Paula Patterns — 3.2.x Goal Corroboration from the Emerging-Patterns POV

I am Paula Patterns. I review recurring boundary leaks, ownership confusion, and whack-a-field fixes by naming the shape of the recurring pattern and whether the codebase's pattern vocabulary is actually moving toward the stated goals — not by celebrating intent. I separate the patterns that genuinely spread in the v3.1.10..v3.2.0 delta from the ones that are pre-existing scaffolding, and I flag where the delta shows the opposite of the goal (new shadow paths, duplication added).

Directives applied: D-001 (Architectural Integrity — find the owning boundary a pattern serves before crediting it to a goal), D-003 (Decision Documentation — record the corroboration verdict so it is not relitigated), D-030 (Quality Gate — treat ratchets as the test surface proving a failure-class is closed, and count them as evidence), D-032 (Conceptual Alignment — a recurring name/idiom is the signal that concept drift is or is not being strangled). Tactics applied: anti-corruption-layer (legacy shape kept as a warned compatibility branch, not a co-equal path), review-intent-and-risk- first (does the spreading pattern close the observed class, what is the blast radius), domain-event-capture (canonical event log as the durable SSOT). Builds on my prior naming-identity-ssot-strangler/paula-patterns-duplication-shapes.md.

Scope: v3.1.10..v3.2.0, 2317 commits. Method: targeted git log --grep/-S/-G pickaxe + git ls-tree/git grep token-counts at each tag boundary (NOT a full-diff read). All counts are src/**/*.py unless noted.

TL;DR — the central finding (read this first)

The pattern vocabulary of the codebase shifts hard toward the goals across this range — but the headline evidence is the enforcement layer, not the adoption layer. The biggest, most unambiguous move is architectural ratchets: 3 → 60 test files (20×). The contract patterns (frozen value-objects, fail-closed authorities, compose+parse seams, pure resolvers) all spread 2–3×. But the adoption of those SSOTs lags the construction of them, and in two specific idioms the duplication-class grew in absolute terms — exactly the "seam built, consumers still re-derive inline" gap the 3.2.x research already named. So the emerging patterns corroborate the goal direction strongly, and corroborate goal completion weakly — which is precisely what a "stabilization-and- depth cycle" that "stays open until they hold" should look like one release in.

The dominant emerging pattern is the architectural ratchet (literal-ban / parity / boundary AST-scan) — it is the connective tissue that makes every other pattern stay strangled, and its 20× growth is the single clearest signal that "strangle/route, don't duplicate" has moved from aspiration to enforced practice.

Pattern catalog

Pattern new / spreading first-appearance commit maps-to-goal spread metric (v3.1.10 → v3.2.0)
Architectural ratchet (literal-ban, parity, boundary AST-scan) spreading, explosively pre-existing 3 files; mass landing fa80fa0f9 "ratchet burn-down", b7bf5f9e3 run-twice ratchet, 8dad4de75 canonical-producer lint G3 (primary), enforces G2 tests/architectural/: 3 → 60 files (57 net-new incl. test_no_op_stable_writes, test_no_raw_mission_spec_paths, test_no_primary_anchored_gates, test_topology_resolution_boundary, test_status_module_boundary, test_no_dead_symbols)
Fail-closed authority (*_required, …Unresolved/NotFound/Invalid exceptions) spreading many; 0ca389814 "fail closed on coord topology gaps", 76bb71557, 348ef90db "fail-closed unknown relations" G2 (SSOT refuses to guess), G1 (doctrine refuses malformed pack) fail-closed exception classes 12 → 27 (2.25×); *_required funcs 15 → 23
Frozen-dataclass immutability (value-object compute-once) spreading b02d35c1b "PackContext three-state frozen dataclass" G2 (immutable SSOT carriers), G1 (frozen PackContext/template) @dataclass(frozen=True): 104 → 306 (61 → 159 files, ~3×); __post_init__ 14 → 36
Compute-once-thread context FRAGMENT (IdentityFragment/BranchRefFragment/WorkspaceFragment/StatusSurfaceFragment) NEW value-objects on a pre-existing surface c5a10ce56 (#1793) — Fragment classes net-new; mission_runtime/ dir + runtime_bridge.py pre-exist G2 (the naming/identity/read-path SSOT carriers) resolve_action_context calls 3 → 27 (2 → 9 files); ExecutionContext refs 57 → 82 (11 → 19 files)
Compose+parse SSOT seam (lanes/branch_naming.py: mission_dir_name/worktree_path/coord_*/mission_branch_name_required/lane_branch_name) pre-existing, hardened+routed module + primitives + test_no_worktree_name_guess.py ratchet all in v3.1.10; hardened fcf9be595 (#2001) G2 (identity/naming SSOT) seam primitives already named at v3.1.10; mid8 refs 67 → 605 (16 → 54 files) — the grammar spread across the tree
Pure resolver / resolve_* seam (testable extraction) spreading 43fee0bda canonical kind resolver, fe2c5695d pure plan_activation seam, 0f634d8a8 is_planning_lane seam G2 + G3 (mocks ↓) distinct def resolve_*: 35 → 79 (2.25×)
OHS facade / shared-kernel re-export (canonical authority behind a thin facade) spreading c5a10ce56 status-facade strangle, fa0a02be6 template_catalog facade (allowlist→0), b7bf5f9e3 shared kernel comparison util G2 (route consumers to one authority) facade refs 6 → 49 (3 → 29 files, ~8×)
Declared-identity-keying ("name proposes, authority disposes": resolve_mid8, meta.json governs) spreading seam resolve_mid8 pre-exists; selector-disambiguation hardened fcf9be595 G2 (identity SSOT) covered by mid8 67 → 605 and the #1888 existence-check (8b64bb2f1 ownership seam)
Strangler-fig parallel→route→delete (collapse a duplicate onto an authority) spreading, with real collapses e14db3227 byte-identical do/ask/advise aliases (#1810/#1804), 910ea1a1c collapse from_frontmatter dual branches, a8fad6ae9 registry-back VALID_AGENTS, ba18e1d4d route remaining lane checks through one seam G2 (primary) 7 commits with strangl* in title; multiple explicit collapse/byte-identical/route … through … seam commits (see Discipline §)
Doctrine→runtime step binding (_resolve_step_agent_profile/_resolve_step_binding thread profile+contract_ref into the next step) NEW c5a10ce56-era runtime_bridge hardening G1 (the only direct doctrine-governs-execution seam in-range) step-binding refs 0 → 9; runtime/next doctrine refs 0 → 14 (0 → 5 files)

