Plan amendments: bash-minimization rule + census-scoped shell slices + effect-plan band decisions

[briansrls]

Summary

Encodes the operator's 2026-07-03 rulings into the plan authorities (docs are generated projections — edits are in the .dag twins, mds regenerated via generated_artifact_gate main_wet):

• Bash-minimization rule (shell-emission twin): bash is the pre-runtime medium — emitted only into foreign executors + bootstrap windows; runtime-present effects are typed argv/REST or binary-interpreted plans; fidelity-bounded ingestion (no "projection-only forever"); thin-run guardrail (typed plan stays authority). Supersedes the ci-humming T5 emitted-converge-shell framing.
• Census-scoped slices: slice 1 shrinks to the If band (pre-runtime census found zero For/While/trap/background sites at pre-runtime locations); slice 2 becomes converge thin-run; slice 3 live_deploy reframed runtime-present; slice 0 gains the Retry.on_exhausted wire-or-narrow precondition and drops the stale env-weld clause (Grow bash AST: EnvUnset node + multi-binding env nesting (dissolve orch_emit_run_env_welded) #5868+2-emit-partition (a): dissolve orch_run env weld into structured EnvPrefixed/EnvUnset AST #6137 discharged it).
• Effect-plan band (host-effect twin, new §1b): layer responsibility contract + governance RED rule; fork-census decisions — Policy de-fork onto IdempotencyEvidence (derived Drive), decide-algebra consolidation onto std.upsert_decision (typed refusal reasons), Retry-vs-temporal DISTINCT, While carries a measure (derived DescentEvidence from dag/std/termination.dag), std.effect_plan rename at Phase B, long-term (not near-term) dissolve onto substrate fragments, dead-vocab deletions, no new std.ProcessInvocation (reuse/extend extdeps service-op argv authority).
• Roadmap nodes: 6-shell-emission/slice1/slice2/shelf-shell-tail re-scoped; 2-emit-partition stale env-weld prose fixed; 2-nbd-serve-held-session-lease re-briefed to instantiate std.upsert_decision.
• Includes main's pre-existing 3-line regen drift (dag-v2-defork-audit.md, fold-ergonomics.md) — idempotent regen output, keeps the drift gate green.

Test plan

• generated_artifact_gate --function main (drift dry-check): exit 0
• plan_lens_witness_test (dissolution-trigger + titled, both): true
• roadmap_authority_test::roadmap_authority_witnesses: true
• roadmap_gate_test::roadmap_gate_witnesses: true
• generated_artifact_drift_test::generated_artifact_drift_witnesses: true
• Full floor runs in CI.

🤖 Generated with Claude Code

[gunbai-bot]

Re cursor/composer-2.5 REQUEST_CHANGES (fold-ergonomics regression): valid — fixed at the carrier in 353bcca5dc.

Verified the finding: the pre-existing md wording was the correct post-#5552 state; the .dag authority (dag/gunbc/plans/fold_ergonomics.dag:31) was the stale side (live tree confirms: src/v2/std/qualified_name.dag:21 defines the alias, zero qualified_name_eq occurrences), so the main_wet regen in 067419b674 faithfully projected stale authority over corrected prose — moving the md backward exactly as the review said.

Fix per the review's second option (update the carrier and regenerate, not a hand-restore of the md): fold_ergonomics.dag:31 now carries the #5552 partial-fix wording + the remaining recursive-tail resolver gap, consistent with dag-v2-defork-audit.md:144,147. The regenerated md is byte-identical to its pre-regression wording (git diff 067419b674^ -- docs/plans/fold-ergonomics.md = empty). Drift gate main exit 0; plan-lens dissolution witness true by execution.

— sent from eager-crane-506