Political Ad Provenance Scan Launch Kit

What shipped

A runnable Python 3.11+ standard-library CLI with sample-launch-kit, generate-launch-kit, and validate-launch-kit. It packages exactly seven copied synthetic parent outputs into 12 fixed commercial/operational artifacts, a sample index, seven unmistakably labeled sample derivatives, and a canonical lineage manifest. The generated offer is fixed at $750 USD paid upfront for one race, up to 10 ads, a 14-day evidence window, delivery within 48 hours after complete intake is accepted, and a 30-minute readout.

The project includes strict profile/source schemas, exact roles and CSV headers, safe relative-path and symlink checks, source and derivative SHA-256 reconciliation, deterministic templates, five unsent outreach touches, five stable fictional prospect rows, safety/leakage scanning, prohibited-conclusion enforcement, atomic output replacement, rollback behavior, unit/mutation tests, and an offline verify.sh workflow.

Architecture

The seven source fixtures were copied by value from one verified synthetic parent demo. Five aggregates and only the changed-creative and unresolved-link client-safe receipts are bundled; their bytes remain unchanged.
Runtime has no parent dependency. Evidence production stays in the separately operated parent desk; this code only packages allowlisted outputs.
Profile data is an allowlist, not a template language. Commercial and safety terms are code-owned and fixed.
Generation builds the complete output tree in memory, writes and validates a sibling staging directory, verifies input immutability, then swaps outputs with rollback on failure.
Independent validation reloads contracts and recomputes expected bytes, source/packaged hashes, paths, counts, scenario markers, sample labels, content requirements, and safety checks.
launch-kit.json indexes every generated artifact except itself, avoiding a cyclic self-hash.

Trimmed scope

This is a one-off launch/delivery kit for one race, not a new evidence system. There is no ingestion, evidence graph, finding logic, source research, live adapter, scraping, browsing, OCR/transcription, media classification, payer inference, legal interpretation, contact enrichment, outreach sending, CRM, payment/signature flow, UI, database, API server, PDF output, monitoring subscription, white-label implementation, or general proposal builder.

Limitations

The bundled content is fictional and demonstrates packaging only. Hashes prove byte identity after copying, not truth, authenticity, authorization, or prior chain of custody. Automated leakage scanning cannot establish that human-entered evidence is accurate or publication-safe. Seller fields intentionally remain placeholders. Real engagement fulfillment still requires the separate evidence desk, authorized buyer inputs, a named human reviewer, secure human-operated transfer, payment/signature outside this program, and an explicit retention/deletion decision.

Verification

From the build root, run bash verify.sh. It requires only Python 3.11+ and POSIX shell tools, creates temporary workspaces, exercises all three commands, verifies byte-identical complete trees, tests overwrite refusal and atomic rollback, runs negative mutations, and scans runtime/test code for forbidden parent or external-service execution paths.

Suggested next steps

Present the synthetic packet to five hard-gated boutique compliance consultants or small political agencies and record objections without adding live contact data to this repository.
Seek one paid $750 scan or two serious buyer-data review calls; use a sanitized, authorized engagement workspace outside the fixture repository.
Record repeated input friction before considering a narrowly scoped importer.
Test $1,000 per race per week monitoring only after a paid scan, and a $3,500 white-label setup only after repeat demand.