@lifeaitools/rdc-skills 0.24.42 → 0.25.1

package/skills/onramp/SKILL.md

@@ -0,0 +1,1459 @@
+---
+name: rdc:onramp
+description: "Usage `rdc:onramp <slug> --name \"<Display Name>\"` — Full-pipeline enrollment orchestrator for a new Place Fund project. Runs all 7 phases: enroll → research + document integration → conflict resolution → brand book + imagery → Regen Score gate → site build → deploy dev. Each phase is idempotent and resumable."
+---
+
+> **⚠️ OUTPUT CONTRACT (READ FIRST):** `guides/output-contract.md`
+> Checklist-only output. No tool-call narration. No raw MCP/JSON/log dumps.
+> One checklist upfront, updated in place, shown again at end with a 1-line verdict.
+
+> If dispatching subagents or running as a subagent: read `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md` first, then `{PROJECT_ROOT}/.rdc/guides/engineering-behavior.md`.
+
+> **Sandbox contract:** This skill honors `RDC_TEST=1` per `guides/agent-bootstrap.md` § RDC_TEST Sandbox Contract. DB writes, deploys, and web searches short-circuit under the flag. Drift resolution, scoring, and file validation run normally.
+
+> **RPMS compliance:** Encodes the Four Laws (One Home Per Fact · One Direction · One Door · One Spine). Each fact is written to its one home. **⛔ NEVER write `places/<slug>/_context.md`** — it is a generated projection owned by `apps/studio/src/lib/context-export-writer.ts`. A post-scaffold gate asserts its absence; the skill halts if found.
+
+# rdc:onramp — Full Pipeline Enrollment
+
+## When to Use
+
+- Enrolling a new Place Fund project end-to-end (enrollment through dev deploy)
+- Resuming an enrollment that stopped at any phase (idempotent re-invoke)
+- Running research + scoring + build for a place already enrolled
+
+**Prod promotion is NOT part of onramp.** Once Phase 7 completes, the pipeline
+is done. To promote to production later: `rdc:deploy <slug> promote`.
+
+## Argument Contract
+
+```
+rdc:onramp <slug> --name "<Display Name>" [--location <json>] [--archetype <archetype>]
+                  [--owner <place-fund|rdc|jv|client>] [--history] [--dry-run]
+                  [--skip-to <phase>] [--no-gate]
+```
+
+| Arg / Flag | Required | Default | Description |
+|------------|----------|---------|-------------|
+| `slug` | **yes** | — | kebab-case, place-anchored. Validated against `^[a-z0-9]+(-[a-z0-9]+)*$`. |
+| `--name` | **yes** | — | Human display name. |
+| `--location` | no | `{}` | JSON object for `enroll_place p_location`. Never guessed. |
+| `--archetype` | **yes** (fresh) | — | Required on fresh enrollment — reject if absent. See `corpus/_shared/onramp/ARCHETYPES.md`. Valid: `land-regen`, `data-center`, `urban-renewal`, or custom slug. |
+| `--owner` | no | `place-fund` | `place-fund` \| `rdc` \| `jv` \| `client` |
+| `--history` | no | off | Scaffold `places/<slug>/HISTORY.md`. |
+| `--dry-run` | no | off | Resolve drift + print plan; no writes. |
+| `--skip-to` | no | — | Jump to a specific phase (1–7). |
+| `--no-gate` | no | off | Skip human gates (brand review). |
+
+## Master Checklist
+
+```
+rdc:onramp <slug> — full pipeline
+Phase 1: Enrollment + RPMS Infrastructure Setup
+  [ ] Preflight — validate slug + --name + --archetype (required on fresh); parse flags
+  [ ] Resolve drift — classify: fresh | already-enrolled | spine-ahead | disk-ahead
+  [ ] DB spine — enroll_place RPC → capture project_node_id
+  [ ] DB spine completeness — project_nodes, project_places, prt_projects, regen_projects, place_phase_state
+  [ ] Archetype stored — places.metadata.archetype set (not TBD)
+  [ ] Enrollment epic — dup-guard + insert_work_item
+  [ ] Disk tree — scaffold places/<slug>/ with ALL required files
+  [ ] Directory verification — PLACE.md, PRODUCT.md, DESIGN.md, HANDOFF.md, RECONCILIATION.md, SCREENING.md, PROCESS-ARC.md, INTAKE-LOG.md, ONRAMP-REPORT.md, corpus/INDEX.md, arc/01-06, tracker/*
+  [ ] Corpus setup — places.corpus_path set, $CORPUS_ROOT dir exists
+  [ ] Process-state arc selection — PROCESS-ARC.md written (25 states, per-archetype)
+  [ ] Screening stub — SCREENING.md written
+  [ ] _context.md absence gate
+  [ ] Phase 1 rubric gate — 15-row pass/fail table printed + ONRAMP-REPORT.md appended
+Phase 2: Research — Full Regenerative Regional Deep Research
+  [ ] Research prompts — load archetype template from corpus/_shared/onramp/research-prompts/
+  [ ] Web research — structured per Five Capitals (Natural, Human, Social, Built, Financial)
+  [ ] Read ALL incoming corpus documents ($CORPUS_ROOT/<corpus_path>/)
+  [ ] Financial model — extract from docs OR construct from analysis (Tier 4, tagged)
+  [ ] Integrate web + document data into 6 arc files (research layer)
+  [ ] PRODUCT.md route registry — archetype-specific website routes (NOT fixed 6-arc)
+  [ ] PLACE.md compiled from findings
+  [ ] corpus/INDEX.md — all sources with tier + capital + arc file
+  [ ] DOCUMENT GATE: every incoming doc read — no unread documents allowed
+  [ ] Phase 2 rubric gate — 14-row pass/fail + ONRAMP-REPORT.md appended
+Phase 2b: People, Organizations & Place Enrichment
+  [ ] Extract all named people + organizations from project documents
+  [ ] RocketReach lookup on each person + org — verify existence
+  [ ] Web research on orgs — website, news, track record, regional presence
+  [ ] Government registry verification for key entities
+  [ ] Write verified entities to tracker/STAKEHOLDERS.md (>20 lines floor)
+  [ ] Flag unverified entities in RECONCILIATION.md
+Phase 3: Conflict Resolution
+  [ ] Build fact inventory — extract every factual claim from all 6 arc files
+  [ ] Cross-source verification — web vs document claims explicitly compared
+  [ ] Cross-document verification — internal document consistency checked
+  [ ] Financial figure verification — arc/05 + arc/06 internally consistent
+  [ ] Stakeholder claim verification — arc/03 matches STAKEHOLDERS.md
+  [ ] Timeline consistency check — dates coherent across all files
+  [ ] Geographic/legal claim verification — boundaries, zoning, rights checked
+  [ ] Resolve all conflicts — each claim verified | resolved | held
+  [ ] RECONCILIATION.md written — structured ledger, zero conflicted remaining
+  [ ] Arc files updated — resolved values written back with citations
+  [ ] corpus/INDEX.md updated — new sources + tier reassessments
+  [ ] Phase 3 rubric gate — 13-row pass/fail + ONRAMP-REPORT.md appended
+Phase 4: Brand Book + Imagery — Full rdc:design Integration
+  [ ] Invoke rdc:design with DESIGN.md brief + PRODUCT.md route registry
+  [ ] DESIGN.md — 24-spread editorial brand book (not a stub)
+  [ ] Palette — full Studio tokens (--<slug>-*) via RAMPA from ACTUAL place imagery
+  [ ] Palette source proof — image filename(s) logged in DESIGN.md
+  [ ] IMAGERY-PROMPTS.md — all slots declared; real search before AI gen
+  [ ] Generate images via regen-media MCP or OpenAI local gpt-image-2
+  [ ] Place images in apps/<slug>/public/images/ + wire imagery.ts
+  [ ] Brand review gate (skip if --no-gate)
+  [ ] Phase 4 rubric gate — 9-row pass/fail + ONRAMP-REPORT.md appended
+Phase 5: Regen Score — Model-Driven Rubric Assessment
+  [ ] Load model specification (VLAS 01 + 03 + archetype benchmarks)
+  [ ] Regional comparable research — iterative gap fill
+  [ ] SCARF×4Ps lattice assessment — 20 cells with evidence
+  [ ] Five Capitals per-capital scoring with evidence citation
+  [ ] Coverage spread gate — max-min <= 40
+  [ ] Annihilator check — 8 binary pass/fail (any BLOCKED → NO-GO)
+  [ ] Recognition formula readiness — 7 terms assessed
+  [ ] Process state completion audit — required states per archetype
+  [ ] Pathway ladder position — current + target rung
+  [ ] Honesty audit — Tier 4 tagged, precedent tiers set
+  [ ] Compose Regen Score from model rubric
+  [ ] Score gate decision — GO/NEEDS WORK/NO-GO
+  [ ] SCREENING.md written — lattice + score card + evidence
+  [ ] Phase 5 rubric gate — 15-row pass/fail + ONRAMP-REPORT.md appended
+Phase 6: Site Build — Archetype Routes + RegenOps Integration
+  [ ] Create apps/<slug>/ — archetype-specific routes from PRODUCT.md
+  [ ] HISTORY as its own route (sourced from HISTORY.md, not folded into arc)
+  [ ] All routes render real arc content + imagery (no placeholders)
+  [ ] Image/video slots from IMAGERY-PROMPTS.md have placeholder components
+  [ ] tsc clean, build clean
+  [ ] PUBLISH.md + DEPLOY block generated
+  [ ] RegenOps screens — report present/missing (advisory)
+  [ ] Phase 6 rubric gate — 10-row pass/fail + ONRAMP-REPORT.md appended
+Phase 7: Deploy Dev — Phase State + Epic Closure
+  [ ] Register in apps + app_deployments (allocate next free port)
+  [ ] Regenerate PM2 ecosystem config on Vultr (--write flag)
+  [ ] PM2 start + HTTP 200 at <slug>.dev.place.fund
+  [ ] place_phase_state advanced to phase 7, gate=passed
+  [ ] Epic: implementation report submitted, transitioned to review
+  [ ] INTAKE-LOG.md — all 7 phase rows present
+  [ ] ONRAMP-REPORT.md — all 7 phase sections present
+  [ ] Phase 7 rubric gate — 9-row pass/fail (final)
+```
+
+**Prod promotion is NOT part of onramp.** Once Phase 7 completes, the pipeline
+is done. To promote to production later: `rdc:deploy <slug> promote`.
