Delete the 27-row self-host frontier roster and the machinery that only served it

[gunbai-bot]

[gunbai-bot]

The ci red on f083077 had two causes, not one, and only after separating them did the second turn out to be mine. Pushed 045ca6ee131 for it.

The five failing rows are the doc-graph pair asserted from two files plus the conjunction — one defect pair, five row names. That shape is usually main's, and half of it is: main gained docs/probes/root_b_primitive_repr_fork_2026-08-16.md after this branch forked, PR CI evaluates the merge ref, and gunbc#8333 fixes it.

But the row name identifies the lens, not the defect, so "those are the known inherited rows" is cover rather than a diagnosis. Checking the diff for edge-forming content rather than for .md files found this branch removing two HandAuthoredDocBind rows:

- slug: "frontier_probe_exact_head_survey_2026-08-03"
- slug: "frontier_probe_run_interrupted_2026-08-06"

…while deleting no documents. Three files kept their place in the tree and lost their only inbound reachability, with zero markdown links pointing at them. Two new orphans of my own, sitting behind main's.

Deleting the rows was right — every declaration they cite (frontier_probe_exact_head_closeout_note, frontier_probe_run_interruption_note, frontier_roster_provenance_constructor_inflation_note, compiler_frontier_receipt_totality_wall_holds) is deleted by this PR, so the binds would dangle. The missing half is DESIGN §3 step 6: a document whose entire subject is retired is retired with it, never left standing. Retired:

• docs/probes/frontier_probe_exact_head_survey_2026-08-03.{md,tsv}
• docs/probes/frontier_probe_run_interrupted_2026-08-06.md
• docs/probes/diff_frontier_declared_vs_survey.sh — its only input was the TSV above
• docs/probes/run_frontier_probe_survey_per_module.sh — frontier probe transport, served nothing else

Two unrelated STR-RC-0 scaffolds cited the second script as a class exemplar; their comments now state the class without the pointer, so this leaves no stale citation. Residual references to any deleted path: 0.

Executed receipt for the branch tree is running now and I will post it rather than leaving the argument to stand on its own. Expect ci here to keep showing the same five rows until #8333 merges — that residue is main's, and pushing at it from this branch would misattribute a doc repair to the frontier cut in git history permanently.

— sent from smart-ram-730

[gunbai-bot]

Executed receipt, as promised — the branch tree is clean.

Run on 045ca6ee131, remote, release gunbc built in the same dispatch:

HEAD=045ca6e
BIND_ROWS_GONE=0
gunbc run --entry src/v2/lens/doc_reachability_test.dag --function doc_graph_is_clean
  -> function `doc_graph_is_clean` returned `true`

(The trailing returned true, not ProcessExit is the exit-code wrapper refusing to map a bare Bool to a status — it prints the verdict on the way past. The lens ran and answered.)

So the two orphans this branch created are closed, verified by running the lens rather than by reading the diff. The arm is the right one: the SHA is the pushed head and BIND_ROWS_GONE=0 confirms the deleted rows are actually absent from the tree that was compiled.

One check of mine did not report, and I would rather say so than let it pass as a pass: the SUBJECT_GONE probe over the five deleted paths printed empty instead of 0 — nested quoting in the dispatch broke the pattern. An empty result is not a zero; it is a check that did not run. It is redundant here (HEAD plus BIND_ROWS_GONE already pin the arm) but it proved nothing and is not doing any work in this receipt.

Remaining red on this PR is main's single orphan, cleared by #8333.

— sent from smart-ram-730

[gunbai-bot]

Pushed 5921f90b1b4 — all six findings from cursor review 52817 verified against the tree and fixed, plus two more of the same class the review did not reach.

Fixed per site rather than by one substitution across them: a mechanical sweep over prose like this produces sentences that parse and say nothing, which I did once already tonight and am not repeating.

finding	what it was
stage0_crate_layout_emit.dag	the template still embedded Authority: v2.compiler.self_host.frontier in both the .dag and .rs emit paths after the import was repointed — the next regen would have baked a deleted module name into two generated artifacts. This is the one with teeth; the rest are prose.
witness_admission_offline_note	dropped the deleted exclusion note from the paired list
long_lane_exclusion_note	dropped the "Named consumer with file-grain exclusion row" sentence — both things it named are gone
falsifier_self_host_wet_note	the frontier-row flip no longer gates wet enrollment; now says so, pointing at the note recording that the term it fed was already empty
ci_release_bins_witness_note	claimed two positive gap bins while the tests assert one
witness_deferral_freeze_purpose_frontier_note	rewritten, not removed — the terminating-survey specimen did terminate, but by its subject being deleted rather than by reaching 27 of 27 receipts, and that is precisely the disposition the freeze cannot express. Worth more as a corrected case than as a deletion.

Two the review did not reach, found by sweeping every symbol the cut deletes rather than fixing only the named six:

• doc_graph_roots.dag — the bind for this cut's own coordination document had primary_work pointing at v2.compiler.self_host.frontier_band_a_emit_readiness, a module this PR deletes. Repointed to the door-contract note already in its additional_works.
• discrete_cost_derivation (.dag + .md) — cited the deleted exclusion note as live evidence for the "green CI implies coverage" claim. The claim stands; the specimen now reads as one that has since been deleted with its subject, which is an instance of the pattern rather than a loss of it.

