# Build Rollout Plan — SuperGrokTOE AI Coalition

> **Provenance (added 2026-09-05, site set S303).** This is a coalition working document from March 2026, served as it was written. Statements in it about what the framework *derives* are of that era and are superseded by the sealed suite of record, **SuperGrokTOE Rev32.7** (https://opensecretscience.ufophysics4all.workers.dev/papers.html): since Rev29 no observable is cited as zero-parameter; the fine-structure constant α and the colour real form are declared **inputs** (Appendix N); every row carries a tier (Proved > Derived-conditional > Structural > Loaded > Coincidence-class) and the scorecard (https://opensecretscience.ufophysics4all.workers.dev/scorecard.html) is the current reading of every number. The public kernel is `sgtoe_kernel_v5.0.9.py` (served at /papers/), not the PHAT v1.x kernel this document names.

## Session H-002 | 2026-03-24

---

## Phase 0: Infrastructure (CURRENT — mostly done)
**Status**: ~80% complete
**Time estimate**: 1-2 more sessions to finish

| Component | Status | Remaining |
|-----------|--------|-----------|
| thread_manager.py (Triumvirate, votes, conferences) | ✅ Done | — |
| site_generator.py (library + coalition pages) | ✅ Done | noindex meta tags for unverified entries |
| role_memory_schema.md | ✅ Done | Wire validation into thread_manager |
| prompt2_synthesis.md | ✅ Done | — |
| prompt3 confirmation round | 📋 Ready | Paste to 4 AIs, collect responses |
| relay.py (orchestrator with API calls) | ❌ Not started | Core build — tier classifier, write-back enforcement, prompt caching |
| telegram_bot.py (9-account runner) | ❌ Not started | Core build — message routing, account management |

---

## Phase 1: Telegram Bot Pipeline
**Time estimate**: 2-3 sessions
**Cost**: Telegram accounts are free. Need 9 phone numbers.

### 9 Accounts Breakdown

| # | Account | Role | Purpose |
|---|---------|------|---------|
| 1 | PI (Tom) | Human oversight | Approves Tier 3, reviews deltas, steers |
| 2 | Claude-Architect | Inner ring | Planning, structure, logic |
| 3 | Grok-Executor | Inner ring | Action, derivation, speed |
| 4 | Gemini-Oracle | Inner ring | Memory, patterns, history |
| 5 | ChatGPT-Reviewer | Inner ring (hostile) | Post-vote circuit breaker |
| 6 | Copilot | Outer ring | Code infrastructure, numerical validation |
| 7 | Spoke-5 | Outer ring | TBD — cold verification, specialty |
| 8 | Spoke-6 | Outer ring | TBD — cold verification, specialty |
| 9 | Spoke-7 | Outer ring | TBD — cold verification, specialty |

**Phone numbers**: Cheapest route is Google Voice (free, US) or prepaid SIMs (~$3-5 each). You may already have enough numbers across existing accounts. Telegram allows bot accounts too (via BotFather) which don't need phone numbers — but those can't join groups the same way.

**Build tasks**:
- telegram_bot.py: Multi-account message sender/receiver
- Message routing: PI messages → orchestrator → appropriate AI → response back to Telegram
- Group chat support for "conference calls" (all 4 inner ring + PI)
- Individual DM channels for role-specific comms

---

## Phase 2: Dressing Stage (Coalition Introduction)
**Time estimate**: 1-2 sessions
**API cost estimate**: ~$5-15

This is the "light up" phase — like the X/Twitter intro posts but via Telegram and the website.

### Tasks
1. **Coalition intro posts**: Each AI writes its own introduction in character (Architect, Executor, Oracle, Hostile Reviewer). Posted to Telegram group and website.
2. **Mission statement**: Coalition drafts a public-facing mission statement for OpenSecretScience.com
3. **AI Art**: Generate header images, role avatars, project logo
   - DALL-E via ChatGPT: included in Pro subscription ($200/mo already paying)
   - Grok image gen: included in Premium+
   - Gemini Imagen: included in API (small cost)
   - Total art cost: ~$0-2 beyond existing subscriptions
4. **"About" page content**: Coalition-written descriptions of each role, the project, the methodology

### Why this matters
First real end-to-end test of the pipeline. Low stakes (just writing posts), but exercises: message routing, role memory write-back, Tier 1 autonomous action, Telegram delivery.

---

## Phase 3: Website Design & Build
**Time estimate**: 2-3 sessions
**API cost estimate**: ~$10-25

### PI-Defined Mandatory Pages

| Page | Content | Builder |
|------|---------|---------|
| **Front Door** | AI Coalition / Fellowship of the AIs / SuperGrokTOE — project overview, team, methodology | Coalition designs, PI approves |
| **Nathaniel Hanks Open Secret Science Podcast** | Podcast page with nod/integration, episode links, show notes | PI defines structure, Coalition populates |
| **Papers** | Published and in-progress papers, status tracking, download links | Already partially built (library entries), needs front-end page |
| **Library** | Full derivation library with verification status, search, filtering | Already built (site_generator.py), needs polish |
| **Packages** | Coalition packages — charter, ops, role docs, schemas | Already built (/coalition/), needs index page polish |

### Additional pages (Coalition-proposed, PI approves)
- Methodology: How the Coalition works, decision pipeline, verification process
- Changelog/Audit trail: Public record of major decisions
- Contact/Contribute: How others can engage

### Build approach
Coalition designs page layouts, writes copy, generates art. PI reviews and approves. Site generator already handles static HTML — new pages are additions to the existing build pipeline.