+
+---
+
+## INTAKE-LOG.md — Transaction Log
+
+Every onramp run writes a timestamped transaction log at `places/<slug>/INTAKE-LOG.md`.
+Each phase records its timestamp, operator, session ID, and notes. Intake #2+ appends
+rows — the skill detects a prior intake and runs as an update (re-reads documents,
+re-scores, rebuilds changed content).
+
+```markdown
+| # | Phase | Timestamp | Operator | Session | Notes |
+|---|-------|-----------|----------|---------|-------|
+| 1 | Enrollment | 2026-07-20T07:11Z | claude-5 | abc123 | fresh state |
+| 1 | Research | 2026-07-20T07:20Z | claude-5 | abc123 | 14 web + 18 docs |
+| 1 | People/Org | ... | ... | ... | N verified |
+| 1 | Conflicts | ... | ... | ... | 2 resolved |
+| 1 | Brand | ... | ... | ... | palette derived |
+| 1 | Score | ... | ... | ... | composite 67 |
+| 1 | Build | ... | ... | ... | tsc 0, build 0 |
+| 1 | Deploy Dev | ... | ... | ... | HTTP 200 |
+```
+
+---
+
+## Phase 1: Enrollment + RPMS Infrastructure Setup
+
+### 1.0 Preflight
+
+Validate before any external call:
+- `slug` matches `^[a-z0-9]+(-[a-z0-9]+)*$` — reject if not
+- `--name` is present and non-empty — reject if not
+- `--archetype` is present and not `TBD` — **reject if not provided on fresh enrollment**
+  (valid values: `land-regen`, `data-center`, `urban-renewal`, or a custom slug
+  matching `^[a-z0-9]+(-[a-z0-9]+)*$`)
+- `--dry-run` set → print plan and exit after drift resolution + rubric, no writes
+
+### 1.1 Resolve Drift State
+
+Read-only DB probe:
+```sql
+SELECT id, slug, name, status, corpus_path, metadata FROM places WHERE slug = '<slug>';
+```
+
+Check disk:
+```bash
+test -d "places/<slug>/" && echo "disk:present" || echo "disk:absent"
+```
+
+| State | Detection | Action |
+|-------|-----------|--------|
+| **fresh** | No DB row AND no disk dir | Run all enrollment steps |
+| **already-enrolled** | DB row exists AND disk dir exists | Run verification rubric; fix gaps; continue to Phase 2. |
+| **spine-ahead** | DB row exists, disk dir ABSENT | Run scaffolder; then verification rubric |
+| **disk-ahead** | Disk dir exists, NO DB row | **⛔ STOP.** Open reconciliation work item; exit. |
+
+### 1.2 DB Spine (`enroll_place`)
+
+```sql
+SELECT enroll_place('<slug>', '<Display Name>', '<location_json>'::jsonb, 'enrolling');
+```
+`project_id = uvojezuorjgqzmhhgluu` REQUIRED on every Supabase MCP call.
+
+**GATE:** Capture `project_node_id` from the returned JSONB. If absent → BLOCKED template, stop.
+
+### 1.3 DB Spine Completeness (NEW)
+
+After `enroll_place`, verify ALL required DB rows exist — not just the `places` row:
+
+```sql
+-- project_nodes row
+SELECT id, slug, node_type FROM project_nodes WHERE slug = 'prtnode-<slug>';
+
+-- project_places join
+SELECT * FROM project_places WHERE place_id = '<places.id>'::uuid;
+
+-- prt_projects row (drives HISTORY.md gate)
+SELECT id, slug, project_type FROM prt_projects WHERE slug = '<slug>';
+
+-- regen_projects row (work_items FK)
+SELECT id, slug FROM regen_projects WHERE slug ILIKE '%<slug>%';
+
+-- place_phase_state
+SELECT * FROM place_phase_state WHERE place_slug = '<slug>';
+```
+
+**Fix missing rows:**
+- `prt_projects` absent → create via RPC or flag as a required manual step
+- `regen_projects` absent → create (needed for epic's `project_id` FK)
+- `place_phase_state` stuck at phase 0 with gate `open` on a re-run → leave as-is
+  (phase advancement happens at end of each subsequent phase)
+- Archetype → store in `places.metadata` as `{"archetype": "<value>"}`:
+  ```sql
+  UPDATE places SET metadata = jsonb_set(COALESCE(metadata,'{}'), '{archetype}', '"<archetype>"')
+  WHERE slug = '<slug>';
+  ```
+
+### 1.4 Enrollment Epic (`insert_work_item`)
+
+Dup-guard, then create if none exists:
+```sql
+SELECT insert_work_item(
+  p_title           := 'epic: enroll <slug>',
+  p_item_type       := 'epic',
+  p_priority        := 'high',
+  p_source          := 'onramp',
+  p_project_node_id := '<project_node_id>'::uuid,
+  p_labels          := ARRAY['onramp','<archetype>']
+);
+```
+
+### 1.5 Disk Tree + Directory Verification
+
+```bash
+node scripts/onramp-scaffold-place.mjs --slug <slug> --name "<Display Name>" \
+  [--archetype <archetype>] [--owner <owner>] [--history]
+```
+
+After scaffold, verify the **full RPMS-compliant tree** — not just "disk present":
+```
+places/<slug>/
+  PLACE.md              — authored identity (One Home)
+  PRODUCT.md            — route registry + model definition (archetype-specific routes)
+  DESIGN.md             — brand brief (input to rdc:design)
+  HANDOFF.md            — build contract
+  RECONCILIATION.md     — conflict ledger
+  SCREENING.md          — mission fit + RCL eligibility (NEW)
+  PROCESS-ARC.md        — 25-state selection matrix for this land type (NEW)
+  INTAKE-LOG.md         — transaction log (NEW)
+  ONRAMP-REPORT.md      — per-phase accumulating report (NEW)
+  corpus/INDEX.md       — source index
+  corpus/               — source documents
+  arc/01-story.md · 02-place.md · 03-foundation.md · 04-process-outcomes.md · 05-investors.md · 06-model.md
+  tracker/DECISIONS.md · DELIVERABLES.md · MILESTONES.md · RISKS.md · STAKEHOLDERS.md
+```
+
+If the scaffolder does not create `SCREENING.md`, `PROCESS-ARC.md`, `INTAKE-LOG.md`,
+or `ONRAMP-REPORT.md`, the skill writes them directly (stub templates).
+
+**Post-scaffold gate:** `_context.md` must be ABSENT.
+
+### 1.6 Corpus Setup (NEW)
+
+- Set `places.corpus_path` in DB:
+  ```sql
+  UPDATE places SET corpus_path = 'TPF/<slug>' WHERE slug = '<slug>' AND corpus_path IS NULL;
+  ```
+  (Adjust the prefix per project — `TPF/` for Place Fund, `RDC/` for internal, etc.)
+- Verify or create the corpus directory at `$CORPUS_ROOT/<corpus_path>/`
+- Wire `places/<slug>/corpus/INDEX.md` header to reference the global corpus path
+
+### 1.7 Process-State Arc Selection (NEW)
+
+Write `places/<slug>/PROCESS-ARC.md` — the model definition.
+Read the land-type selection matrix from `$CORPUS_ROOT/VLAS/plans/03-process-states.md` §5
+and produce a table for THIS place:
+
+```markdown
+# <Display Name> — Process-State Arc Selection
+> Archetype: <archetype> | Generated: <date>
+> Canon: VLAS/plans/03-process-states.md §5
+
+| State | Name | Class | Selection | Reason |
+|-------|------|-------|-----------|--------|
+| S00 | Inquiry | CORE | required | — |
+| S01 | Screening | CORE | required | — |
+| S02 | Regional Read | CORE | required | — |
+| S03 | Stakeholder Convening | CORE | required | — |
+| S04 | FPIC | CORE — ANNIHILATOR | required | S→0 if unmet |
+| S05 | Asset Entry | CONDITIONAL | S05a LRLT | land-regen: trust entry |
+| ... | ... | ... | ... | ... |
+| S11 | Stewardship Active | CONDITIONAL | required | land-regen: enterprise models live here |
+| S15 | Capital Formation | CONDITIONAL | required | land-regen: PRT/Place Fund raise |
+| ... through S24 ... |
+```
+
+Every `not-applicable` entry MUST carry a reason. A dropped state without a stated
+reason is a defect, not a design (03 §5 rule).
