Resolvia

AI investigates. Evidence decides.

1. What Resolvia is

Resolvia is an evidence-driven dispute resolution workspace. It keeps claims, evidence, provenance, Truth Graph projections, deterministic policy, and AI proposals as distinct concepts so agents investigate without inventing truth.

2. Why WebMCP matters

WebMCP lets an external agent inspect the same authoritative case state the human sees through structured tools—not by scraping the UI.

3. The demo scenario — RV-1028

Open /cases/RV-1028. The seeded synthetic case is deliberately incomplete: provider confirmation and customer receipt evidence are still missing, so deterministic readiness is NOT READY.

4. WebMCP tools

Tool	Role
resolvia_get_case	Case inspection
resolvia_get_truth_graph	Truth Graph projection
resolvia_list_resolution_gaps	Missing verification gaps
resolvia_check_resolution_readiness	Deterministic readiness
resolvia_prepare_evidence_request	Draft evidence request only

5. What agents CAN do

• Inspect case state
• Reason over Truth Graph and gaps
• Identify missing evidence
• Check deterministic readiness
• Prepare a human-reviewable evidence request (draft)

6. What agents CANNOT do

• Resolve cases
• Issue refunds
• Promote evidence
• Modify the Truth Graph
• Bypass deterministic policy

7. Architecture

Browser (WebMCP tools)
  → same-origin POST /api/webmcp/invoke
  → session identity + authorizeCaseAccess
  → read-only case / Truth Graph / readiness projections
  → draft-only evidence request (requiresHumanApproval=true)

Authoritative writes stay outside WebMCP:
evidence ingestion · deterministic policy · human approval

Gemini (when configured) may propose analysis as an append-only AgentRun. It cannot assign provenance, transition the case, or execute external actions.

8. Run locally

Requirements: Node.js 24.13+ and npm 11.8+.

npm.cmd ci
npm.cmd run dev

Open http://localhost:3000/cases/RV-1028. Prefer localhost over 127.0.0.1 under Next.js 16.

Optional: copy .env.example → .env.local and set placeholders only. No credentials are required for the deterministic demo.

9. WebMCP testing

1. Use supported Chrome / ChatGPT in-app browser with WebMCP enabled.
2. Open RV-1028 and confirm the five tools are discovered.
3. Replay:
   • “Can RV-1028 be resolved yet?â€� → readiness NOT READY
   • “What is missing?â€� → gaps
   • “Show me why.â€� → Truth Graph
   • “Prepare the request.â€� → draft; human approval required

Local Chrome example flags: --enable-features=WebMCPTesting / --enable-blink-features=WebMCP.

10. Prior work vs WebMCP challenge work

See docs/webmcp/competition-delta.md.

	Commit
Pre-WebMCP baseline	d98271271662faac34faa77ee5389381e52fdc20
WebMCP implementation	94460d01e1341bb376c9d66396635e39fba97d3c

This public repository is a sanitized competition snapshot derived from a private development repository. It does not include private git history.

11. Security / human-control model

• Agent tools are read/draft only
• Same-origin invoke + case authorization
• No browser Firestore authority
• Consequential actions require human approval
• Demo data is synthetic RV-1028 only

12. Demo URL

https://resolvia-webmcp-competition-3pzbr52hla-et.a.run.app

LIVE_URL = https://resolvia-webmcp-competition-3pzbr52hla-et.a.run.app

License

MIT License — see LICENSE.

Copyright (c) 2026 Resolvia contributors.