
• DESIGN.md (+8: three paragraphs at their anchors + two index identities), docs/design-ledgers.md (+2 entries in roster order) — regenerated by tools.generated_artifact_gate.main_wet, repo-built binary, fresh srv2 clone of the pinned SHA under a 16 GiB docker cgroup bound (the modeled host-budget source)
• docs/plans/meaning-externalization-design-placement.md — DELETED per the operator terminal ruling; no HandAuthoredDocBind

Receipts: regen produced the projections from the repaired merge tree 0566b5fd; a no-op verification run of main_wet on the final head (empty working-tree diff) is attached in comments. BuildBuddy empty-diff anomaly recorded and root-caused in comments (OOM rc=137 masked by a pipe; budget must fit the slot).

Merge gate: CI green + side-chat exact-head approval at 5e7e6bb2ef (standing operator grant: merge with side-chat approval). Repository-wide merge freeze in effect: this PR is the only permitted merge until ROOT-1 lands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QhdaHkmZPoQoeUWz5V2mzF

[gunbai-bot]

Correction to the anomaly diagnosis recorded in the regen commit message: the BuildBuddy empty-diff runs were NOT a stale tree. The discriminator run proves the tree was exact (8965ebfa; design_document.dag grep=1, DESIGN.md grep=0, correct pre-regen) and shows main_wet was OOM-killed at rc=137 mid-run: the derived budget (MemTotal×0.8 ≈ 6.1 GB) exceeds what the ~7.5 GB runner VM can actually give after build residue, so the kernel killed the process before any artifact write — and in the earlier dispatches the | tail pipe swallowed exit 137, rendering a killed run as success-with-empty-diff (the pipe-discards-exit-code class). Note the inversion: the first probe's accidentally INT_MAX-clamped ~2 GiB budget was survivable (cap-bounded eviction completed); the 'corrected' larger budget caused the kill — the budget must be derived from what the slot can actually deliver, not from MemTotal. The srv2 receipt (fresh clone, 16 GiB docker cgroup bound, clean completion, exact two-file diff) stands as this PR's regeneration receipt. Follow-up for the DEMAND-1 preflight: on BuildBuddy either set the budget well under the VM's real capacity (typed eviction beats SIGKILL) or use the srv2 container route. — sent from jolly-swift-749

[gunbai-bot]

No-op regeneration receipt for the approval target: on a fresh srv2 clone of 5e7e6bb2ef (repo-built binary, 16 GiB cgroup bound), tools.generated_artifact_gate.main_wet ran to completion writing all committed artifacts and git status --porcelain returned empty — the committed projections are byte-exact for this tree, which contains both current-main content and the ROOT-1 edits. This is the ruling's step-3 receipt. Remaining gates: CI on this head, then side-chat exact-head approval. — sent from jolly-swift-749