Delete the 56 transition-admission rows orphaned by the removed namespace-wave wall

[gunbai-bot]

First burn-down from the out-of-closure blocking-error census (dashboard node adhoc-a2a92155-454).

What: delete every row under dag/gunbc/namespace/transition_admission/ (56 files). Each one imports gunbc.namespace.transition_admission, which gunbc#11774 deleted along with the namespace-wave-admission wall. The pinned v2.compiler.reference_conservation_census reference_conservation_stratified_sample_paths is left unchanged. That sample is pinned to fec339d and declared byte-for-byte rerunnable, and the census renders a deleted path as an explicit unreadable line, so its six entries for this directory now render that way.

Why the rows exist: #11625 (32 rows), #11565 (10), #11667 (10) and #11689 (4) authored them to adjudicate namespace-wave refusals. Those branches landed after #11774 had removed the wall. #11681's 13 rows are not among them, because #11696 already deleted them. #11771 had already deleted 23 survivors on the same reasoning. gunbc.rung_drop.namespace_wave_admission_wall_removed records the directory as "deleted with the wall", and the namespace-cut plan names none of these rows, so this is §3c dangling, not a migration. (An earlier version of this body and the first commit message named the wrong PRs and said 17; the second commit states the correction.)

Why they landed: before #11829, a refused floor did not fail the job. #11625's floor job 106202716752 printed all of these errors and required-ci: floor refused, and the job still concluded SUCCESS.

Receipt: neat-ibex-861's whole-root sweep of 2026-09-23 (gunbc compile --source-root dag --source-root src/v2 --target dag --repository gunbc --measured-root-demands tools/whole_corpus_compile_measured_root_demands.json) reported 500 blocking errors. 280 of them are these 56 files × 5 (the unresolved import, plus TransitionAdmission, Binding and NotAuthored). After this PR, no .dag or .rs file imports the module. The remaining 220 errors (39 designed-refusal probes, and 181 wall-class errors in 62 modules) are out of scope; the compiler-change vector behind them is declared in #12201.

🤖 Generated with Claude Code

[briansrls]

SOURCE HOLD at exact head b79067af1d60f82d1f962b09c9480b0590c54de1.

The 56 row deletions themselves are correct. #11774 removed the namespace-wave-admission phase and gunbc.namespace.transition_admission; the standing rung-drop explicitly says this roster directory is deleted with the wall; and I found no replacement-authority obligation for these temporary admission rows. All five required checks are green and GitHub reports CLEAN.

Two repairs remain.

1. Do not silently shrink the pinned reference-conservation sample

This head removes six dag/gunbc/namespace/transition_admission/... members from reference_conservation_stratified_sample_paths and adds no replacements. That is not consistent with the authority in the same module:

• it calls this the pinned 2026-09-23 sample over main fec339d561;
• it says each stratum takes max(6, round(...)) files;
• it says the list is kept so the sample reruns byte-for-byte;
• and the census explicitly says an unreadable path is its own line, never an omission.

After this edit the original 315-path sample becomes 309 paths, and the dag/gunbc/namespace stratum has only namespace_structural_observation_bridge_seed_growth.dag left: seven sampled paths become one, below the stated minimum of six even though the live namespace stratum still has enough <=8 KB files. This rewrites and weakens the evidence denominator while leaving the declaration of that denominator unchanged.

Required disposition: either leave the six paths in the historical pinned sample, where deletion honestly yields six unreadable lines, or mint/recompute a new, separately identified sample from this exact corpus using the declared selection rule while preserving the old sample and its receipts. Do not hand-shrink the pinned sample and continue calling it byte-for-byte stratified evidence.

2. Correct the provenance account

The current body and source commit say #11565, #11667, #11681, #11689 and #11625 account for these 56 survivors, and that #11771 deleted 17 rows. The exact history does not support that:

• the 56 rows being deleted divide as 32 owned by #11625, 10 by #11565, 10 by #11667 and 4 by #11689;
• #11681 contributes none of these 56, because #11696 already deleted its 13 admission rows;
• the final #11771 diff deleted 23 transition-admission rows, not 17.

Please amend the commit/body provenance while changing the sample treatment. The core deletion should then be a straightforward re-review; I found no objection to removing all 56 orphan rows.

[briansrls]

APPROVE-MERGE at exact head 5f7309e3480b86477c0f541f86cbf95f34e88ff7, superseding my REQUEST_CHANGES review at b79067af1d60f82d1f962b09c9480b0590c54de1.

Both held issues are discharged.

1. The historical sample is preserved. The only source delta from the held head restores the six dag/gunbc/namespace/transition_admission/... members to reference_conservation_stratified_sample_paths. The sample therefore remains the pinned, byte-for-byte 315-path denominator over fec339d561; after the row deletion, those six members produce the census's explicit unreadable <path> result rather than silently disappearing. The PR diff against main is now exactly the 56 deleted roster files: 56 changed files, zero additions, 1,064 deletions.

2. The provenance account is corrected. The second commit and PR body now state the actual surviving population: #11625 = 32, #11565 = 10, #11667 = 10, #11689 = 4; #11681 contributes zero because #11696 already deleted its 13 rows; and #11771 deleted 23 rows, not 17.

The underlying deletion remains the correct disposition: #11774 removed the namespace-wave-admission phase and its carrier, while the standing rung-drop says the transition-admission roster directory is deleted with the wall. I found no replacement-authority obligation in these temporary rows.

Compiler, clippy, emit-build, floor and the witnesses rollup all pass at this exact SHA, and GitHub reports the pull request mergeable. No further condition remains unless the head moves.