+
+### 1.8 Screening Stub (NEW)
+
+Write `places/<slug>/SCREENING.md` if absent:
+```markdown
+# <Display Name> — Screening (S01)
+> Status: pending | Date: <date>
+
+## Mission Fit
+<!-- TODO: assess alignment with regenerative mission -->
+
+## RCL Eligibility
+- Filter 1 (entity form): <!-- TODO -->
+- Filter 2 (field of use): <!-- TODO -->
+
+## Red Lines
+<!-- TODO: identify any hard stops (e.g. FPIC not obtainable) -->
+
+## Decision
+<!-- screening-pass | screening-decline (with reason) -->
+```
+
+### 1.9 Phase 1 Rubric Gate
+
+Print this table with actual values filled in. **A FAIL on any row stops the phase.**
+
+```
+PHASE 1 RUBRIC — <slug>
+| # | Check                    | Expected                          | Actual | Pass |
+|---|--------------------------|-----------------------------------|--------|------|
+| 1 | places row               | status=enrolling, corpus_path set | ?      |      |
+| 2 | project_nodes row        | prtnode-<slug> exists             | ?      |      |
+| 3 | project_places join      | links place to node               | ?      |      |
+| 4 | prt_projects row         | project_type set (not null)       | ?      |      |
+| 5 | regen_projects row       | exists (work_items FK)            | ?      |      |
+| 6 | place_phase_state        | row exists, initialized           | ?      |      |
+| 7 | archetype in metadata    | not TBD, matches --archetype      | ?      |      |
+| 8 | epic                     | exists, not archived              | ?      |      |
+| 9 | disk: places/<slug>/     | exists with all required files    | ?      |      |
+| 10| disk: PROCESS-ARC.md     | state selection matrix, >20 rows  | ?      |      |
+| 11| disk: SCREENING.md       | exists (stub OK for Phase 1)      | ?      |      |
+| 12| disk: INTAKE-LOG.md      | exists, Phase 1 row written       | ?      |      |
+| 13| disk: ONRAMP-REPORT.md   | exists, Phase 1 section written   | ?      |      |
+| 14| corpus: $CORPUS_ROOT dir | dir exists or created             | ?      |      |
+| 15| _context.md              | ABSENT                            | ?      |      |
+```
+
+Append Phase 1 results to `places/<slug>/ONRAMP-REPORT.md`:
+```markdown
+## Phase 1: Enrollment — <date>
+### Drift state: <fresh|already-enrolled|spine-ahead>
+### DB spine: places ✅ | project_nodes ✅ | project_places ✅ | prt_projects ⚠️ | regen_projects ✅
+### Archetype: <archetype>
+### Disk tree: N files created, M skipped (existing)
+### Rubric: [full table above, filled in]
+```
+
+Append Phase 1 row to `places/<slug>/INTAKE-LOG.md`.
+
+---
+
+## Phase 2: Research — Full Regenerative Regional Deep Research
+
+Research has THREE layers. All must complete before writing arc files.
+
+### 2.1 Research Prompt Engineering — Stored with the Model
+
+Each archetype has a **research prompt template** at
+`corpus/_shared/onramp/research-prompts/<archetype>.md`. The template structures a
+full Five Capitals regional deep research:
+
+| Capital   | Research areas                                                        | Min sources |
+|-----------|-----------------------------------------------------------------------|-------------|
+| Natural   | Ecoregion, soils (SSURGO), hydrology, vegetation (NVC), wildlife,     | 3 Tier 0-2  |
+|           | climate band, watershed, marine/tidal, habitat connectivity           |             |
+| Human     | Indigenous territory, FPIC holders, community demographics,           | 2 Tier 0-2  |
+|           | health/education infrastructure, cultural heritage (NHPA)             |             |
+| Social    | Governance structure, co-governance pathway, community orgs,          | 2 Tier 0-2  |
+|           | First Nations relationships, stakeholder network                      |             |
+| Built     | Existing infrastructure, permitted structures, heritage listings,     | 2 Tier 0-2  |
+|           | grid/energy, transport, marine access, communications                 |             |
+| Financial | Title chain, encumbrances, easements, tax status, water rights,       | 2 Tier 0-2  |
+|           | zoning, indicative raise, capital structure, comparable sales         |             |
+
+If the template does not exist for this archetype, the skill writes a default based on
+the Five Capitals table above and flags it in ONRAMP-REPORT.md.
+
+Run structured WebSearch calls per capital (not generic 4-6 searches). Target government
+databases, heritage registers, peer-reviewed studies. Use the web-research MCP for
+deep research or built-in WebSearch for targeted queries.
+
+### 2.2 Incoming Document Integration (MANDATORY)
+
+**⛔ HARD GATE: Every document in the incoming corpus folder MUST be read before arc files are written.**
+
+Locate the incoming corpus folder:
+- Check `$CORPUS_ROOT/<corpus_path>/` first (set in Phase 1)
+- Check Google Drive incoming folder
+- Check `places/<slug>/corpus/` for pre-existing documents
+
+For each document:
+1. **PDFs**: Extract text via `pypdf` (`from pypdf import PdfReader`)
+2. **Excel/XLSX**: Extract via `openpyxl` or `pandas`
+3. **CSV**: Read directly
+4. **Word/DOCX**: Convert to MD via `rdc:convert` or pandoc
+5. **Images**: Catalog for Phase 4 imagery
+
+**What to extract from project documents:**
+
+| Document Type | Extract |
+|--------------|---------|
+| Business Plan | Mission, pillars, revenue streams, team bios, partners list |
+| Pro Forma / Financials | CapEx total, phased investment, revenue projections, cost structure |
+| Execution Proposal | Phased timeline, team roles, partner LOIs |
+| Development Package | Property details, permits, water rights, infrastructure specs |
+| Due Diligence List | Regulatory requirements, title review, environmental assessments |
+| Competitor Analysis | Market positioning, pricing, differentiators |
+
+### 2.3 Financial Model — Invent if Absent
+
+If no financial model is provided in incoming documents:
+1. **Analyze**: comparable properties/projects in the region, infrastructure replacement
+   value, land value from tax assessments, revenue potential from archetype-typical
+   enterprise models (VLAS doc 04 §7 — the 14 stewardship enterprises × primary capital)
+2. **Build** an illustrative Tier-4 financial model: CapEx estimate, phased investment,
+   revenue streams, cost structure, pro forma projection
+3. **Tag EVERY figure** `(Illustrative — constructed from analysis, no project-supplied source)`
+4. Write to `arc/06-model.md` with full methodology disclosure
+5. Flag in ONRAMP-REPORT.md: `Financial model constructed from analysis — no incoming financials`
+
+Use the financial model analysis template at `corpus/_shared/onramp/financial-model-template.md`
+if available.
+
+### 2.4 Write Arc Files (Research Layer)
+
+The 6 arc files are the **research corpus** — organized by Five Capitals:
+
+| Arc file              | Primary capital(s) | Research feeds                         |
+|-----------------------|--------------------|----------------------------------------|
+| `01-story.md`         | narrative spine    | Place identity + vision + mission      |
+| `02-place.md`         | Natural, Built     | Ecoregion + infrastructure + heritage  |
+| `03-foundation.md`    | Human, Social      | People + governance + FPIC context     |
+| `04-process-outcomes.md`| Social, Built    | Programs + stewardship + enterprise models |
+| `05-investors.md`     | Financial          | Capital structure + raise + title chain |
+| `06-model.md`         | Financial          | Pro forma + revenue (supplied or constructed) |
+
+**Every number in 05 and 06 MUST cite its source document with page reference.**
+Tier 4 claims tagged `(Illustrative)`.
+
+### 2.5 Website Route Registry — Archetype-Specific (NOT Fixed 6-Arc)
+
+The arc files are research. The **website routes are a separate artifact** defined in
+`places/<slug>/PRODUCT.md`, derived from the archetype's telling
+(VLAS doc 02 — the six movements).
+
+Each archetype has a route template at `corpus/_shared/onramp/routes/<archetype>.md`.
+Different archetypes produce different site structures:
+
+- **land-regen** routes: `/`, `/place`, `/story`, `/foundation`, `/stewardship`,
+  `/invest`, `/history`, `/model`
+- **data-center** routes: `/`, `/place-writes-the-spec`, `/library`, `/pathway`,
+  `/civic-compute`, `/resources`
+- **HISTORY** is always its own route/artifact (sourced from `HISTORY.md`),
+  never folded into the generic 6-arc
+
+`PRODUCT.md` declares: `route → arc-file(s) → content mapping → archetype`
+
+If the route template does not exist, the skill writes a default land-regen route set
+and flags it in ONRAMP-REPORT.md.
