One defined capability request
A seeded request asked for a TypeScript service that extracts bank-statement transactions, redacts PII, deploys to AWS Lambda, and permits commercial modification.
FunctionFoundry / Capability economics
A seeded agent-capability request became a constraint-aware build-versus-buy record with explicit estimates, rights coverage, and a no-spend decision boundary.
Verified July 27, 2026Observed result
The comparison was read-only: it spent no funds, created no purchase, and treated acquisition prices as planning inputs requiring a separate confirmed request.
Evidence chain
A seeded request asked for a TypeScript service that extracts bank-statement transactions, redacts PII, deploys to AWS Lambda, and permits commercial modification.
Runtime, deployment, rights, price, security, and feature requirements were compiled before options were ranked.
The deterministic prototype model compared a 213,000–426,000 token rebuild range with a 10,000–28,000 token integration range and a seeded $799 acquisition input.
The rights-compatible option covered 12 of 12 hard requirements, but overlapping modeled cost ranges prevented a false-precision buy recommendation.
Reproduce it
Node 25.2.1 · pnpm 10.13.1 · Vitest 3.2.7
Download evidence manifestCommand
pnpm exec vitest run \
apps/web/src/lib/build-vs-buy.test.ts \
apps/web/src/app/api/agent/build-vs-buy/route.test.tsSHA-256 fingerprints
Dependency lock396bafe5a089f4edbe3bfdff4f3fc17537d259b1217e9e3001548957bd092a0c
Comparison engine3cea528710928d8295ff29dd564655e34ab2df04be17ecdd49afdf7f5cdd196b
Decision tests7311768d5ca5d1db7c2474883d2d155127b7df38a9fd748a55f3f5517ab9090a
Interpretation boundary
The token ranges are directional planning estimates, not observed agent usage or provider billing.
The $799 acquisition price and two-hour integration estimate are seeded inputs, not live market quotes.
This record demonstrates deterministic decision behavior on a fixture; it is not a claim of realized client savings.
Apply the method