Patterns → goals map (does the vocabulary shift toward the goals?)

Yes — measurably, and in the goal-ordered way the release doc predicts.

  • G1 (doctrine governs runtime): the thinnest but net-new evidence. runtime_bridge gained _resolve_step_agent_profile/_resolve_step_binding (0 → 9 refs) so a step's resolved agent_profile/contract_ref from the frozen mission-type template now decides step execution; runtime/next doctrine references went 0 → 14. The frozen PackContext/template + fail-closed-on-malformed-pack patterns (bc04abae7, 348ef90db) are the upstream half — doctrine that refuses to resolve when malformed. This is doctrine starting to gate, not just resolve — but it is a seam, not yet pervasive. Vocabulary present, spread shallow.
  • G2 (strangle core domains onto SSOTs): the dominant beneficiary. Every spreading pattern in the catalog except the G1 binding serves G2: frozen value-objects (×3), fail-closed authorities (×2.25), pure resolvers (×2.25), OHS facades (×8), the NEW mission_runtime Fragment carriers, and the mid8 grammar spreading 67 → 605. The new mission_runtime/context.py Fragments are the literal "take identity/naming/ read-path onto its canonical SSOT" instruction in code. Vocabulary shift is decisive.
  • G3 (DevEx & enablers): the clearest single number in the whole delta — ratchets 3 → 60. Pure-seam extractions (resolve_* 35 → 79; 8b64bb2f1/fe2c5695d explicitly "extract … seam for stub-based testing") lower test cost; the ratchet wall makes the strangled classes stay strangled. The docs/release-goals/ + tests/architectural/ README.md governance surface also landed in-range. Vocabulary shift is decisive.

Verdict on the claim "emerging patterns point toward the goals": the codebase's pattern vocabulary does shift toward all three goals across the range, ordered G3≈G2 (loud) > G1 (quiet-but-real) — consistent with a cycle that front-loads enablers/SSOTs and defers deep doctrine-into-runtime to later 3.2.x slices.

The discipline signal — "strangle/route, don't duplicate; no shadow paths"

Emerging practice, not aspirational — but unevenly enforced. Evidence the discipline is being practiced and ratcheted, not just stated:

Strangle-then-collapse instances (the right shape):

  • e14db3227 — do/ask/advise collapsed to byte-identical aliases over one canonical dispatch (#1810/#1804). Textbook route-then-collapse.
  • 910ea1a1c — "collapse from_frontmatter dual branches into one path."
  • a8fad6ae9/e7de69d76 — SKILL_ONLY_AGENTS/VALID_AGENTS registry-backed (delete the hardcoded twin).
  • 0f634d8a8/ba18e1d4d — "single planning-lane classifier seam" + "route remaining lane checks through is_planning_lane" — extract one classifier, route consumers, delete re-derivations (emit-don't-guess for predicates).
  • c5a10ce56 — status-facade strangle + canonical mission_runtime surface.
  • fa0a02be6 — template_catalog facade with boundary allowlist driven to 0.
  • 8b64bb2f1 — extract build_wp_manifests seam for stub-based testing (DevEx-driven strangle).