+
+### 2.6 Source Tier Tracking
+
+Every source gets a row in `corpus/INDEX.md`:
+`| source | tier | capital | arc file | date accessed |`
+
+### 2.7 Document Gate
+
+Before proceeding to Phase 2b:
+```
+DOCUMENT GATE:
+  [ ] N/M incoming documents read (list each with filename + page count)
+  [ ] Financial model: supplied | constructed from analysis
+  [ ] Team and partners extracted
+  [ ] Timeline extracted: phases with dates
+  [ ] All data integrated into arc files with source citations
+  [ ] PRODUCT.md route registry written (archetype-specific)
+```
+
+If any incoming document was NOT read: **STOP and report which ones were skipped and why.**
+
+### 2.8 Phase 2 Rubric Gate
+
+```
+PHASE 2 RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | Natural capital sources        | >= 3 Tier 0-2                         | ?      |      |
+| 2 | Human capital sources          | >= 2 Tier 0-2                         | ?      |      |
+| 3 | Social capital sources         | >= 2 Tier 0-2                         | ?      |      |
+| 4 | Built capital sources          | >= 2 Tier 0-2                         | ?      |      |
+| 5 | Financial capital sources      | >= 2 Tier 0-2                         | ?      |      |
+| 6 | All incoming docs read         | 0 unread                              | ?      |      |
+| 7 | corpus/INDEX.md                | all sources with tier + capital        | ?      |      |
+| 8 | arc/01-06 each > 40 lines      | substantive content, not stubs        | ?      |      |
+| 9 | Every number in 05/06 cited    | source:page reference present         | ?      |      |
+| 10| Tier 4 claims tagged           | (Illustrative) on all                 | ?      |      |
+| 11| Financial model                | supplied or constructed + disclosed    | ?      |      |
+| 12| PRODUCT.md route registry      | archetype-specific routes declared    | ?      |      |
+| 13| places.corpus_path             | set in DB                             | ?      |      |
+| 14| HISTORY.md                     | exists if prt_projects.project_type qualifies | ? |      |
+```
+
+Append Phase 2 results to `places/<slug>/ONRAMP-REPORT.md`:
+```markdown
+## Phase 2: Research — <date>
+### Sources: N total (Tier 0: X, Tier 1: Y, Tier 2: Z, Tier 3: A, Tier 4: B)
+### Per-capital: Natural ✅|⚠️ | Human ✅|⚠️ | Social ✅|⚠️ | Built ✅|⚠️ | Financial ✅|⚠️
+### Financial model: supplied | constructed from analysis
+### Documents read: N of M incoming
+### Route registry: <archetype> — N routes declared
+### Gaps: [list any research areas with insufficient sourcing]
+### Rubric: [full table above, filled in]
+```
+
+Append Phase 2 row to INTAKE-LOG.md.
+
+---
+
+## Phase 2b: People, Organizations & Place Enrichment
+
+After research, before conflict analysis. Enrich and verify all named entities.
+
+### 2b.1 Extract People from All Sources
+
+Scan ALL project documents + arc files for named people:
+- Team members (founders, directors, managers)
+- Partners and advisors (named individuals)
+- Indigenous / First Nations contacts and elders
+- Government officials and agency contacts
+- Community leaders and stakeholders
+- Historical figures (for HISTORY.md context)
+
+Build a raw entity list: `name | mentioned_in | role_claimed | org_affiliation`.
+
+### 2b.2 Extract Organizations from All Sources
+
+Scan for named organizations:
+- Project partners (development, construction, technology)
+- First Nations bands, tribal councils, indigenous orgs
+- Government agencies (federal, state/provincial, municipal)
+- NGOs and conservation organizations
+- Financial institutions, investors, fund managers
+- Contractors, consultants, service providers
+
+Build a raw org list: `org_name | mentioned_in | relationship | domain_claimed`.
+
+### 2b.3 Categorize by Relationship Type
+
+Assign each entity a relationship category from the VLAS stakeholder taxonomy:
+- `core-team` — named on the project team
+- `partner` — LOI or MOU holder
+- `advisor` — advisory board or named advisor
+- `indigenous-rights-holder` — FPIC counterparty (ANNIHILATOR — must be identified)
+- `government` — regulatory or permitting authority
+- `community` — local community stakeholder
+- `investor` — capital provider or prospective
+- `contractor` — service delivery
+- `historical` — historical figure (HISTORY.md only, no verification needed)
+
+### 2b.4 RocketReach Person Lookup
+
+For each non-historical person, run verification:
+```
+mcp__claude_ai_RocketReach__rocketreach_lookup_person or rocketreach_search_people
+```
+Record: current_title, current_org, LinkedIn URL, email, verification_status.
+If no match: mark `unverified` — DO NOT invent data.
+
+### 2b.5 RocketReach Company Lookup
+
+For each organization:
+```
+mcp__claude_ai_RocketReach__rocketreach_lookup_company
+```
+Record: domain, employee_count, industry, location, verification_status.
+Cross-check claimed relationship against actual org profile.
+
+### 2b.6 Web Research on Organizations
+
+RocketReach gives you the company card. **Web research gives you the story.** For each
+significant org (core-team, partner, investor, indigenous-rights-holder), run targeted
+web research:
+
+1. **Website analysis:** find the org's website, read their about/mission/team pages.
+   Does the org's stated mission align with the project's claims about them?
+2. **News search:** search for recent news about the org. Any red flags (lawsuits,
+   regulatory actions, financial trouble, controversy)?
+3. **Social/community presence:** does the org have a real presence in the region?
+   LinkedIn company page, social media, community references?
+4. **Track record:** has this org done similar projects before? What was the outcome?
+   Search for their name + the project type (e.g., "org-name conservation" or
+   "org-name eco-tourism").
+5. **Financial standing:** for investor/financial entities, any public filings,
+   annual reports, or regulatory registrations (SEC, provincial securities)?
+
+For each org, record: `org | website | mission_fit | news_flags | regional_presence |
+track_record | verification_status`.
+
+Update arc files with findings — especially arc/03-foundation.md (partner orgs) and
+arc/05-investors.md (financial entities). **A partner org with no web presence is a
+yellow flag; an investor entity with no regulatory registration is a red flag.**
+
+### 2b.7 Government Registry Verification
+
+For key entities (project company, First Nations band, conservation trust):
+- Check government business registries (state/provincial)
+- Verify First Nations band recognition (federal indigenous affairs)
+- Confirm non-profit/charity status where claimed
+- Record registration numbers where available
+
+### 2b.8 Cross-Reference Against Arc Files
+
+Verify consistency between entity claims and arc file content:
+- Team bios in arc/03-foundation.md match RocketReach profiles
+- Partner orgs in arc/04-process-outcomes.md exist and match claimed domain
+- Investor entities in arc/05-investors.md are real institutions
+- Flag any entity mentioned in arc files but NOT in the raw extraction (missed entity)
+
+### 2b.9 Map Stakeholder Network
+
+Build a relationship map: who connects to whom, through what mechanism.
+- Direct relationships (employment, partnership, advisory)
+- Indirect relationships (shared board membership, common investor)
+- FPIC chain: which indigenous rights-holders must consent, and who is the contact
+- Governance chain: project → trust → community → rights-holder
+
+### 2b.10 Write STAKEHOLDERS.md
+
+Write verified entities to `places/<slug>/tracker/STAKEHOLDERS.md`:
+- **People section:** name, title, org, LinkedIn, verification status, relationship category
+- **Organizations section:** name, domain, size, relationship, verification status
+- **Network section:** key relationships mapped
+- **FPIC counterparties:** explicitly listed (even if unverified — the gap is the finding)
+
+**Content floor:** STAKEHOLDERS.md must have >20 lines of substantive content.
+If fewer than 20 lines after enrichment, the rubric FAILs.
+
+### 2b.11 Flag Unverified Entities
+
+Write ALL unverified entities to RECONCILIATION.md under a `## Unverified Entities` section:
+- Person: name, claimed role, what was searched, why no match
+- Org: name, claimed relationship, what was searched, why no match
+- Each flagged entity gets an escalation path: "verify via [specific next step]"
+
+### 2b.12 Update Arc Files with Verified Data
+
+Where RocketReach returned richer data than the arc files contain:
+- Update arc/03-foundation.md with verified titles and org affiliations
+- Update arc/05-investors.md with verified investor entity details
+- Tag each update: `(Verified via RocketReach <date>)`
+
+### 2b.13 Phase 2b Rubric Gate
+
+```
+PHASE 2b RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | People extracted               | all named people from all docs        | ?      |      |
+| 2 | Orgs extracted                 | all named orgs from all docs          | ?      |      |
+| 3 | Entities categorized           | relationship type assigned each       | ?      |      |
+| 4 | RocketReach person lookups     | run on all non-historical people      | ?      |      |
+| 5 | RocketReach company lookups    | run on all orgs                       | ?      |      |
+| 6 | Web research on orgs           | website + news + track record checked | ?      |      |
+| 7 | Government registry checks     | key entities verified                 | ?      |      |
+| 8 | Arc file cross-reference       | no missed entities                    | ?      |      |
+| 9 | Stakeholder network mapped     | relationships documented              | ?      |      |
+| 10| STAKEHOLDERS.md                | >20 lines substantive content         | ?      |      |
+| 11| FPIC counterparties identified | listed even if unverified             | ?      |      |
+| 12| Unverified entities flagged    | in RECONCILIATION.md with next steps  | ?      |      |
+| 13| Arc files updated              | verified data written back            | ?      |      |
+```
+
+Under `RDC_TEST=1`: skip RocketReach + registry lookups; write placeholder STAKEHOLDERS.md.
+
+Append Phase 2b results to ONRAMP-REPORT.md + INTAKE-LOG.md.
+
+---
+
+## Phase 3: Conflict Resolution
+
+Systematic fact verification across all sources — every factual claim in every arc file
+must be cross-checked, classified, and resolved before scoring.
+
+### 3.1 Build Fact Inventory
+
+Scan all 6 arc files and extract every factual claim into a structured inventory:
+- **Quantitative claims:** acreage, elevation, population, financial figures, dates
+- **Entity claims:** people, orgs, roles, relationships
+- **Legal claims:** title status, easements, zoning, permits, water rights
+- **Ecological claims:** species presence, habitat type, soil classification, watershed
+- **Historical claims:** dates, events, ownership transitions
+
+Each claim gets a row: `claim | arc_file | source | tier | category`.
