# Energy Decision Stack: Synthesized Action Plan

**Generated:** April 2, 2026
**Source:** Five independent LLM analyses (Thiel lens, Altman lens, Musk lens, Amodei/Anthropic lens, Error Audit) synthesized into a single prioritized plan.

---

## The Unanimous Verdict

Every perspective agrees: the thesis is contrarian, specific, and well-constructed. Every perspective also agrees: the package cannot convert to capital, a hire, or a design partner on the strength of analysis alone. The single missing piece cited by all five is a measured result from a real workflow at a real company.

---

## Priority 1: Fix Internal Contradictions (Done / Hours)

These were identified in the error audit and executed in the v32.1 patch session:

- [x] **Malformed CSV** (v31 claim ledger line 31): unquoted commas in C30 claim text caused 7-field parse instead of 5. Fixed by quoting.
- [x] **"404 roles" vs "404 positions"**: Canonical definition is 404 positions = 373 roles + 24 workflows + 7 artifacts. Fixed in index.html meta tag, methodology_faq.md, methodology_faq_sensitivity_paragraph.txt, 00_SENSITIVITY_INDEX.md, SENSITIVITY_AT_A_GLANCE.txt.
- [x] **Stale exhibit/doc counts in index.html**: "9 interactive exhibits" corrected to 10; "14 linked supporting documents (22+)" corrected to "25 linked supporting documents (32 in the full bundle)."
- [x] **Stale internal reference in v33_upgrade_roadmap.md**: `thread_context L310` reference updated to point to superseded file location.
- [x] **Cycle-time contradiction**: deal_readiness_memo says "4-6 weeks → 8-12 days"; operator_memo and eval_spec say "15-20 business days → ≤5 business days." These measure different scopes (end-to-end including mobilization vs. core packet prep). Fixed by adding scope labels to both docx tables.
- [x] **75% vs 90% detection gap**: Proof artifact shows 3/4 planted issues detected (75%), but operator_memo and eval_spec set >=90% as the success bar. Added explicit disclaimer to proof HTML footer acknowledging the gap and what it takes to close it.

---

## Priority 2: One Measured Pilot (Weeks, Not Months)

Every perspective — Thiel, Altman, Musk, Amodei — converges on this as the single most important next step.

**What to measure (from your own eval_spec):**

| Metric | Baseline | Target |
|--------|----------|--------|
| Cycle time (core packet prep) | 15-20 business days | <=5 business days |
| Analyst hours per cycle | 120-160 hrs | <=40 hrs |
| Citation fidelity | manual spot-check | >=95% traceable to source |
| Numeric reconciliation accuracy | unknown | >=99% |
| Error escape rate | 8-12% rework | <3% rework |
| Conflict detection rate | manual | >=90% of planted/known issues |
| Reviewer burden | full rebuild | review + signoff only |

**Who:** One mid-cap E&P approaching spring/fall 2026 redetermination. The deal_readiness_memo already identifies this buyer.

**Why this is the unlock:** A single before/after result converts every part of this package from "modeled" to "measured." It is the difference between admiration and conviction.

---

## Priority 3: Shrink the Wedge Language Everywhere

Thiel is most explicit, but all perspectives agree: "AI for energy" is too broad. The winning pitch narrows to one painful loop, one buyer, one sales motion.

**Current framing (too wide):** "AI for energy decision workflows"

**Target framing:** "We own reserve-based lending redetermination support for mid-cap E&Ps."

**Where to update:**
- index.html hero section and meta tags
- deal_readiness_memo opening paragraph
- audience_specific_briefs.md (investor brief)
- distribution_package.md email templates
- Any pitch deck or outreach materials

The Altman lens adds a useful expansion frame: "the evidence operating system for regulated capital workflows" — energy is the wedge, but the platform story is broader. Use the narrow version for fundraising and sales; use the broader version for talent and partnerships.

---

## Priority 4: Describe the Moat in Software Terms

Every perspective flags that "domain expertise" sounds like consulting. Reframe the defensibility as:

- **Workflow memory** — structured context that persists across redetermination cycles
- **Exception corpus** — catalog of edge cases, human corrections, and failure modes that grows with each deployment
- **Source-to-claim tracing** — provenance chain from raw document to output claim, auditable by reviewer
- **Eval harness** — automated accuracy and detection benchmarks that run on every output
- **Cross-company learning loops** — anonymized patterns across multiple deployments that improve the system

The sentence: "Every cycle makes the system harder to copy" is the moat story.

---

## Priority 5: Make the Distribution Story Concrete

From the Thiel lens (Palantir model): high-ticket, technically deep, founder-led sales. Answer three questions:

1. **Who signs the first check?** VP Finance / Treasurer at a $1-5B E&P with 2+ borrowing-base cycles per year.
2. **Why do they buy now?** Spring 2026 redetermination cycle is approaching. Covenant headroom is tight in the current commodity environment. Every day of delay risks covenant breach.
3. **How do you get the next 10?** Each completed cycle produces a case study. The E&P CFO network is small and relationship-driven. Word of mouth from one successful deployment is the distribution channel.