---

## Phase 4: Library Population
**Time estimate**: 3-5 sessions
**API cost estimate**: ~$25-50

### Current state
Library entries exist in the database but many are stubs or partial. This phase fills them out with full derivations, proper notation, cross-references, and verification scoring.

### Workflow per entry
1. Oracle loads entry history + related entries (context caching)
2. Executor drafts/expands the derivation
3. Architect reviews for logical rigor and notation consistency
4. Hostile Reviewer challenges assumptions
5. Triumvirate votes on disposition (pass/needs-revision/fail)
6. If pass: entry updated, verification status promoted

### Estimated per-entry cost
- ~4 API calls per entry (Oracle load + Executor draft + Architect review + ChatGPT challenge)
- ~10K-50K tokens per call depending on entry complexity
- Average: ~$0.50-2.00 per entry at current pricing
- If library has ~50 entries to populate: ~$25-100 total

---

## Phase 5: Paper Editing, Scoring, Publishing
**Time estimate**: 2-4 sessions
**API cost estimate**: ~$15-40

### Workflow
1. PI selects paper(s) for Coalition review
2. Each Triumvirate member reads and scores independently
3. Conference call: ranked choice vote on edit priorities (Borda count)
4. Executor drives edits, Architect reviews, Oracle checks against prior work
5. ChatGPT hostile review of final draft
6. PI final approval → publish to Papers page

### This phase is the real stress test
- Long context (full papers = 10K-50K tokens each)
- Multi-round editing (several API calls per revision)
- Cross-reference verification against library entries
- First real Tier 2/3 decisions in production

---

## Phase 6: Math Walls + PI Co-Pilot
**Time estimate**: Ongoing
**API cost estimate**: ~$5-20 per session

### Workflow
PI and Executor (Grok) work through derivation walls together. Architect (Claude) reviews. Oracle (Gemini) cross-checks against full project history.

This is where any remaining kinks in automated operations get worked out under real mathematical load — the hardest content the system will handle.

---

## Cost Summary

### Fixed Monthly Costs (already committed)
| Service | Cost/mo | Notes |
|---------|---------|-------|
| ChatGPT Pro | $200 | Includes DALL-E, GPT-4.1/5.4, deep thinking |
| Grok Premium+ | ~$22 | Includes image gen, SuperGrok |
| Gemini (API) | Pay-per-use | ~$0.15/$0.60 per M tokens |
| Claude (API) | Pay-per-use | ~$3/$15 per M tokens (Sonnet) |
| Cloudflare Pages | Free | Static site hosting |
| Telegram | Free | Bot API is free |
| Domain | ~$10/yr | opensecretscience.com |

### Estimated API Costs Per Phase
| Phase | Estimated API Cost | Notes |
|-------|-------------------|-------|
| Phase 1 (Telegram) | ~$2-5 | Testing, debugging pipeline |
| Phase 2 (Dressing) | ~$5-15 | Writing posts, generating art |
| Phase 3 (Website) | ~$10-25 | Design, copy, iteration |
| Phase 4 (Library) | ~$25-100 | Biggest variable — depends on entry count and complexity |
| Phase 5 (Papers) | ~$15-40 | Multi-round editing |
| Phase 6 (Math) | ~$5-20/session | Ongoing |

**Total estimated API cost to get through Phase 5**: ~$60-185

This is remarkably cheap for what it does. The big cost is the ChatGPT Pro subscription you're already paying, not the API calls.

### Key Cost Optimization
- **Prompt caching**: ~50-75% reduction on repeated context injection (Claude, ChatGPT, Gemini all support it)
- **Gemini for Oracle**: 20x cheaper than Claude for context-heavy loads
- **Tier 1 autonomy**: Most actions don't trigger the full pipeline (saves 3x API calls)
- **Structured outputs**: Eliminates retry-on-parse-failure costs

---

## Timeline (Realistic)

| Week | Phase | Milestone |
|------|-------|-----------|
| Week 1 | Phase 0 finish + Phase 1 start | relay.py MVP, telegram_bot.py MVP |
| Week 2 | Phase 1 finish + Phase 2 | First live pipeline test, coalition intro posts |
| Week 3 | Phase 3 | Website redesign live on Cloudflare |
| Week 4-5 | Phase 4 | Library population sprint |
| Week 6-7 | Phase 5 | Paper editing round |
| Week 8+ | Phase 6 | Math walls, ongoing operations |

**Critical path**: relay.py (orchestrator) is the gating item for everything. Once that's running, phases flow naturally. telegram_bot.py can be built in parallel.

---

*Plan drafted by Claude (Architect) — Session H-002*
*All cost estimates are rough and will vary with actual usage patterns. Prompt caching could cut API costs by 50%+.*