+
+### 3.2 Cross-Source Verification — Web vs Documents
+
+For each factual claim, explicitly compare web research sources (Tier 0-2) against
+project-supplied documents (Tier 3-4):
+- Same fact, different numbers → `conflicted`
+- Web confirms document → `verified`
+- Web contradicts document → `conflicted` with both values logged
+- No web source found for document claim → `unverified`
+- Web source found, no document mention → `web-only` (flag for arc update)
+
+Log each comparison in RECONCILIATION.md: `claim | web_value | doc_value | status`.
+
+### 3.3 Cross-Document Verification
+
+Compare claims across incoming documents themselves:
+- Business plan vs pro forma (financial figures consistent?)
+- Execution proposal vs development package (timeline, team, scope consistent?)
+- Due diligence list vs actual permits/rights in arc files
+- Flag any internal document contradictions
+
+### 3.4 Financial Figure Verification
+
+Dedicated pass on all financial claims in arc/05-investors.md and arc/06-model.md:
+- CapEx total: consistent across all mentions?
+- Revenue projections: same base assumptions everywhere?
+- Phased investment amounts: sum correctly?
+- Pro forma: internally consistent (revenue - costs = margin)?
+- All Tier 4 figures tagged `(Illustrative)` — verify none are presented as verified
+
+### 3.5 Stakeholder Claim Verification
+
+Cross-check STAKEHOLDERS.md against arc file claims:
+- Every person mentioned in arc/03 appears in STAKEHOLDERS.md
+- Titles/roles match between arc files and verified RocketReach data
+- Partner orgs in arc/04 match STAKEHOLDERS.md org entries
+- No phantom stakeholders (mentioned in arc files, not in STAKEHOLDERS.md)
+
+### 3.6 Timeline Consistency Check
+
+Verify all temporal claims are internally consistent:
+- Project phases in arc/04 have realistic durations
+- Milestone dates in tracker/MILESTONES.md align with arc file narratives
+- Historical dates in HISTORY.md are chronologically valid
+- Regulatory timeline (permits, approvals) is plausible for jurisdiction
+
+### 3.7 Geographic and Legal Claim Verification
+
+Verify location-specific claims:
+- Property boundaries / APN match county records (Tier 0)
+- Zoning classification matches municipal records
+- Water rights claims match state/provincial records
+- Conservation easement claims match land trust records
+- Heritage/NHPA listings match federal/state registers
+
+### 3.8 Resolve All Conflicts
+
+For each `conflicted` or `unverified` claim, choose one resolution:
+- **verified** — web + doc agree, or authoritative source confirms
+- **resolved** — conflict existed, winner chosen with rationale documented
+- **held** — cannot resolve now; escalation path documented (who to ask, what to check)
+
+Update arc files with resolved values. Tag resolved claims:
+`(Resolved: <winning_source>, see RECONCILIATION.md RC-NNN)`
+
+### 3.9 Write RECONCILIATION.md
+
+Structure as a ledger with sections:
+1. **Verified Claims** — fact + sources that agree
+2. **Resolved Conflicts** — fact + both values + winner + rationale
+3. **Held Claims** — fact + why unresolvable + escalation path + impact if wrong
+4. **Unverified Entities** (from Phase 2b)
+
+Zero `conflicted` facts may remain. `held` facts must each have a documented escalation path
+and an impact assessment (what breaks if this fact is wrong).
+
+### 3.10 Update Arc Files with Corrections
+
+Write resolved values back into the arc files:
+- Replace conflicted figures with resolved values
+- Add source citations for newly verified facts
+- Remove or tag unverifiable claims
+- Ensure arc/05 and arc/06 financial figures are internally consistent post-resolution
+
+### 3.11 Update corpus/INDEX.md
+
+Add any new sources discovered during verification. Update tier assignments where
+a source's authority was re-assessed during conflict resolution.
+
+### 3.12 Phase 3 Rubric Gate
+
+```
+PHASE 3 RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | Fact inventory built           | every arc file scanned, claims listed | ?      |      |
+| 2 | Web vs doc comparisons         | explicitly logged per claim            | ?      |      |
+| 3 | Cross-document consistency     | internal doc contradictions resolved  | ?      |      |
+| 4 | Financial figures verified     | arc/05 + arc/06 internally consistent | ?      |      |
+| 5 | Stakeholder claims verified    | arc/03 matches STAKEHOLDERS.md        | ?      |      |
+| 6 | Timeline consistency           | dates coherent across all files       | ?      |      |
+| 7 | Geographic/legal verified      | boundaries, zoning, rights checked    | ?      |      |
+| 8 | Zero conflicted facts          | 0 remaining                           | ?      |      |
+| 9 | Zero unverified facts          | 0 remaining (or reclassified held)    | ?      |      |
+| 10| Held facts documented          | each with escalation + impact         | ?      |      |
+| 11| Arc files updated              | resolved values written back          | ?      |      |
+| 12| corpus/INDEX.md updated        | new sources + tier reassessments      | ?      |      |
+| 13| STAKEHOLDERS.md                | >20 lines substantive content         | ?      |      |
+```
+
+Append Phase 3 results to ONRAMP-REPORT.md + INTAKE-LOG.md.
+
+---
+
+## Phase 4: Brand Book + Imagery — Full rdc:design Integration
+
+Phase 4 **invokes `rdc:design`** — it does not duplicate design work. The skill prepares
+inputs (DESIGN.md brief + PRODUCT.md route registry from Phases 2/3) and calls `rdc:design`
+which produces the full brand system.
+
+### 4.1 Inputs to rdc:design
+
+- `places/<slug>/DESIGN.md` — the brief (identity, voice, audience, mood — written in Phase 2)
+- `places/<slug>/PRODUCT.md` — route registry, content map, site IA (archetype-specific)
+- `places/<slug>/arc/01-story.md` — narrative spine (Story of Place)
+
+### 4.2 The 24-Page Brand Book (Baru Model)
+
+`rdc:design` writes/updates `places/<slug>/DESIGN.md` as a **24-page editorial brand book**:
+
+| Spread | Section | Content |
+|--------|---------|---------|
+| 1 | Cover | Project name + tagline + hero image |
+| 2-3 | Story of Place | The narrative — who was here, what happened, what's next |
+| 4-5 | Market Position | Competitor landscape, pricing tier, differentiation |
+| 6-7 | Audience Segments | Primary + secondary audiences, psychographics |
+| 8-9 | Voice + Tone | Register, rhythm, prohibitions, example copy |
+| 10-11 | Color Palette | 10-12 tokens derived from ACTUAL place imagery, NOT generic green |
+| 12-13 | Typography | Display + body families, hierarchy, specimen |
+| 14-15 | Imagery Direction | Photography style, subject matter, treatment, existing imagery catalog |
+| 16-17 | Mood + Materiality | Visual register, textures, physical materials, digital materiality |
+| 18-19 | Content Architecture | Site map, route-to-arc mapping, content types per route |
+| 20-21 | Communications | Social strategy, press approach, stakeholder communications |
+| 22 | Token Reference | Full CSS custom property table with hex values |
+| 23 | Pattern Reference | Component patterns, animation vocabulary, interaction model |
+| 24 | Credits + Colophon | Contributors, sources, version |
+
+### 4.3 Palette — Full Studio Tokens via RAMPA
+
+**⛔ NEVER generate a generic dark green palette.**
+
+1. Search online for existing imagery of the project (previous names, partner sites, location photos)
+2. Use RAMPA or image analysis tools to extract dominant colors from ACTUAL place imagery
+3. Map colors to the place's identity — whatever is SPECIFIC to THIS place
+4. Generate the full token system:
+   - **Studio tokens** — CSS custom properties in the `--<slug>-*` namespace
+   - **Color system** — primary, secondary, accent, neutral, semantic mapped to Five Capitals
+5. Log which image(s) the palette was derived from (source proof in DESIGN.md)
+
+### 4.4 Imagery — Search First, Generate Second, regen-media Pipeline
+
+**Before generating ANY imagery with AI:**
+1. Search the web for existing non-copyright imagery of the project
+2. Catalog found images in `IMAGERY-PROMPTS.md` with source URLs and license status
+3. Only generate AI imagery for slots that have no real photography available
+
+**Required image slots (minimum per archetype — adjust per PRODUCT.md route set):**
+
+| Slot | Aspect | Used in | Prefer real vs AI |
+|------|--------|---------|------------------|
+| `hero-main` | 16:9 wide | Hero section | REAL |
+| `place-aerial` | 16:9 wide | /place header | REAL |
+| `ecology-wildlife` | 4:3 | /place ecology | REAL |
+| `heritage-structure` | 4:3 | /foundation or /story | REAL |
+| `programs-activity` | 1:1 | /stewardship or /regeneration | AI OK |
+| `og-image` | 1200x630 | Social share | Composite |
+
+**Generation pipeline — three-stage flow:**
+
+```
+Stage 1: Generate → output/imagegen/<slug>/
+  └ AI: regen-media MCP (generate_flux / generate_midjourney)
+        OR Codex image_gen (gpt-image-2, --enable image_generation)
+        OR OpenAI API direct
+  └ Each image: <slot-name>.png (e.g. hero-main.png, ecology-salmon.png)
+  └ REPORT.md in the same dir: slot, prompt used, model, dimensions, date
+
+Stage 2: Review → $CORPUS_ROOT/TPF/<slug>/artifacts/images/
+  └ After task review, move approved images to the corpus artifacts dir
+  └ This is the durable home (One Home Per Fact — images are artifacts)
+  └ Rejected images stay in output/imagegen/<slug>/ with _rejected suffix
+
+Stage 3: Wire → apps/<slug>/public/images/
+  └ Copy from corpus artifacts into the app's public dir
+  └ Wire imagery.ts (or equivalent) to reference the images
+  └ Verify all <Image> src imports resolve after build
+```
+
+- Video placeholders: declare slots in IMAGERY-PROMPTS.md even if content isn't ready
+- The `output/imagegen/<slug>/` directory is the generation landing zone — always
+  project-organized, never flat
+- IMAGERY-PROMPTS.md is the manifest: it declares what to generate, and REPORT.md
+  in the output dir records what was actually generated
+
+### 4.5 Brand Gate
+
+If `--no-gate` is NOT set: stop and report. Re-invoke with `--skip-to 5`.