---

## Priority 6: Audience-Specific Packaging

The core work is the same, but the wrapper differs by audience:

**For Thiel / Founders Fund / VC:**
- Lead with the monopoly wedge, not the market size
- One buyer, one workflow, one measured result, one compounding moat
- Scott Nolan at Founders Fund covers energy/infrastructure — this is his lane

**For Altman / OpenAI:**
- Reframe as "unblocking AI infrastructure by accelerating power, interconnection, permitting, lender, and rate-case workflows"
- Maps to Stargate bottlenecks and OpenAI's stated resource constraints
- The evidence OS framing resonates here

**For Musk / xAI:**
- Strip all OpenAI/Claude references from any version sent this direction
- Reframe around Tesla Megapack energization, xAI data center power procurement, SpaceX facility permitting
- xAI is actively hiring structured-finance and power-generation roles in Memphis

**For Amodei / Anthropic Institute:**
- Lead with the proof room, the misses, and the update plan — not the outreach package
- Turn the package into a reproducible benchmark: ship scoring code, version the pipeline, add changelog
- Replace synthetic proof with 3-5 redacted real workflows with human-labeled outcomes
- Do a head-to-head model comparison (Claude vs. GPT vs. Gemini) on the same eval set
- The Analyst role at The Anthropic Institute is nearly a direct match for this work

---

## Priority 7: Reproducibility Upgrades (v33 Roadmap)

From the Amodei lens and the existing v33_upgrade_roadmap.md:

1. **Ship the scoring/sensitivity code** — the Python script exists but isn't in the package
2. **Publish row-level provenance** — source_basis, confidence_level, uncertainty_band columns in the CSV
3. **Complete the inter-rater study** — 3-5 analysts independently score 30-50 roles, measure agreement
4. **Replace formulaic employment estimates** — expand BLS-sourced rows from 60 to 120+
5. **Fix the company_control_zone definition** — reconcile outcome-oriented vs. prep-work-oriented definitions

---

## The One-Line Pitch (Per Audience)

**VC / Thiel:** "Everyone thinks AI hits energy first in the field. It actually hits first in the evidence stack that governs financing and regulation. We are starting with reserve-based lending redeterminations because the questions recur, the buyer is obvious, the workflow is measurable, and every cycle makes the system harder to copy."

**OpenAI / Altman:** "We found a non-obvious wedge in energy, turned it into a workflow product, have one paying design partner, and can prove we cut a real cycle from weeks to days with auditable outputs and human signoff intact."

**xAI / Musk:** "The real bottleneck to scaling physical AI is not just chips or transformers; it is the document-heavy financing, permitting, interconnection, compliance, and review loops around power and infrastructure. We built the evidence OS that compresses one of those loops 3-10x, with citations, audit logs, and human signoff."

**Anthropic / Amodei:** "We study how AI changes real work, not just benchmark scores. We built a reproducible framework for measuring AI impact on high-stakes evidence workflows in energy, and we can show exactly where the model succeeds, where it fails, and where the human must stay in the loop."

---

## What Was Fixed Today (v32.1 Patch Log)

| File | Issue | Fix |
|------|-------|-----|
| `superseded/energy_decision_stack_claim_ledger_v31.csv` | Line 31 (C30) unquoted commas causing 7-field parse | Quoted claim_text field |
| `index.html` | Meta tag says "404 roles" | Changed to "404 positions" |
| `index.html` | Says "9 interactive exhibits" | Corrected to "10" |
| `index.html` | Says "14 linked supporting documents (22+)" | Corrected to "25 linked supporting documents (32 in the full bundle)" |
| `docs/methodology_faq.md` | "404 roles" in sensitivity paragraph | Changed to "404 positions" |
| `docs/methodology_faq_sensitivity_paragraph.txt` | "404 roles" | Changed to "404 positions" |
| `docs/00_SENSITIVITY_INDEX.md` | "404 energy roles" in data specs | Changed to "404 energy positions (373 roles, 24 workflows, 7 artifacts)" |
| `docs/SENSITIVITY_AT_A_GLANCE.txt` | "404 energy industry roles" | Changed to "404 energy positions (373 roles, 24 workflows, 7 artifacts)" |
| `docs/v33_upgrade_roadmap.md` | Stale reference to `thread_context L310` | Updated to point to superseded file location |
| `docs/deal_readiness_memo.docx` | Cycle time row says "4-6 weeks" without scope | Added "(end-to-end incl. mobilization)" label |
| `docs/operator_memo_treasury_lender_readiness.docx` | Cycle time row says "15-20 business days" without scope | Added "(core packet preparation)" label |
| `proof/treasury-lender-readiness.html` | Footer claims 75% detection "matches the claim" without acknowledging 90% success bar | Replaced with explicit acknowledgment of the gap and path to close it |