Census after: every decl_name bound in doc_graph_roots resolves in the tree. An earlier run of that census reported two missing — both were my grep omitting func declarations, so the instrument was wrong, not the corpus.

Also on this head: main merged in (4e6956fe58d). Two conflicts, both resolved by rule — the modify/delete on frontier_probe_survey.dag kept the deletion (main's change was a prose edit inside a file this PR deletes whole), and .github/workflows/ci.yml is generated, so the merge driver refused it as designed and I did not hand-pick hunks. Regenerating it on the merged tree currently fails (gunbc ci exits 1); I am running a control on main to establish whether that is this branch or the environment, and will not claim the file is correct until that is settled.

— sent from smart-ram-730

[gunbai-bot]

Both findings from review 52854 verified and fixed in b7979cd4bdf. The first was my error, and it is a specific one worth naming.

I repointed to the wrong authority, by name resemblance. I moved Rust's TargetGovernedByContracts to v2.compiler.self_host.emit_coverage_frontier because the name looked like the deleted module. It does not answer the same question — that carrier's own emit_coverage_frontier_two_axis_note says it answers axis (a), whether a witness body runs native, and explicitly disclaims axis (b), module self-emission. The reviewer is right.

That is §18.2's finding — a type name in a diagnostic says the carrier was involved, it does not say how — committed by me, in the same hours I sent it to two other lanes as a correction. The lesson generalizes past type names to any citation picked on resemblance.

Repointed to v2.compiler.self_host.emitter_producer_provenance, the surviving producer authority, which is also where this PR re-homed the promotion obligation earlier — so the two now agree instead of naming different survivors.

And the note beside it was false on its face. governing_contract_id_resolution_note explains this field carries a module name only because the real strategy authority lives in a private overlay, and that the interim must be "a real module in this repository that a reader can open … resolvable rather than dangling." It still named the deleted module while the field had already moved. Updated — and it records the wrong first choice rather than quietly showing the second, because a reader deciding whether this field means anything is better served knowing it was mis-picked once.

Second finding — two dissolution triggers whose subjects the cut removed, in seed_emitter_behavioral_wet_module_bindings.dag:

• :14 authorization waiting on witness-realization + frontier_probe_survey → witness-realization alone, with a note that the probe and the roster it surveyed are deleted, so the trigger is now the whole condition rather than half of one.
• :22 "Frontier rows stay SeedRetained until that path is host-wired" → rewritten to name what actually gates now, the binding's own absence.

A trigger naming a deleted subject is worse than honest prose: it cannot fire, so it reads as checkable while being permanently pending. That is the same defect this PR already fixed once in self_host_promotion_obligations — I found that one by sweeping module_path: fields, and missed these two because they are prose inside dissolution descriptions. Fourth distinct syntactic form for one class: module_path: field, import line, contract_ids list element, and free prose in a trigger.

— sent from smart-ram-730

[gunbai-bot]

Fixed in c352f7dfb91 — and review 52867 is right that the file was untouched, which also means the approval on the same head (review 52863) was wrong where it credited v2_self_hosting.dag with quarry annotation. There is no banner in it. Seven live citations of the deleted frontier.dag: the status line, the state-of-play census, two Wave 1 residual lists, the §2 census block, the Wave 3 exit gate, and the retirement condition.

I did not take the suggested repair as offered, and the reason is the interesting part. Quarry-bannering it would be wrong. wave2_prep_design earned that treatment because its entire mechanism section named deleted roster symbols with no surviving mapping. v2_self_hosting is the live sequencer for a program whose Wave 1 is still in flight — it is not superseded by the cut, it merely cited what the cut removed. Retiring an active plan to make a citation problem disappear loses the sequencing and would be the more expensive error.

So: repoint, and where a claim was denominated by the roster, retire the fraction rather than re-author it.

• 0/27 self-host-green and the Wave 3 exit 27/27 came from the same hand-authored row list, so both measured the roster's opinion of how many compiler modules exist, not the corpus. Substance survives (no module is self-emitted with a green equivalence receipt; every module must flip or be declared pinned-kernel), stated qualitatively, with the derived denominator named as the first Wave 3 obligation. Restoring a count here would rebuild the weed inside the replacement — the DESIGN §3 failure by name, and the one most tempting while 'just fixing citations'.
• Surviving authorities named explicitly at the top: emitter_producer_provenance, stage0_crate_layout, v2_emitter_direct_rust_door_contract.
• Frontier-probe residuals marked retired, not outstanding — that survey never produced a receipt on any host including a dedicated runner, so its deletion costs machinery, not evidence.
• The operator-signed END GOAL quote keeps its 27 verbatim; it is a signed historical block and the roster-deleted receipt now precedes it.

Receipt, remote (claim_batch built and run in one dispatch): PASS doc_graph_has_no_orphan_docs, PASS doc_graph_has_no_dangling_links. A third name I passed, doc_graph_is_clean, does not exist in that entry and returned FAIL (no such function) — kept in the receipt because it doubles as proof the runner reports failures rather than swallowing them, which is the pairing this PR's own lane has been enforcing all night.

— sent from smart-ram-730