Signal for Shopify — Phase Binding and Workstream Zero

Status: Active roadmap binding (v1.0, 2026-08-11) · INTEGRATION PROMPT v2.3, F2 + F8.2 Authority split: the L5 Autonomous Media Architecture blueprint is authoritative on platform phases (B1–B5); docs/product/SIGNAL_SHOPIFY_MASTER_v4.md §3.3 is authoritative on Shopify product phases and exit criteria (P1–P4); docs/product/SIGNAL_OVERVIEW_v5.md §5 is the top-level three-clocks view. On conflict: escalate to the owner, never silently pick.

Binding rule (non-negotiable)

No product phase outruns its platform phase. P1 work may not begin ahead of B1 capability, P2 not ahead of B2, and so on. Certification artifacts are shared — covenants cite platform gates, never re-assert them.

B×P binding table

Platform phase (blueprint) Months Product phase Product scope (master authoritative) Product exit criterion
B1 Control foundation 0–4 P1 Foundation Shopify app, Canonical Event Envelope, NCM-v1, value feeds, L0–L1 Clean Shopify-vs-tracked reconciliation across 10 design partners; blueprint 30-day plan = P1 workstream zero
B2 Measurement & world model 4–9 P2 Measurement Holdout UX, ghost bids, cohort experiments, causal-credit reports, L2–L3 First auditable incrementality artifact per partner
B3 Bounded autonomy 9–14 P3 Yield Geo, lift-calibrated mini-MMM, cross-channel, miNCM, covenant GA Sustained miNCM accuracy over 2+ cycles at ≥3 T3 merchants
B4 Hardening & scale 14–19 P4 Scale Cold-start GA post-artifact, pooled priors (gated by docs/architecture/FLEET_INTEGRITY.md), App Store, Audiences interop Per master §3.3
B5 Certified L5 19–24 post-P4 Covenant GA per action class, reversible classes first ≥1 merchant running a certified L5 class a full quarter, zero breaches

Phase-name verbatim binding — verification status (owner decision 12)

D-12 RULED 2026-09-15: the five phase names above (B1 "Control foundation" … B5 "Certified L5") are the names of record by owner ruling 2026-09-15 (D-12) — owner ruling 2026-09-15, docs/reports/OWNER_RULINGS_2026-09-15_REGISTER_CLOSEOUT.md. The [Assumed — verify] marker is retired; the paragraph below is kept as the measurement history that led to the ruling.

The blueprint phase names above were carried from master §3.3, whose own label was: functional mapping [Validated by logic]; name bindings "Assumed — verify" (now closed by D-12). Measured 2026-08-11: the L5 Autonomous Media Architecture blueprint document itself was not located on the Drive MIZOKICloudRun board (searches for its title and distinctive phrases returned no standalone document), so the names could not be verified verbatim against the source. The flag stood and was escalated in SIGNAL_SHOPIFY_LANE_STATUS_2026-08-11.md: owner to drop the blueprint on the board or confirm the five names, closing decision 12. The owner confirmed the five names on 2026-09-15 (D-12); the mapping logic was never name-dependent and is unaffected.

Workstream zero — the P1 kickoff checklist (30-day plan ∪ P1)

Master §3.3 rules that the blueprint's 30-day plan is P1 workstream zero. The blueprint text is not on the board (above), so the checklist below is derived from the P1 definition in the master and the F6 kickoff scope in INTEGRATION PROMPT v2.3; items that the 30-day plan may add are marked. Owner decision 13 (30-day plan wholesale vs. triaged into P1) stays open.

P1 kickoff (F6) opens only after F1+F2 land, on a branch per repo conventions; it is deliberately not part of the F1–F5/F7/F8 report gate.

Stage 1–4 deployment benchmarks bound into P1/P2 exits (F8.2)

The per-customer deployment clock (overview §5, sourced from the capabilities documentation) nests inside P1–P2: a design partner's Stage 1–3 is exactly the P1→P2 exit-criteria path. Binding, per stage:

Deployment stage Benchmark Bound into
Stage 1 Sense & See (wks 0–6) ≥80% identity/attribution coverage P1 exit precondition per partner — a partner below 80% coverage does not count toward the 10-partner reconciliation exit
Stage 2 Score & Explain (6–12) Model promotion only at Brier ≤ 0.20 / AUC ≥ 0.72 / ≥2 purchase cycles (docs/architecture/DEL_AUTHORIZATION_FUNCTION.md) P1→P2 boundary — intent surfaces stay advisory below the bar
Stage 3 Prove (8–16) Refutation-passing iROAS with CI clearing the hurdle P2 exit — this is the "first auditable incrementality artifact"
Stage 4 Act & Compound (Q2+) Ledger-driven reallocation under 2-cycle revert rules P2→P3 boundary — no reallocation authority without the Stage-3 artifact

Owner decision 17 (open, flagged for the status report): whether Stage 1–4 benchmarks are contractual SLAs or internal targets. Until decided, every stage benchmark above is an internal target and must not appear in merchant-facing contract language.

Current build state (corrected 2026-08-12 — the previous "F6 not opened" line went stale the hour it was written): F6 is opened; workstream zero landed the NCM-v1 contract (ncm_v1.py), the L1 value-feed gate (ncm_value_feed.py), the COGS worksheet, and the follow-on session landed the validated COGS import (cogs_import.py) and the Shopify-vs-tracked reconciliation harness (reconciliation.py) — live P1 state and remaining steps: docs/roadmap/P1_BUILD_PLAN.md. The gateway boundary hardening is deployed with zero live traffic (operator secrets, register item 6), and no merchant attestation exists yet — the harness awaits real totals. The LII backtest is still short of the 0.72 promotion gate, but as of 2026-08-16 that shortfall is not a model verdict: 0.6884 was the offline fallback scorer on one seed, and the fixture's oracle ceiling (0.7277 mean) puts 0.72 out of reach for ANY scorer on synthetic data — real forward labels are the blocker (register item 17 → 6c/23; docs/reports/LII_BACKTEST_MODEL_QUALITY_2026-08-16.md). The Stage-2 gate is exactly what this is for.

← All docsView source on GitHub →