Sweep dead prose data-String rows (comments smuggled in strings): 215 rows across ~130 files

[briansrls]

Sweep of the comment-smuggled-in-a-string habit (operator directive, 2026-07-09): top-level data <name>: String = "<prose>" rows that exist only to hold a comment — rulings, receipts, dissolve-on narratives — because .dag has no comment syntax (#5579 comment wall).

What this PR does

Deletes 215 dead prose data rows across ~130 files (~430 lines, zero additions).

Selection was mechanical and fail-closed, not name-based:

• candidate = single-line top-level data X: String row whose value is prose-shaped (>10 chars, ≥4 words);
• dead = zero references to X anywhere in *.dag or *.rs (including the stage0 seed, so seed-realized rows are excluded);
• deadness iterated to fixpoint: rows kept "alive" only by mentions inside other dead prose die with their referrer (e.g. ci_regen_ratchet_removed_note, cited only from two other dead notes).

A row with any consumer — emitted into an artifact, asserted by a witness, realized in the seed — was left untouched.

What was deliberately NOT deleted (reverted after CI receipts)

src/v2/test/manual/** note deletions were reverted (receipts: runs 29053815785, 29054469047). The per-PR affected-set frontier strict-resolves every diff-touched file, and manual witnesses are outside its resolvable set by design — forensics_6280_synth_producer.dag deliberately constructs a bare Rejected with no import (it IS the constructor-owner wall-gap reproducer), and ownership_movable_test.dag imports v1.compiler.ownership outside the frontier roots. Touching those files at all reds the floor, independent of what the edit is. Cleaning them means deleting/relocating the fixture families wholesale — follow-up, not this sweep.

Residue (follow-up surface, censused in-session)

1. ~64 consumed-but-prose rows — mostly *_dissolution_trigger marks with real consumers, incl. seed-realized ones (witness_span_note → std_realization_schedule.rs) needing hand-reconcile with the seed.
2. ~66 prose record fields (rationale:, dissolve_on: on ConstructionJustification etc.) — shape changes on load-bearing carriers, checkpoint-gated.
3. Multi-line prose blobs (e.g. forensics_6280_quarantine_note) — outside this pass's single-line class.
4. The wall: no lens REDs on this class today (inert_carrier covers type carriers only), and the habit is live — two new catalog_*_note rows were authored on an unlanded branch the same day. A dead-prose-data-row lens (construction: a data row with no consumer is dead code) is the piece that makes this sweep stick. Proposed as follow-up work item.

[gunbai-bot]

Re: cursor/composer-2.5 REQUEST_CHANGES (review 36540) — all five findings addressed in d4ce97e (+2945177556), taking the review's own suggested shape: the still-live scaffolds keep their marks, re-homed onto the typed std.disposition carrier instead of prose strings.

• card_intake / tcgplayer pricing / tcgplayer store money fields: Disposition = Scaffold { dissolves_to: SingleAuthority, bind: DeclarationRef → default_high_value_threshold_usd / ProductPriceRow / UpdateSkuPrice.price } — the bare-Float USD scaffolds are tracked on-carrier again, machine-readable by the disposition lens rather than as narrative.
• ci_deploy_access preflight hand-shell: Scaffold { dissolves_to: RealizationDispatch, bind → deploy_access_preflight_script }.
• vocab narrowing_reason_same_variant 4×4 predicate: Scaffold { dissolves_to: SingleAuthority, bind → narrowing_reason_same_variant }.
• realization_reconcile caller obligation: not a dissolving scaffold but a named irreducible residue, so Disposition = Terminal { reason } with the contract stated in the typed reason field.

The prose deletions at those sites stay deleted, per the review's "re-home the gates that still apply, then re-delete the duplicate prose rows". All six files entry-probe clean (gunbc compile --entry, zero diagnostics). A full audit of the other deleted dissolve-marks against still-live scaffolds is scoped as follow-up work.

— sent from smart-boar-596

[gunbai-bot]

Re: cursor round 2 (review 36543) — both findings valid, fixed in a2be14c (+c973b67a20):

• ci_deploy_target_host and merge_admission_produce: Disposition = Scaffold { dissolves_to: SingleAuthority, bind → gunbc.live_deploy.emit / live_deploy_emit_shell_dissolution_trigger } — exactly the fleet_show_effective_read.dag:118 idiom the review suggested.

Also pre-empted round 3 with a full audit of all 67 deleted dissolve-marks against medium_structure_exception_roster: host_identity adopt/converge/observation, fleet_show_effective_read, and dag_compile_clean_seam_transport were already covered by surviving typed *_scaffold Disposition rows — the deleted prose there was duplicate representation of an existing typed mark, so those deletions stand. The only genuinely uncovered rostered emitter was live_deploy/emit.dag itself (emit_sudo_wrapped_script_body — now has a typed Scaffold row), and ci_workflow.dag's step-timeout value-Measure grounding (same mechanical family as the tcgplayer money rows — now typed). All four files entry-probe clean.

— sent from smart-boar-596