---
name: build-token-value-estimate
description: >-
  Estimate total AI tokens spent building since a start date (default 2025-11-01)
  and intrinsic value from token-cost equivalents, using multi-op OP_CONTEXT
  fan-out across ClaudeCode_v1 + CodexCode_v1 + GrokBuild_v1 (not exclusive).
  Triggers: "/build-token-value", "estimate tokens since", "how many tokens built",
  "intrinsic value of build", "token cost of work since Nov", "AI spend estimate",
  "build ROI tokens", "multi-op token valuation".
---

# build-token-value-estimate

Estimate **(A) total tokens used to build** since a horizon date and **(B) intrinsic value**
expressed as token-cost dollars (API-list equivalent) plus optional human-replacement band.
Always Tarski-honest: name what is measured vs proxied vs undecidable.

## Invariants

1. **Multi-op is the default** when ≥2 of ClaudeCode / CodexCode / GrokBuild are ready.
   Fan-out OP_CONTEXT; merge as a multiset (OTC). Cascade only if the user says "one op only".
2. **Claude `stats-cache.json` is primary** for Claude tokens when present.
3. **Grok / Codex session stores are proxies** (bytes → tokens × utilization) unless billing APIs exist.
4. **Intrinsic value ≠ invoice.** Report L1 cash, L2 API-list, L3 human separately.
5. **Never invent precision.** Prefer ranges; mark measured / proxy / residual.
6. **Product name is Pharosiraptor** (never "Pharos").

## Horizon

| Input | Default |
|---|---|
| `since` | `2025-11-01` |
| `pricing_as_of` | see `data/methodology.json` |

## Collect evidence

```text
1. ~/.claude/stats-cache.json → modelUsage, dailyModelTokens, dailyActivity
2. Session trees: ~/.claude/projects/** · ~/.grok/sessions/** · ~/.codex/sessions/**
3. Optional: Pharosiraptor GET /api/ai-assistant/backends
```

```bash
python scripts/collect_evidence.py --since 2025-11-01 --out evidence.json
python scripts/estimate_value.py --evidence evidence.json --out estimate.json
```

## Pricing (USD / 1M tokens)

| Tier | input | output | cache_read | cache_write |
|---|---:|---:|---:|---:|
| opus | 15 | 75 | 1.50 | 18.75 |
| sonnet | 3 | 15 | 0.30 | 3.75 |
| haiku | 1 | 5 | 0.10 | 1.25 |

Grok / Codex blended proxy: **$3–10 / MTok** on utilization-adjusted session tokens.

## Multi-op fan-out

| Op | Lens | Ask |
|---|---|---|
| ClaudeCode_v1 | META+GENESIS | Validate Claude fold; non-cache vs cache bounds |
| CodexCode_v1 | MGS+GENESIS | Session-byte → token util factor range |
| GrokBuild_v1 | SHADOW+MGS | Name Tarski residuals; attack overclaims |

Prompts for top-10 agentic systems live in `prompts/`. Compose prompt: `prompts/00-multi-op-compose.md`.

## Value layers (never collapse)

| Layer | Meaning | Typical class |
|---|---|---|
| **L1 cash paid** | Invoices / seats | tarski-residual without billing |
| **L2 API-list** | tokens × public list prices | measured + proxy |
| **L3 human opportunity** | eng hours × rate | values-question |
| **Artifact market** | IP / revenue | oracle-required |

## Residuals (always name)

- Cash paid without billing export
- Grok/Codex exact tokens without vendor usage API
- Subscription vs API list divergence
- Cache tokens dominate Claude totals
- Artifact market value is oracle-required
- Work on other machines / harnesses out of scope

## Dashboard + clone

- Static dashboard: `index.html` + `data/estimate.json`
- Dumb-HTTP bare git: `git clone https://build-token-value.pages.dev/git/build-token-value.git`