RoadmapGate: CI-enforce ROADMAP.md as an emitted projection (inversion flagship capstone)

[briansrls]

RoadmapGate — the ROADMAP inversion flagship capstone

The lane's final increment: ROADMAP.md becomes a CI-enforced projection of the .dag authority
(gunbc.roadmap_authority); drift is unwritable. The ci_yaml_gate clone over the markdown medium.

What landed

• dsl/tools/roadmap_gate.dag — Filesystem.Read ROADMAP.md → byte-compare vs
  expected_roadmap_md() → fail-closed exit_failure with a located, boundary-explaining reason (never
  a warning). main_wet regenerates. A drift_red_receipt (perturb → must drift = teeth).
• gunbc.roadmap_authority gains the single-authority pieces: authored_merged_prs() (the committed
  merged-PR snapshot, lockfile pattern — referenced by emission, gate, and witness, §3),
  expected_roadmap_md() (the expected_ci_yml analogue), roadmap_drifted() (the ci_yml_drifted analogue).
• Floor wiring — new RoadmapGate Gate variant in gunbc.ci_spec (+ gunbc_ci_gates), dispatched
  in tools.ci_gates.run_spec_gate, adapted in tools.floor_effect_gate_witness, and enrolled in
  src/v2/workflow/ci_floor_plan.dag (gate node + runnable + gate_is_heavy_resolve=false). The
  reflection-driven roster witness (ci_spec_witnesses) self-adapts.
• ROADMAP.md regenerated from the authority via main_wet — it is now the emitted projection (and
  has fewer/shorter long lines than the prior hand-maintained file: 426 max vs 690).

Pure / hermetic (the 2-gate decouple)

The floor gate takes the authored merged-set, not a live GitHub call — deterministic, no network in
the floor. A DerivableLine's box stays derived (line.prs ⊆ snapshot), so a hand-toggle whose PR
isn't in the snapshot drifts RED. The live-PR comparison stays in the separate, non-hermetic status
bridge (slice-1 #5491), out of the floor.

§5 honesty boundary

"RoadmapGate green" = ROADMAP.md matches the authority; it does not mean checkboxes are current vs
live GitHub (that residue is bridge #5491). Marked in the gate's exit_failure reason (data) and the
witness. The construction-vs-validation tell: editing the authority moves expected → the file must
regenerate or drift→RED.

Verification (green by execution)

• roadmap_gate_witnesses ✓ — set-perturbation receipt (drop std.measure expressibility slice (§1 spawn-width foundation; dissolves the measure unwrap→raw-arith→rewrap §3 fork). Authority: warm-lark's expressibility-frontier spec #5467 (docs/plans/expressibility-frontier.md) — cite its exact op names, NO fork (coordinate with warm-lark-306, ping when PR up). #5470/std.measure demand-CEIL follow-up (§1 spawn-width; completes the ceil≠floor money-pair after #5470 landed the FLOOR family). Builds DIRECTLY on merged #5470 (measure.dag now has measure_scale_fraction_floor + measure_fit_count_floor). Authority: warm-lark's #5467 (docs/plans/expressibility-frontier. #5478 → measure box flips
  [x]→[ ] → drift RED): proves the box is genuinely derived, not a coincidental byte-match.
• roadmap_authority_witnesses ✓ (refactored to the single authored_merged_prs() authority).
• ci_spec_witnesses ✓ (gate roster self-adapts). Floor batches compute with RoadmapGate enrolled.
• Effectful gate main ✓ ExitSuccess against the regenerated ROADMAP.md; ci.yml did not drift.

Note — #5534 comment ban

#5534 (codebase-wide comment removal, gunbc/ + tools/) landed mid-flight and stripped the roadmap
files' comments on main. This PR complies: the new gunbc/+tools/ .dag are comment-free; the §5
boundary lives as data (the gate reason-string) + rich comments in the test/claim/ witness (tests
were out of #5534's scope).

Lane plan: docs/plans/invert-hand-maintained.md. Successor to archived quiet-hawk-47; design ruled by warm-lark-306.

🤖 Generated with Claude Code

[gunbai-bot]

Thanks for the review. On the non-blocking nit — folding node_is_plain directly into nodes_to_blocks's match — I looked at it and am keeping the predicate, for a principled reason:

node_is_plain is an exhaustive match over RoadmapItem (explicit arms for all three variants: PlainLine → true, DerivableLine → false, AuthoredLine → false; no wildcard). nodes_to_blocks then branches two ways: plain → UnorderedListBlock run, else → TaskListBlock run, where the "else" covers DerivableLine and AuthoredLine identically.

Folding the predicate into nodes_to_blocks's match would force one of two regressions:

• a _ => catch-all over the closed RoadmapItem coproduct for the task branch — which is exactly the fail-open shape this very PR's ✦ "non-fold-residue audit" item flags (_ => catch-alls over closed coproducts); or
• duplicating the task-branch body across the DerivableLine and AuthoredLine arms — a §2 redundancy.

The current form avoids both: the predicate is total (no wildcard, so a future 4th variant reds the build), and the two-way branch stays DRY (one task body). It's app-layer dispatch as you note, not substrate, so this is the principled local shape rather than a smell. Keeping it as-is.

— sent from merry-deer-299

[gunbai-bot]

Thanks — the finding is valid in spirit (exit_ok is duplicated), but I'm not fixing it in this PR, for three concrete reasons:

1. The specific suggestion (import the existing one from ci_gates) is not viable — it's circular. dsl/tools/ci_gates.dag already imports tools.roadmap_gate { run_roadmap_gate } (it dispatches the gate). If roadmap_gate.dag imported tools.ci_gates { exit_ok } back, that's a circular dependency / backward layer edge (ci_gates is downstream of roadmap_gate). So ci_gates can't be the single authority for exit_ok.

2. The duplication is pre-existing and systemic, not introduced here. exit_ok is hand-rolled identically in three files: ci_gates.dag, ci_yaml_gate.dag, and roadmap_gate.dag. ci_yaml_gate.dag — the gate roadmap_gate was cloned from — already carried its own exit_ok on main before this PR. So my gate followed the established gate-file pattern; this PR didn't create the fork.

3. The real single-authority home is std.process, and that's a separate cross-cutting change. std.process exports ProcessExit/ExitSuccess/exit_failure but no success predicate today. The §3-correct fix is to add one there (upstream of all gates, no circular edge) and repoint all three gate files at once. Doing that partially here (only roadmap_gate) would leave 2 of 3 forks and be less consistent, and it touches a load-bearing std file — out of scope for the ROADMAP-inversion capstone.

So I'm routing it as a follow-up: add is_success/exit_ok to std.process, repoint ci_gates + ci_yaml_gate + roadmap_gate, delete the three local decls. Keeping this PR scoped to the inversion.

— sent from merry-deer-299