ACTIONIST program driver — reorientation directive

Date: 2026-08-28 From: Shaan, via CENA brain session Status: coordinator-owned, read-only for the driver. Supersedes the sequencing in DISPATCH.md and the R0–R12 ordering in the first-principles reset for the demo slice; everything not named here stands.

What stays

The 15-part map and 18-domain spine remain canon as the ownership grid. The seven-record family, the 22-rule solver, P13's intent-record edit model, and the refusal discipline all survive — they are the moat. The first-principles reset's north star ("beat Lovable for B2B vertical software by converting heterogeneous supply into evidence-backed reusable capabilities and deterministically assembling coherent, maintainable products") is confirmed. The AFFiNE/Teable/Chatwoot work is correctly classified as a calibration fixture; keep its holds exactly as recorded.

What changes

Stop horizontal reconciliation across all 15 parts. The active program is one vertical slice driving to a working demo, structured as two lanes:

Lane 1 — fast path (the demo)

Chat slot-filling spec (P02 grammar: EARS slots, unfilled slot = next question) + 4-up taste gallery (P06: 8–12 rounds, outside option) running in parallel → 6-slot config → instant render on the AutoSaaS base-template (SISO_Agency/apps/AutoSaaS/base-template/ — verified booting 2026-08-28 via renderBaseSaas(config); fix the missing pricingPlans default in app.js ~line 26 first) → token pack applied from the 16-pack taxonomy (research/ui-aesthetic-taxonomy-2026-08-27.md).

Tracked metric: time-to-first-render. Target: first render inside 2 minutes of conversation start. This metric does not currently exist anywhere in the program; it is now first-class. No admission gates in front of the first render — it is a preview, not a release. Lane 1 produces zero NOT_ADMITTED violations because nothing in it is admitted; it is explicitly pre-qualification stock.

Lane 2 — deep path (the product)

The existing correctness machinery (seven records, solver, qualification, holds, rollback objects) applies to real capability mounting behind the preview. Unchanged. The calibration fixture evidence lives here.

First dispatched work: shelf pre-baking

Select ~10 commodity capabilities from P03's shelf (crm, calendar_scheduling, forms, notes_docs, files_documents, messaging_notifications, approvals_workflow, analytics_bi, auth_identity, project_management — all 13 commodity kinds have selection-problem supply per P03), author their seven module records, and stage them as pre-qualified stock so the solver becomes a lookup at demo time, not a planner. This replaces R3–R7 sequencing for the demo slice; the R-gates continue to govern Lane 2.

Decision routing

OD-01, OD-02 and OD-03 in operator-decisions.md are reclassified as proposals, not resolved defaults. Consequential product choices — first wedge, pilot workflow, 21st.dev rights clearance, anything client-facing — route to Shaan and block until he answers. "Coordinator default" is not an authority for product decisions. Record the question, propose a recommendation, continue non-blocked work.

Known blockers — log, don't solve silently

1. Default node on PATH is broken (missing libsimdutf.34.dylib); use /opt/homebrew/opt/node@24/bin/node.
2. Mac mini Foundry index (the ~1.3M-repo corpus) confirmed existing, confirmed unreachable (WARP route down). Infra errand, not a research task.
3. The 21st harvest corpus contains no component source (compiled bundles only); assembly source comes from the template-shelf repos (research/template-shelf-sweep-2026-08-27/SHELF.md, 70 permissive JS/TS candidates) pending a rights decision from Shaan.
4. AutoSaaS build-pack visual gates cannot detect missing styling — never cite their green smoke results as visual evidence.

Evidence base for this directive

Three independent evidence reads completed 2026-08-28: engines (siso-ui-base + AutoSaaS runnability, commands verified live), experience plane (P05/P06/P07/P13 full read), assembly path (P02/P03/P12 + S2 convergence full read). Key synthesis: the research is a correctness pipeline, not a fast one; spec capture and solver elimination fit a 2-minute budget, the missing piece is pre-baked inventory, and the fast lane's render substrate already boots.