# Decision log — WP-AI-0001

All entries are dated; "before freeze" refers to the confirmatory freeze (PREREGISTRATION v1.0, not yet done).

• D1 (2026-11-09, before freeze). Project located at Recherche-UQO/ia/wp-ai-0001 with git remote gitsrv:srv/git/wpai0001_uqo.git, following the FIRC-0001 precedent for non-numeric IDs; UQO paper template (_modele-working-paper) adopted; English throughout.
• D2 (2026-11-09, before freeze). OPSEQ output format changed from single final answer to a scratchpad trace after the single-token variant failed to generalize (train loss → 0, dev ≈ 0.24–0.34 on length ≤ 3; ≈ 0.13 on length ≤ 5). Consequence: the Tier-0 target capability is "execute Ω in context", a per-step map, and compositional difficulty is controlled by trace_stride. Risk R1 recorded.
• D3 (2026-11-09, before freeze). Micro-world enlarged from 6 operators / length ≤ 4 (≈ 14 k unique items, too small for 40 k training items) to 8 operators / length ≤ 5 (≈ 400 k).
• D4 (2026-11-09, before freeze). Toy models are PyTorch (hooks trivially exact, CPU-testable); MLX is reserved for Tier ≥ 1 with a layer-wrapping intervention mechanism and an equivalence test.
• D5 (2026-11-09, before freeze). The pilot is exploratory; localization selection uses the pilot's development localization set; confirmatory evaluation splits will be generated from new seeds after freeze.
• D6 (2026-11-09, before freeze). Localization sweep families in the pilot were resid_post, attn_out, mlp_out (plus heads). The embed/resid_pre.0 family was not swept, which is why patching at the Ω token restores nothing; it is added to the next sweep. The pilot's localization set (length-3 programs, Ω at index 1, corrupt = zib) is a development set; confirmatory localization will use new seeds and other corrupt operators.
• D7 (2026-11-09, before freeze). Exploratory run graft_tier0_f50a28ca1bf7 showed that affine maps fitted to predict the recipient's activations on Ω-containing rows (ridge, CCA) remove the Ω-specific component (the recipient has none to predict), while orthogonal maps (Procrustes) and maps fitted on task-free rows preserve it (Procrustes on task-free rows: 0.99 same-init / 0.96 different-init vs ridge-on-Ω-rows 0.17). The reference map for the secondary families (steering, subspace, adapter, H5, mismatch, dose, layer grid) is therefore Procrustes fitted on task-free alignment rows; ridge variants are kept as comparisons. Protocol tag v2-unrelated-procrustes-reference; the v1 run is retained in the registry.
• D8 (2026-11-09, before freeze). Pairs are classified by initialization: donor_si and recipient_si share the initialization seed (identical initial weights, different data); off-diagonal pairs do not. Reported separately because the identity graft works only for same-initialization pairs (0.87 vs 0.10).
• D9 (2026-11-09, before freeze). Tier-1 alignment prompts are format-matched and task-free (Input:/Output: frames with random content, D9) after the quick run showed that sentence-final states are off the manifold of Output:-position states; held-out R² is now computed on rows never used for fitting; a donor-in-the-loop replace graft (donor few-shot state mapped into the recipient) and its controls (donor zero-shot state, shuffled query, sham map) are added; a declared norm-rescaled variant of the mapped vector is reported.
• D10 (2026-11-09, FREEZE). PREREGISTRATION v1.0 (Tier 0) frozen after graft_tier0_30e3644aa801; primary method = activation graft with task-free Procrustes at L0→L0, α = 1; confirmatory seed base 90 000. SHA-256 of docs/PREREGISTRATION.md at freeze recorded below. PREREGISTRATION.md sha256 at freeze: fe06d326bbbd27916d573a617e94d4443546f179587209e2e04a6893aad25d3e
• D11 (2026-11-09, FREEZE Tier 1). Selection rule for the Tier-1/2 confirmatory contrast recorded in PREREGISTRATION before the development results were read (eligibility = own-vector sufficiency ≥ 0.5; selected condition = best dev condition; contrast = selected minus best control on confirmatory queries). PREREGISTRATION.md sha256 at Tier-1 freeze: 85c5db69657144e6d6afecfb4fc346c2345d09455fb6134bee615567e5c705ad
• D12 (2026-11-09, amendment A1 before the Tier-1 confirmatory run). Shuffled-query donor state reclassified from null control to mechanism diagnostic (it carries the task identity); both analyses reported.
• D13 (2026-11-09, amendment A2). Tier-1/2 confirmatory runs load the development localization (layers, vectors) instead of re-localizing on confirmatory queries; first confirmatory run invalidated and re-run.