QA Mission — Tidy-First Sequencing

Sequencing note, 2026-07-13. Written while planning the second of the "two missions, perf first" pair (the CI test-topology-performance mission shipped as PR #2609; this covers the sibling #2071 CT-friction slice).

The question

Would running one or more degod / deshim / dead-code-removal missions first make the planned #2071 test-QA mission more efficient?

Bottom line: yes, but only for the slice of test friction that is downstream of src structure — and the CaaCS co-change data should pick that slice, not a blanket "degod everything first." A broad degod-first sequencing would delay the QA mission for partial benefit and serialize it behind the whole #1797 / #2173 program. A narrow dead-code/deshim pass plus interleaved targeted degod is almost pure win.

Why it depends: two independent causes of test friction

The discriminator is why a given test is painful. The two causes respond differently to structural cleanup:

  1. Test-intrinsic friction — fakeable DoDs, over-mocking a clean unit, retry-to-green, quarantine because of a real flake, legacy contracts. This is the bulk of #2071's core (CT3 #2074, CT4 #2075, CT5 #2076, the CT7 gate hole #2564, the legacy-contract backfill #2553/#2323, the quarantine debt #2295/#2309/#2342). Degodding src does nothing for these — the test is bad independent of the code it exercises. Do not sequence structural work ahead of this.

  2. Structure-induced friction — tests that are brittle because the unit-under-test has no clean seam: heavy fixtures, deep mock stacks, and co-change with src on every refactor. Here degod-first pays off twice. Once the god-module is split into ports + a pure core (the #2173 infra-logic pattern, delivered by the #1797 degod waves), those tests collapse to small mock-free units. Rewriting them before the seam exists means rewriting them again after — you pay the de-scaffold cost twice.

The discriminator: CaaCS co-change

Do not guess which friction is which. The test-change-coupling (CaaCS) analysis ranks src ↔ test co-change; its high-co-change clusters are exactly the structure-induced friction where degod-first helps. Before committing the QA mission's scope, read that ranking:

  • If the fragile #2071 tests cluster on the known god-surfaces (cli/commands/ doctor.py #2059, cli/commands/merge.py #2057/#2026, agent/mission.py #2056, charter/context.py #2532, next_step #2603), that overlap is the enabler list.
  • If they are spread across otherwise-clean units, structural work will not move the needle and the QA mission should just fix the tests directly.

(The cli/commands/tasks.py crime-scene — the original highest-co-change surface, #2034/#2116 — has already been degodded, PR #2308, 4569→1206 LOC. That is the proof-of-mechanism: its tests got a golden-CLI seam first, then thinned.)

ROI ranking for this goal

Cleanup Effect on the QA mission Verdict
Dead-code removal Deletes the code and its tests outright — zero rewrite, near-zero risk First. Cheapest friction reduction available.
Deshim Hardening tests around a scheduled-dead surface is wasted work; delete the shim + its tests now First. Tidy-first: shim deletions in the current cycle.
Targeted degod of surfaces the QA mission will rewrite Creates the seam so the test rewrite is durable, not double-work Interleave, gated on the CaaCS overlap; scope to only those surfaces.
Broad god-module degod (whole doctor.py / merge.py / mission.py) Real but long, merge-serialized; the #1797 / #2173 program already owns it Do not gate the QA mission on it — separate track.

Concrete issue clusters (tracker, 2026-07-13)

Dead-code / deshim enablers — cheapest, do first (tidy-up / tech-debt):

  • #2463 — drop pre-3.2.x legacy mission support (empty-mid8 / dual-era bare-slug branches). Dead-code removal; deletes its dual-era tests.
  • #2293 — unshim: burn down category_b_grandfathered_legacy dead-symbol carry-over (237 symbols).
  • #2499 — consolidate compat/registry.py shim registry into the Contract Registry loader.
  • #2561 — runtime_bridge compat-delegate surface (repoint monkeypatches, retire the delegate).
  • #2559 — dead-code gate is blind to first-party module.attr dynamic access. This is tooling that makes the deletions above provably safe — a force-multiplier for the whole dead-code sweep.

Targeted degod enablers — interleave per CaaCS overlap (tidy-up / tech-debt):

  • #2603 de-god next_step (remove # noqa: C901), #2604 reduce _mt_commit_wp_file complexity, #2532 decompose charter/context.py, #2059 decompose cli/commands/doctor.py, #2057/#2026 decompose cli/commands/merge.py, #2056 decompose agent/mission.py, #2465 workflow.py resolver consolidation, #2560 runtime_bridge strangler slice, #2595/#2600 extractions.

Test-intrinsic (the QA mission's own core — degod does NOT help):

  • #2071 epic; #2074/#2075/#2076 (CT3/4/5); #2564 (CT7 gate hole); #2553/#2323 (legacy-contract backfill); #2295/#2309/#2342 (quarantine).
  1. A small dead-code + deshim sweep first (its own quick mission or a WP-0 cluster) — #2463, #2293, #2499, #2561, plus the #2559 tooling gate. It shrinks the QA mission's surface before it starts, at near-zero risk, because deleting code deletes its tests.
  2. Fold small, local structural cleanup INTO the QA mission as a campsite-first WP cluster, scoped to exactly the surfaces it rewrites — the CaaCS-implicated god-surfaces from §"discriminator". Route any god-module too big for a campsite step to its tracked degod issue (#2059/#2057/#2056/ #2532); do not inline the full decomposition.
  3. Do not block #2071 behind the full #1797 / #2173 program. Pin the tests on genuinely-clean-but-badly-written units and fix them directly.

Relationship to the 3.2.x spine

This refines, it does not contradict, the milestone roadmap dependency spine, where #1797 (degod/unshim DELIVERY) and #2071 (test-QA friction) already sit as peer blockers of the #1619 root. The spine says both must land before #1619; this note adds the intra-pair sequencing the spine leaves implicit: they are not merely parallel — a targeted subset of #1797 (the CaaCS-implicated god-surfaces + the dead-code/deshim sweep) is a cheap enabler of #2071, while the bulk of #1797 stays on its own track. Concretely: the dead-code/deshim sweep leads, the CaaCS-overlap degod interleaves as campsite work, and the remaining god-module decompositions proceed independently under #1797 / #2173.

Edit this page