+
+### 4.6 Phase 4 Rubric Gate
+
+```
+PHASE 4 RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | DESIGN.md                      | 24-spread outline, not a stub         | ?      |      |
+| 2 | Palette source                 | derived from THIS place's imagery     | ?      |      |
+| 3 | Palette source proof           | image filename(s) logged in DESIGN.md | ?      |      |
+| 4 | Studio tokens                  | --<slug>-* CSS properties defined     | ?      |      |
+| 5 | IMAGERY-PROMPTS.md             | all slots declared with aspect+mood   | ?      |      |
+| 6 | Image slots filled             | hero, aerial, ecology, heritage min   | ?      |      |
+| 7 | Image sources                  | real search done before AI gen        | ?      |      |
+| 8 | No generic dark green          | palette specific to THIS place        | ?      |      |
+| 9 | Brand review gate              | passed (or --no-gate)                 | ?      |      |
+```
+
+Append Phase 4 results to ONRAMP-REPORT.md + INTAKE-LOG.md.
+
+---
+
+## Phase 5: Regen Score — Model-Driven Rubric Assessment
+
+The Regen Score is **driven by the VLAS model specification** (corpus 01), not a standalone
+scoring system. The archetype assigned in Phase 1 determines which model rubric applies,
+which process states are required (corpus 03 §5), which Five Capitals are emphasized,
+and which annihilators must be checked. **The model IS the rubric.**
+
+Scoring is NOT a one-pass read-and-number exercise. Each dimension requires regional
+research correlation — you look at the region, discover what comparables exist, check
+how this place fits the regional pattern, and circle back for deeper evidence where the
+first pass left gaps. Multiple research passes per dimension are expected.
+
+### 5.1 Load Model Specification
+
+Load the model's scoring context from the VLAS corpus and archetype templates:
+- Read `$CORPUS_ROOT/VLAS/plans/01-model-spec.md` — the universal template (§6 SCARF×4Ps,
+  §7 Five Capitals, §8 the arc, §12 recognition formula + annihilators)
+- Read `$CORPUS_ROOT/VLAS/plans/03-process-states.md` — state library + §5 land-type
+  selection matrix (Data Center / Land Regen / Urban Renewal)
+- Read `corpus/_shared/onramp/benchmarks/<archetype>.md` — archetype-specific benchmarks
+- Read `places/<slug>/PROCESS-ARC.md` — this place's state selection (from Phase 1)
+
+The model spec defines **what gets scored and how**. If the archetype benchmark set does
+not exist, the skill writes a default from the model spec §7 (Five Capitals) + §6
+(SCARF×4Ps lattice) and flags it for review.
+
+### 5.2 Regional Comparable Research — First Pass
+
+Before scoring, establish the regional context this place sits within:
+- **Comparable projects:** search for similar archetype projects in the same region
+  (state/province, ecoregion, economic zone). What are they? What did they cost?
+  What revenue do they generate? What regulatory path did they follow?
+- **Regional economic indicators:** median land values, construction costs per sqft,
+  tourism/hospitality revenue baselines, agricultural yields for the bioregion
+- **Regulatory landscape:** what permits/approvals do comparable projects in this
+  jurisdiction typically require? What are typical timelines?
+- **Conservation/ecological benchmarks:** ecoregion health baselines, species
+  inventories for the bioregion, watershed health metrics from regional agencies
+
+This is research — use WebSearch, web-research MCP, or existing corpus. Log all
+new sources in corpus/INDEX.md with tier assignments.
+
+### 5.3 SCARF×4Ps Lattice Assessment
+
+Fill the 5×4 grid from model spec §6 for this place. Each cell is a question the
+model answers for a real project:
+
+```
+| SCARF \ 4P     | Place            | People           | Purpose          | Process          |
+|----------------|------------------|------------------|------------------|------------------|
+| Source          | ?                | ?                | ?                | ?                |
+| Condition       | ?                | ?                | ?                | ?                |
+| Appropriateness | ?                | ?                | ?                | ?                |
+| Resilience      | ?                | ?                | ?                | ?                |
+| Feedback        | ?                | ?                | ?                | ?                |
+```
+
+For each cell:
+- Fill with evidence from arc files (cite arc file + source tier)
+- Mark as **gate-generating** (produces a hard gate), **filter-generating** (produces
+  a soft filter), or **neither**
+- Cells that are gate-generating feed directly into the annihilator check (5.7)
+- Empty cells are scoring gaps — flag for additional research or mark as
+  `declared absence (reason)` (doc 01 §0.3: *a declared absence is an artifact*)
+
+Write the filled lattice to `places/<slug>/SCREENING.md`.
+
+### 5.4 Five Capitals Per-Capital Scoring
+
+Score each capital per model spec §7 — canonical definitions, never redefined.
+Each capital has: **definition** → **what it means for this archetype** → **patterns
+that build it** (doc 04) → **indicators** → **evidence citation**.
+
+| Capital   | Unit | Score 0-100 | Evidence | Source Tier | Arc File(s) |
+|-----------|------|-------------|----------|-------------|-------------|
+| Natural   | NCU  | ?           | ?        | ?           | arc/02      |
+| Human     | HCU  | ?           | ?        | ?           | arc/03      |
+| Social    | SCU  | ?           | ?        | ?           | arc/03, 04  |
+| Built     | MCU  | ?           | ?        | ?           | arc/04      |
+| Financial | FCU  | ?           | ?        | ?           | arc/05, 06  |
+
+**Financial is a translation layer** — it captures monetized value from gains in the
+other four (doc 01 §7). Score on whether the financial model supports the other capitals,
+not as an independent measure of profitability.
+
+**Coverage rule (doc 01 §5):** a plan that maxes one capital and starves another is
+**visibly lopsided and fails.** No capital may score below 40 while another scores
+above 80 (the coverage spread gate). Whole-place or nothing.
+
+**Weighting honesty:** any weighting used is **pedagogical, not measured.** Label it
+`Illustrative`. Measured outcomes come from verification (S20), never the planner.
+
+Each per-capital score MUST cite: which arc file, which source (tier), what evidence
+justifies the number. No bare scores. Tier 4 figures that are the SOLE evidence for a
+sub-score → cap that sub-score at 50.
+
+### 5.5 Annihilator Check — Binary Pass/Fail
+
+The recognition formula is a **product** — any zeroed term collapses everything.
+Gates are **annihilators, not low scores** (doc 01 §12.2, doc 03 §2).
+
+Check each annihilator and record binary status:
+
+| Annihilator | State | Check | Status | Evidence |
+|-------------|-------|-------|--------|----------|
+| FPIC        | S04   | Consent counterparties identified + pathway | ? | STAKEHOLDERS.md |
+| Readiness   | S13   | Place ecology can hold development | ? | arc/02 |
+| Baseline    | S08   | Five Capitals baseline captured or plan exists | ? | arc files |
+| Covenant    | S14   | Legal pathway identified | ? | arc/05, tracker/ |
+| Permanence  | S20   | π_legal × π_ecology both non-zero pathway | ? | HISTORY.md, arc/02 |
+| Verification| S20   | Measurement plan for ΔI exists | ? | PROCESS-ARC.md |
+| Offtake     | S21   | Revenue/offtake pathway identified | ? | arc/06 |
+| Dividend    | S22   | ≥10% community, before investor allocation | ? | arc/05 |
+
+At enrollment, most annihilators cannot be fully passed — but the **pathway to passing
+them** MUST be identified. A place with no FPIC counterparty identified and no plan to
+identify them is a NO-GO regardless of other scores.
+
+Status per annihilator:
+- **CLEAR** — evidence exists or a credible pathway is documented
+- **AT RISK** — gap identified, pathway unclear but not impossible
+- **BLOCKED** — no pathway visible → NO-GO signal
+
+Any annihilator **BLOCKED** → composite capped at NO-GO (<55) regardless of capital scores.
+
+### 5.6 Recognition Formula Readiness
+
+Assess each term of the normative issuance kernel (doc 01 §12, doc 09 §2):
+
+`Q = A_cap × ΔI_c × P × S × v^α × a^β × (1-U)`
+
+At enrollment, this is a **readiness assessment**, not a computed Q:
+
+| Term | Question | Evidence | Readiness |
+|------|----------|----------|-----------|
+| A_cap (scale) | Area/asset quantified? | Survey, title, GIS | ? |
+| ΔI (integrity uplift) | Baseline plan + measurement methodology? | PROCESS-ARC.md | ? |
+| P (permanence) | Legal + ecological durability pathway? | arc/05, HISTORY.md | ? |
+| S (stewardship/FPIC) | Consent + governance quality? | STAKEHOLDERS.md | ? |
+| v (velocity) | Time-series plan for improvement rate? | tracker/MILESTONES.md | ? |
+| a (acceleration) | Sustained care plan beyond initial cycle? | arc/04 | ? |
+| (1-U) (uncertainty) | Source quality + measurement confidence? | corpus/INDEX.md tiers | ? |
+
+Each term: **Ready** / **Partially ready** / **Not ready** / **N/A (with reason)**.
