What shipped
A focused Python 3.11+ standard-library CLI with exactly three commands: sample-proof-kit, generate-proof-kit, and validate-proof-kit. It builds a buyer-ready $750 proof-review package containing qualification, baseline, evidence-supply, SOW, delivery, QA, review, demo, outreach, objections, pricing, sample-navigation, and canonical lineage artifacts.
The bundled fixture has exactly 14 inputs: one strict service profile, one strict source manifest, and 12 immutable copied-by-value parent artifacts. Generation emits exactly 27 outputs: 13 top-level human artifacts, one sample README, 12 labeled derivatives, and proof-kit.json, which hashes the other 26 outputs under an acyclic rule.
Architecture
- Standard-library-only source package runnable without installation, credentials, environment variables, network, services, or parent checkout.
- Closed fixed-value JSON contracts with duplicate-key rejection, exact schemas, canonical dates/money, safe paths, exact inventories, UTF-8/LF enforcement, and filesystem/content boundary checks.
- Parent hashes and semantic facts compiled into independent validation so coordinated source/manifest drift cannot pass.
- One fixed renderer per commercial artifact; no arbitrary templates or configurable pricing language.
- Exact Markdown, CSV, and JSON synthetic-derivative rules with source and derivative lineage.
- Independent validation re-renders every expected byte rather than trusting the existing
proof-kit.json. - Sibling staging, input snapshots, validated publication, rollback, and constrained cleanup for deterministic atomic generation.
- Safe sample replacement: default refusal for non-empty targets and forced replacement only for a workspace that already validates as the exact fixture.
Suggested next steps
Do not add engineering scope before testing demand. The next-day test is five warm or directly reachable agency prospects with active receptionist pilots: apply all eight gates, present the fixed $750 paid-upfront offer, use the synthetic 90-second demo only when useful, and record the exact objection. Success is one paid $750 review; a strong secondary signal is one concrete active-pilot data-fit review. Expand to ten qualified offers before changing the product, and archive rather than build speculative integrations if neither signal appears.