Phase 0 Research — Write-Side Context-Factory Adoption (Mission B)

Basis documents (normative Phase-0 evidence, verified on HEAD efb28158f):

artifact family (C-007), the Q3 pivotal verdict, the C-003 sizing, the #2016 fold, the 7-WP partition.

confirmation, reduction LOC, verification-by-deletion list, fragment-retirement targets, and the idempotency-divergence finding.

kitty-specs/read-path-error-fidelity-adoption-01KV8NPC/research/investigation-3-readwrite/.

  • research/write-site-inventory.md — architect-alphonso: the exact write-site inventory classified by
  • research/reduction-census.md — randy-reducer: the re-verified write-path census, 0→load-bearing
  • Carried from Mission A: docs/engineering_notes/context-factory-readwrite-symmetry/00-SYNTHESIS.md +

Resolved questions

1. Q3 — do the write-half fragment fields exist + resolve correctly? YES (pure consumer-routing). workspace.primary_root (context.py:144/resolution.py:698), status_surface.status_write_dir (context.py:162/resolution.py:724-728 → resolve_status_surface, the coord-aware fail-closed authority — routes to status/coord, NOT primary_root, C-007-correct), branch_ref.destination_ref (context.py:130/resolution.py:705-722). The factory needs no completion; the only mission_runtime/ touch is FR-006 deletion. C-001 honored.

2. C-003 scope → Option A (bounded), ~90–130 LOC. R1–R5 root walks + P1 placement + S1 surface + FR-006 retirement. The S2 write-surface-SELECTION ladder (the #1716 ~2094-LOC authority root) computes the same value the factory already does → cleanly deferrable (reduction-not-symmetry).

3. FR-004 (write-target) DEFERS — the one idempotency divergence (D-2). The flattened-arm destination_ref (= target_branch, CWD-invariant) vs the inline _current_branch(repo_root) (= git HEAD) diverge on off-target-branch topology — a latent-bug-fix that CHANGES the on-disk write target, violating NFR-004 for the bounded cut. Deferred to the #1716 slice with before/after verification. All bounded-cut sites resolve a root/anchor (flip no destination) → NFR-004 clean.

4. #2016 → cross-ref, not fold. Read-path bootstrap already fixed by Mission A WP09 (d4f0cf581).

5. 0→load-bearing (SC-002): all four targets (primary_root, status_write_dir, destination_ref, prompt_source) have zero live consumers today (grep-proven) — adoption flips them load-bearing; prompt_source + the dead surface= param are retired (FR-006).

Decisions of record

is the precondition for the deferred FR-004 flip.

  • D-1 (C-003 scope, decision 01KV9WY760JEXVEQ4KXF7F2VSB): Option A bounded; defer S2/#1716.
  • D-2 (FR-004 idempotency defer): write-target → #1716 slice (NFR-004 binding).
  • D-3 (#2016): cross-ref (Mission A WP09).
  • D-4 (W9/W10): route both to primary_root = the consolidation.
  • D-5 (NFR-001 gate): topology-true root+surface equivalence test in the bounded cut; target equivalence

Testing approach (binding)

Function-over-form + verification-by-deletion (delete the inline re-derivations; green suite + the previously-0-reader fragments now exercised = the proof). TDD-first. Topology-true fixtures (full 26-char ULID, real coord-worktree + real submodule). Idempotency-preserving (NFR-004 — before/after on-disk topology identical for the bounded cut). ruff/mypy clean, ≤15, no suppressions.