BUILD NOTES / HONEST BY DESIGN

One complete recovery.
Clear about what’s real.

PayTrace Recover helps merchants resolve cases where a customer paid but didn’t receive the digital access they purchased. The current prototype demonstrates that promise through one paid course that stays locked. It joins payment and access evidence, asks the merchant to approve a specific correction, and checks the outcome. Membership and download recovery are future directions, not implemented integrations.

This public site is a browser simulation.

Records are fictional, specialist consultations are scripted, and checkout moves no money. The working server and integration adapters are in the project source. PayPal sandbox, live AI, and Render accounts have not been configured or tested end to end.

Try nine different outcomes

  1. A · Correct payment, missing access. Investigate, rehearse, approve, apply, and verify. Then open Maya’s first lesson. During rehearsal, the original stays locked.
  2. B · Wrong order reference. A matching price cannot prove the purchase. The case stops and access stays locked.
  3. C · Verification read fails once. Access is granted, but the case remains open. Retry the reads and watch the write count stay at one.
  4. D · Missing payment evidence. The review holds. Supply the sample record, then investigate again.
  5. E · Access already restored. A store retry beat us to the fix. Verify with zero repair writes.
  6. F · Evidence changes after review. Investigate and rehearse, then use the sample-change control. The old recommendation expires; a new review catches the mismatch.
  7. G · Wrong account signed in. Route the customer to the purchasing account; do not grant another account access.
  8. H · Access term expired. Review the purchased access term before proposing a change.
  9. I · Payment pending. Wait for payment confirmation; a pending record cannot authorize access.

Five connected improvements

Access Watch checks a fictional portfolio against your chosen delivery window. Select a learner, filter related failures, and investigate the specific diagnosis. Each case keeps its own approval. The customer response is an editable evidence-based template; it only claims restoration after current verification and is never sent by this app.

The browser monitor runs while the page is visible. The Node backend checks saved purchases every 30 seconds while running. Related errors suggest a pattern, not a proven shared cause. Complaint timing remains unknown unless recorded.

What the specialists contribute

Ledger checks payment evidence. Relay traces the failed handoff without inventing its cause. Patch proposes the smallest access correction. Prism challenges contradictions and checks the outcome. Expand the challenge card for objections, a short exchange, and the exact evidence snapshot reviewed. The optional model adapter makes six bounded calls: three specialist findings, Prism’s challenge, one response, and a final review. An unresolved objection holds the repair. AI cannot authorize or execute a correction.

From sample to sandbox

  1. Run the server. The source includes a Node 24 application with SQLite persistence and a server-side lesson-access check against the sample entitlement store. Run npm ci, npm test, then npm start.
  2. Configure PayPal sandbox. Add sandbox client credentials to the server environment. Unlock the developer session, restart the story, and use the optional PayPal sandbox button. The server creates and captures the order, then deliberately fails its own access update.
  3. Configure AI review. Add an OpenAI API key and a Structured Outputs-compatible Responses model. The interface labels model-generated findings distinctly. Invalid or incomplete output fails visibly.
  4. Run Render Workflows. Deploy the Node backend and workflow entry point. Configure the shared workflow secret and exact task slugs. Investigation finishes before approval. Approval starts a separate correction-and-verification run.

Detailed environment instructions, provider references, and acceptance checks are in docs/SETUP.md in the project source. Hosting the Node service and running provider integrations may incur charges. This build has not provisioned those services.

The Render and AG Studio build direction

Render will coordinate investigation and a separate approved recovery workflow. AG Studio is planned for linked evidence widgets, case filtering, repair previews, and scoped agent questions. The sales-page workspace is a custom HTML design preview, and the recovery console uses custom HTML and JavaScript. AG Studio is not installed in this version.

The Node verifier now makes an authenticated HTTP request for protected lesson content using a short-lived capability scoped to one learner, course, and case. It checks the returned content and revokes the token. This exercises a separate content endpoint and stored access record in our demo service. It does not prove access on an external course platform or a real customer’s device. Production learner identity and a live sandbox acceptance run remain future work.

The development sequence is integration validation, case ownership and durable repair records, sandbox checkout and protected content, reliable Render workflows, the AG Studio workspace, customer-side verification, and deployed acceptance checks. See docs/DEVELOPMENT_PLAN.md in the source for the complete plan.

What has been checked

The automated suite covers isolated rehearsal, mismatched purchase fields, stale approval, repeat repairs, failed verification, session isolation, origin enforcement, operator gating, and mocked provider calls. The challenge tests also cover missing evidence, zero-write verification, unresolved objections, incomplete AI responses, and changed snapshots. Mocked tests support the implementation; they do not replace real sandbox and workflow acceptance tests.

What still needs real-world evidence

The project includes a five-seller interview kit. Ask about the last actual incident, the records they checked, its frequency, their existing workaround, and what would make a supervised trial credible. No interviews have been conducted and no demand or revenue claims are made.

A production version also needs a real course-platform connector, merchant and learner identity, verified webhook handling, shared transactional storage, retention controls, and refund-related access policies. This prototype is intentionally limited to one fictional course and one correction type.

The three-minute story

Start with the locked learner. Investigate the evidence. Rehearse while the learner remains locked. Approve one exact grant. Verify and open the lesson. Finish by showing the conflicting-payment stop and the failed-read retry. If live integrations are configured, show actual PayPal capture and Render run records alongside the result.

Return to the recovery desk ↗