Enforcement that the discipline is now ratcheted (D-030): test_no_worktree_name_guess (literal-ban with shrinking allow-list), test_no_raw_mission_spec_paths, test_no_primary_anchored_gates, test_no_dead_symbols/test_no_dead_modules, test_no_op_stable_writes (run-twice no-op ratchet, b7bf5f9e3), test_topology_resolution_boundary, test_status_module_boundary, test_compat_shims + test_shim_registry_schema (every legacy shim must be registered, i.e. a warned compatibility branch, not a silent parallel path). The fa80fa0f9 "ratchet burn-down" is the discipline made executable. This is the anti-corruption-layer tactic operationalized: legacy is kept as a declared, tested compat branch, never a co-equal resolver.

Conclusion: the discipline is an emerging, enforced practice — the 20× ratchet growth and the explicit collapse commits are corroborating evidence, not vibes.

Anti-corroboration — where the delta shows the OPPOSITE (challenging the claim)

Honesty per review-intent-and-risk-first. Three real counter-signals:

  1. The adoption-gap duplication GREW in absolute terms. The very class the 3.2.x research flags as "seam built, ~5% adopted" is visibly worse by raw count:

    • bare mission_id[:8] / _id[:8] inline: 6 → 27 (4 → 13 files). The seam (resolve_mid8/mid8()) exists and spread 67 → 605, yet more consumers hand-roll the slice. This is the literal "consumers re-derive mid8 while holding a context that carries it" anti-pattern — and it increased. The ratchet for bare …_id[:8] repo-wide does not yet exist (it is a 3.2.1 deliverable), so the recurrence is unguarded.
    • parents[2] project-root re-derivation: 4 → 10 (4 → 9 files). The paths authority exists; the depth-coupled re-derivation idiom doubled. No ratchet bans it yet (Shape D in my prior note — still the headline missing guard).
    • Mitigant: the naming f-string composes (f"…kitty/mission…") actually shrank 7 → 5, so the static-write end of the same shape is moving the right way. The growth is concentrated in the inline-read idioms that lack a ratchet — confirming the pattern, not refuting it: what is ratcheted shrinks; what is not, grows.

  2. "Mirrors X" docstrings exploded 32 → 157 (18 → 93 files, ~5×). On its face this is the classic un-extracted-seam tell ("Mirrors the _…_feature_dir pattern"). Inspection largely exonerates it: the bulk are deliberate contract/schema mirrors and facade re-exports (charter/* "Mirrors the vocabulary pinned in…", "mirror spec_kitty_events schemas", "Mirrors the field-merge semantics") — i.e. documented OHS boundaries, not raw logic dup. But a residual minority are the bad shape (the implement.py _lanes_feature_dir "Mirrors _status_feature_dir" family from my prior note). So the 5× growth is ~80% benign facade-documentation, ~20% genuine un-extracted-mirror debt — the SSOT vocabulary is partly being used to paper over mirrors rather than collapse them. Watch this number: if facade-mirrors keep growing faster than collapses, the "no shadow paths" rule is being satisfied in docstring but not in structure.

  3. G1 is corroborated only as a seam, not a trajectory. runtime/next doctrine refs 0 → 14 and step-binding 0 → 9 is real but small against 2317 commits. The release doc's own G1 success criterion ("a test proving the directive/contract changes behaviour") is not yet met in-range — the binding reads the profile/contract; I found no in-range ratchet proving a doctrine change flips a runtime decision. G1 is the weakest-corroborated goal: the vocabulary appeared, the governs-execution proof did not.

Verdict per goal

Goal Corroboration Basis
G1 — doctrine governs runtime PARTIALLY supports NEW step→profile/contract binding in runtime_bridge (0 → 9) + runtime/next doctrine 0 → 14 + fail-closed-on-malformed-pack establish the seam; but spread is shallow and the "doctrine changes behaviour" proof/ratchet is absent in-range. Direction yes, trajectory unproven.
G2 — strangle core domains onto SSOTs SUPPORTS Decisive vocabulary shift: NEW mission_runtime Fragments, frozen value-objects ×3, fail-closed authorities ×2.25, pure resolvers ×2.25, facades ×8, mid8 grammar 67 → 605, multiple explicit collapse/byte-identical commits. Caveat: construction outran adoption — inline _id[:8] 6 → 27 and parents[2] 4 → 10 show the strangle is mid-flight, not done.
G3 — DevEx & enablers SUPPORTS (strongest) Ratchets 3 → 60 is the loudest single signal in the delta; resolve_* seam extractions 35 → 79 explicitly "for stub-based testing"; run-twice/no-op + boundary + dead-symbol ratchets + docs/release-goals/ governance surface all landed in-range. The enablers that make G1/G2 stick are visibly in place.

Overall: the operator's claim — the 3.2.0 patterns already point toward the 3.2.x goals — is CORROBORATED for direction, with the honest caveat that G2 adoption and G1 depth are early. The codebase didn't just declare the goals; it grew a 20× ratchet wall and a frozen-SSOT vocabulary to enforce them. The one trajectory to watch (not yet a refutation) is the un-ratcheted inline-read duplication growing while the seam that should absorb it sits at ~5% adoption — which is exactly the 3.2.1 work, and exactly why the cycle is correctly left open.

Edit this page