+
+### 5.7 Process State Completion Audit
+
+Read `places/<slug>/PROCESS-ARC.md` (Phase 1) and check which states have evidence
+of completion vs which are still required.
+
+For each state S00–S24 marked `required`:
+1. What evidence exists that this state has been addressed?
+2. Is the evidence sufficient to pass the state's exit gate?
+3. Which states are CORE vs CONDITIONAL for this archetype?
+
+Cross-reference against the land-type selection matrix (doc 03 §5):
+- Data Center: S05b site control, S11 N/A, S15 N/A
+- Land Regen / Conservation: S05a LRLT, S11 core, S15 core
+- Urban Renewal: S05c no land trust, S11 optional, S15 project-dependent
+- Eco-hospitality / Coastal-watershed: map to closest column + archetype-specific overrides
+
+**A composite ≥75 with a CORE state missing is invalid.**
+
+ANNIHILATOR states (S04 FPIC, S08 Baseline, S13 Readiness, S14 Covenant, S20 Verification,
+S21 Recognition, S22 Return Flow) with no evidence → composite invalid regardless of number.
+
+### 5.8 Pathway Ladder Position
+
+Determine current and target rung (doc 01 §9 — five rungs, ascending):
+1. Where does this place sit today? (current rung, with evidence)
+2. What is the target rung? (from PROCESS-ARC.md or project vision)
+3. Is an operating example at the target rung known? If not, say so honestly.
+4. What is the rising floor — the regulatory trajectory in this jurisdiction?
+5. Is the project building above or at the current floor?
+
+The aspirational rule: if no operating example exists at the top, **say so.** Building
+to the current floor is building to be obsolete.
+
+### 5.9 Regional Correlation — Second Pass (Gap Fill)
+
+After first-pass scoring, identify capitals that scored below 75 individually:
+- For each weak capital, do a targeted regional research pass:
+  - What specific evidence is missing?
+  - Can it be found via web research in the region?
+  - Are there comparable projects whose public data fills the gap?
+  - Correlate: does new regional data change the picture for other capitals too?
+- Update arc files with newly found evidence
+- Re-score the weak capital with new evidence
+- Log the research cycle in ONRAMP-REPORT.md
+
+This may require **multiple rounds** — you discover the region's shape, correlate what
+matters for this place in this region, then circle back for deeper digs on topics that
+emerged. Research is iterative. A gap that CANNOT be filled after targeted research
+stays as-is and contributes to the NEEDS WORK or NO-GO verdict.
+
+### 5.10 Honesty Audit
+
+Verify all claims meet the honesty standard (doc 01 §11):
+- Every Tier 4 figure tagged `(Illustrative)` — no unverified figure scored as fact
+- Tier 4 figures that are the SOLE evidence for a capital score → cap at 50
+- Every pattern, instrument, and rung carries a precedent tier
+  (Operating · Pilot · Proposal · Standard · Regulatory)
+- `announced-not-operating` flags on any case study cited
+- Financial Verified Mode: no unsourced figure ships as fact
+- The `[I]` register is identified — what is illustrative vs verified
+- Log: "N of M scoring evidence points are Tier 4 — ceiling applied to [capitals]"
+
+### 5.11 Compose Regen Score from Model Rubric
+
+Aggregate all scoring components — NOT a simple 5-dimension mean. The model rubric
+drives the structure:
+
+```
+REGEN SCORE — <slug> — <archetype>
+Model: <archetype> (loaded from $CORPUS_ROOT/VLAS/plans/ + archetype benchmarks)
+
+SCARF×4Ps Lattice: NN/20 cells filled with evidence (gate-generating: N)
+
+Five Capitals (per-capital, evidence-cited):
+  Natural:   __/100  [arc/02, Tier _]
+  Human:     __/100  [arc/03, Tier _]
+  Social:    __/100  [arc/03-04, Tier _]
+  Built:     __/100  [arc/04, Tier _]
+  Financial: __/100  [arc/05-06, Tier _]
+  Coverage spread: PASS/FAIL (max-min <= 40)
+
+Annihilators (binary — any BLOCKED → NO-GO):
+  FPIC:        CLEAR / AT RISK / BLOCKED
+  Readiness:   CLEAR / AT RISK / BLOCKED
+  Baseline:    CLEAR / AT RISK / BLOCKED
+  Permanence:  CLEAR / AT RISK / BLOCKED
+  Offtake:     CLEAR / AT RISK / BLOCKED
+  Dividend:    CLEAR / AT RISK / BLOCKED
+
+Recognition Formula: __/7 terms ready
+Process States: __/__ required states addressed
+Pathway: Rung __/5 current → Rung __/5 target
+Research Passes: N rounds (gaps identified + re-researched)
+
+Composite: __ — GO ≥75 | NEEDS WORK 55-74 | NO-GO <55
+```
+
+Write the full score card to `places/<slug>/SCREENING.md` (appending to the lattice
+from 5.3). This is the durable scoring artifact.
+
+### 5.12 Score Gate Decision
+
+| Composite | Verdict | Action |
+|-----------|---------|--------|
+| ≥ 75 | **GO** | Proceed to Phase 6 (Build) |
+| 55-74 | **NEEDS WORK** | List specific gaps; skill STOPS. Re-invoke after addressed. |
+| < 55 | **NO-GO** | Project does not meet threshold. Skill STOPS. Escalate to Dave. |
+
+If NEEDS WORK: list exactly what would move the score to GO — which capital needs
+evidence, which annihilator needs a pathway, which process state needs attention.
+A NO-GO is a success of the model if the place genuinely isn't ready.
+
+Append the full scoring narrative to ONRAMP-REPORT.md:
+```markdown
+## Phase 5: Regen Score — <date>
+### Model: <archetype> (benchmark set: loaded/generated)
+### SCARF×4Ps: NN/20 cells filled (N gate-generating)
+### Five Capitals:
+| Capital   | Score | Evidence | Tier 4 count | Ceiling | Key citation |
+|-----------|-------|----------|--------------|---------|-------------|
+| Natural   | NN    | X        | Y            | yes/no  | arc/02:LNN  |
+| Human     | NN    | X        | Y            | yes/no  | arc/03:LNN  |
+| Social    | NN    | X        | Y            | yes/no  | arc/04:LNN  |
+| Built     | NN    | X        | Y            | yes/no  | arc/04:LNN  |
+| Financial | NN    | X        | Y            | yes/no  | arc/06:LNN  |
+### Coverage spread: PASS/FAIL
+### Annihilators: [list each with status]
+### Recognition Formula Readiness: __/7 terms
+### Process States: __/__ required addressed
+### Pathway: Rung __→__ | Rising floor: [jurisdiction context]
+### Composite: NN — [GO / NEEDS WORK / NO-GO]
+### Gaps (if NEEDS WORK): [specific actionable items per capital]
+```
+
+### 5.13 Phase 5 Rubric Gate
+
+```
+PHASE 5 RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | Model spec loaded              | VLAS 01 + 03 + archetype benchmarks   | ?      |      |
+| 2 | Regional comparables researched| >= 2 comparable projects found        | ?      |      |
+| 3 | SCARF×4Ps lattice filled       | 20 cells answered with evidence       | ?      |      |
+| 4 | All 5 capitals scored          | 0-100 each with evidence citation     | ?      |      |
+| 5 | Coverage spread gate           | max-min <= 40                         | ?      |      |
+| 6 | Annihilators checked           | all 8 binary pass/fail recorded       | ?      |      |
+| 7 | No BLOCKED annihilator         | all CLEAR or AT RISK                  | ?      |      |
+| 8 | Recognition formula assessed   | all 7 terms readiness recorded        | ?      |      |
+| 9 | Process states audited         | required states checked per archetype | ?      |      |
+| 10| Pathway ladder position set    | current + target rung with evidence   | ?      |      |
+| 11| Regional gap-fill pass done    | weak capitals re-researched           | ?      |      |
+| 12| Honesty audit passed           | Tier 4 tagged, precedent tiers set    | ?      |      |
+| 13| Composite >= 75                | GO                                    | ?      |      |
+| 14| SCREENING.md written           | lattice + score card + evidence       | ?      |      |
+| 15| Score in ONRAMP-REPORT.md      | Phase 5 section appended              | ?      |      |
+```
+
+Append Phase 5 results to ONRAMP-REPORT.md + INTAKE-LOG.md.
+
+---
+
+## Phase 6: Site Build — Archetype Routes + RegenOps Integration
+
+### 6.1 App Structure — Archetype-Specific Routes
+
+Create `apps/<slug>/` following the Baru model (`apps/baru-website/`) but with
+**archetype-specific routes** from `PRODUCT.md` (not a fixed 6+1).
+
+- Routes declared in `PRODUCT.md` route registry (Phase 2.5)
+- HISTORY is always its own route (sourced from `HISTORY.md`), never folded into the arc
+- Section components with REAL content from arc files — not placeholder text
+- Financial model page with ACTUAL numbers from the pro forma (tagged Illustrative where Tier 4)
+- Real imagery from Phase 4 (not placeholder gradients)
+- Every image/video slot from IMAGERY-PROMPTS.md has a placeholder component even if
+  the asset isn't ready — the slot EXISTS in the route
+
+### 6.2 RegenOps Screens — Same Flow, TinTin/AI at Every Step
+
+The onramp pipeline runs IN regenops-app. Required screens (report missing, advisory not block):
+
+- **Intake/Screening** — S00/S01: project arrives, mission fit, archetype selection
+- **Research Dashboard** — S02/S07/S10: Five Capitals progress, source tier counts, gaps
+- **Conflict Resolution** — fact-by-fact review, tier comparison, reconciliation log
+- **Regen Score Card** — 5-dimension radar, composite, per-dimension evidence drill-down
+- **Phase State Timeline** — visual arc showing S00–S24 progress for this place
+- **Enrollment Overview** — place card, archetype, score, current phase, next action
+
+**MD read/write file tree (REQUIRED in RegenOps project view):**
+- Browsable tree of `places/<slug>/` markdown files
+- Click to read any .md inline
+- Edit and save .md files (PLACE.md, arc/*, tracker/*, DESIGN.md, PRODUCT.md)
+- Read-only for generated files (`_context.md`)
+- Uses FS MCP WebDAV bridge for cloud sessions; direct filesystem for CLI
+
+**TinTin/AI at every step:**
+- Each screen has a TinTin dispatch button (`POST /api/tintin/dispatch`)
+- AI assist: "research this gap", "score this dimension", "draft this arc section",
+  "generate image for this slot"
+- Dave stays in the RegenOps flow; AI is the assistant, not the driver
+
+### 6.3 PUBLISH.md + DEPLOY Block
+
+Per `app-deploy-manifest.md`, generate `PUBLISH.md` with `<!-- DEPLOY -->` block
+before deploy. Use `scripts/gen-deploy-blocks.mjs` or write manually.
+
+### 6.4 Build + Commit Discipline
+
+1. `pnpm --filter @regen/<slug> install`
+2. `npx tsc --noEmit` — must pass
+3. `pnpm --filter @regen/<slug> build` — must produce all routes
+4. Commit: infra (lockfile) first with `RDC-Bypass:`, then product with `Work-Item:`
+
+### 6.5 Cross-File Route Consistency
+
+After route creation, verify route names are consistent across ALL project files —
+not just PRODUCT.md vs the app. Grep `places/<slug>/` for every route path and
+fix any stale references:
+
+- `DESIGN.md` content map → must reference current route names
+- `arc/*.md` route headers → must reference current route names
+- `ONRAMP-REPORT.md` route listings → must reference current route names
+- `IMAGERY-PROMPTS.md` slot routes → must reference current route names
+
+**This is a rubric FAIL if any file references a route that no longer exists.**
+The skill renamed routes but left stale references → that is the defect this step prevents.
+
+### 6.7 Route Content Verification
+
+After build, verify each route renders REAL content:
+- No placeholder text ("Lorem ipsum", "Coming soon", "TODO")
+- Financial figures match arc/05-investors and arc/06-model
+- Team/partner data matches tracker/STAKEHOLDERS.md
+- All `<Image>` src imports resolve — no broken images
+
+### 6.8 Phase 6 Rubric Gate
+
+```
+PHASE 6 RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | apps/<slug>/ exists            | Next.js App Router                    | ?      |      |
+| 2 | Routes match PRODUCT.md        | all declared routes present           | ?      |      |
+| 3 | Cross-file route consistency   | 0 stale route refs in DESIGN/arc/report| ?     |      |
+| 4 | tsc --noEmit                   | exit 0                                | ?      |      |
+| 5 | pnpm build                     | all routes built                      | ?      |      |
+| 6 | Route content verification     | no placeholders, real arc content     | ?      |      |
+| 7 | Financial figures match arc    | 05/06 cited figures rendered on site  | ?      |      |
+| 8 | Image imports resolve          | no broken <Image> src                 | ?      |      |
+| 9 | HISTORY route                  | own route, sourced from HISTORY.md    | ?      |      |
+| 10| PUBLISH.md + DEPLOY block      | exists, port: registry                | ?      |      |
+| 11| RegenOps screens               | report present/missing (advisory)     | ?      |      |
+```
+
+Append Phase 6 results to ONRAMP-REPORT.md + INTAKE-LOG.md.
+
+---
+
+## Phase 7: Deploy Dev — Phase State + Epic Closure
+
+### 7.1 Registry Setup
+
+```sql
+-- 1. apps table (parent)
+INSERT INTO apps (slug, display_name, description, monorepo_path, github_org,
+  github_repo, runtime, owner_slug, status, pnpm_filter) VALUES (...);
+
+-- 2. app_deployments (child — FK on app_slug)
+-- Environment enum: 'dev' (NOT 'development')
+-- Port: query max(pm2_port) + 1 from app_deployments
+INSERT INTO app_deployments (app_slug, environment, host_type, url,
+  pm2_name, pm2_port, branch, build_command, status) VALUES (...);
+```
+
+### 7.2 Deploy to Vultr PM2
+
+```bash
+export CI=true
+git fetch -q origin develop && git reset -q --hard origin/develop
+pnpm --filter @regen/<slug> install --frozen-lockfile
+pnpm --filter @regen/<slug> build
+node scripts/generate-dev-configs.mjs --write   # --write flag REQUIRED
+pm2 start /srv/regen/ecosystem.config.js --only <slug>  # .js not .cjs
+```
+
+### 7.3 Verify + Phase State Advance
+
+- `curl -s https://<slug>.dev.place.fund/` → HTTP 200 + correct `<title>`
+- All routes return 200
+- Content spot-check: financial figures match arc files
+
+After successful deploy:
+```sql
+UPDATE place_phase_state SET current_phase = 7, phase_name = 'Deploy Dev',
+  gate_status = 'passed', updated_at = now()
+WHERE place_slug = '<slug>';
+```
+
+### 7.4 Epic Closure
+
+- Submit implementation report with `codeflow_post` via `submit_implementation_report`
+- Tick all checklist items via `update_checklist_item`
+- Transition epic to `review` via `update_work_item_status`
+- Validator closes to `done`
+
+### 7.5 INTAKE-LOG.md Finalization
+
+All 7 phase rows written with timestamps, operator, session ID.
+
+### 7.6 Phase 7 Rubric Gate
+
+```
+PHASE 7 RUBRIC — <slug>
+| # | Check                          | Expected                              | Actual | Pass |
+|---|--------------------------------|---------------------------------------|--------|------|
+| 1 | app_deployments row            | exists with pm2_port allocated        | ?      |      |
+| 2 | PM2 running                    | pm2 show <slug> = online              | ?      |      |
+| 3 | HTTP 200                       | <slug>.dev.place.fund returns 200     | ?      |      |
+| 4 | Content spot-check             | title correct, figures match arc      | ?      |      |
+| 5 | place_phase_state              | advanced to phase 7, gate=passed      | ?      |      |
+| 6 | Epic status                    | review or done                        | ?      |      |
+| 7 | Implementation report          | submitted with codeflow_post          | ?      |      |
+| 8 | INTAKE-LOG.md                  | all 7 phase rows present              | ?      |      |
+| 9 | ONRAMP-REPORT.md               | all 7 phase sections present          | ?      |      |
+```
+
+Append Phase 7 results to ONRAMP-REPORT.md (final section).
+
+---
+
+## Resumability
+
+| Phase | Skip condition |
+|-------|---------------|
+| 1 — Enrollment | DB row + disk tree + epic all exist + Phase 1 rubric gate all PASS (15 checks) |
+| 2 — Research | Phase 2 rubric gate all PASS (14 checks) + ONRAMP-REPORT.md Phase 2 section |
+| 2b — People/Org | STAKEHOLDERS.md >10 lines + unverified flagged in RECONCILIATION.md |
+| 3 — Conflicts | Phase 3 rubric gate all PASS (7 checks) + zero conflicted/unverified |
+| 4 — Brand | Phase 4 rubric gate all PASS (9 checks) + --<slug>-* tokens defined |
+| 5 — Score | Phase 5 rubric gate all PASS (6 checks) + composite ≥75 |
+| 6 — Build | Phase 6 rubric gate all PASS (10 checks) + tsc + build clean |
+| 7 — Deploy Dev | Phase 7 rubric gate all PASS (9 checks) + HTTP 200 + phase_state advanced |
+
+---
+
+## Rules
+
+- `enroll_place` is the ONLY door for spine rows
+- `insert_work_item` is the ONLY door for work items
+- `_context.md` must be ABSENT after scaffolding
+- disk-ahead → STOP, never proceed
+- **Every incoming document MUST be read** — no unread docs allowed past Phase 2
+- **Every financial number MUST cite its source document** — no invented figures
+- **Palette MUST be derived from actual place imagery** — no generic dark green
+- **Search for existing project imagery BEFORE generating AI images**
+- Source tiers: 0=Recorded legal, 1=Government, 2=Independent, 3=Project upload, 4=Stakeholder claim
+- Tier 4 claims tagged `(Illustrative)` on all public surfaces
+- Split infra (lockfile) and product commits — scope guard enforces this
+- Under `RDC_TEST=1`: DB writes, web searches, builds, and deploys short-circuit
+
+## Capture Lessons
+
+Before the final verdict, follow `.rdc/guides/lessons-learned-spec.md`. Write `.rdc/lessons/<YYYY-MM-DD>-onramp-<slug>.md` if the run taught something non-obvious.