EVIDENCE, CONNECTED.
One record.
Every dependency.
See what an application check relies on, what changed,
and exactly what needs a second look.
Evidence map
Source records → purpose-specific checks → review status
Application checks
Select a program to inspect its evidence and consent.
CONTROLLED EXPERIMENT
What happens if…
Run a repeatable change against the same evidence. The baseline stays intact until you apply a scenario.
Threshold and freshness changes are comparison-only policy versions. Other scenarios can be applied to the workspace.
BEFORE / AFTER
Understand the change.
Choose a scenario and run a comparison to see which checks move, and why.
Activity trail
Session-only observations. This log is editable browser memory, not a tamper-proof audit system.
BUILT WITH A SPECIFIC QUESTION
What breaks when evidence changes?
YC’s “Plaid for Government Benefits” request describes reusable, consent-based records. Reusable verification already exists. ClaimWeave explores a smaller developer-facing question: can a reviewer inspect evidence freshness, conflicting claims, purpose-specific consent, and policy version changes in one reproducible workspace?
Closest work found
| Existing work | Published focus | Our prototype’s focus |
|---|---|---|
| Reclaim Protocol | Reusable source-backed records across programs | Inspect failure states using synthetic evidence |
| CMS Emmy | Consent-based verification for benefit workflows | Portable, deterministic policy regression examples |
| Student Verification Suite | Agent-driven student verification | Visible dependency checks and consent withdrawal |
Honest boundaries
No live integrations, identity verification, OCR, runtime AI, cryptographic proofs, or actual government rules. Source names and records are fabricated. Import is for synthetic JSON only. Export is a local download, not a submission. The scenario engine never makes a real eligibility decision. Withdrawing consent blocks future in-app exports; it cannot retract a file already downloaded.
Attribution
New implementation created November 8, 2026 with OpenAI Codex coding assistance. Research sources were consulted for comparison; their code was not copied. No claim of patent novelty, independent security review, or YC endorsement.
Read YC’s Requests for Startups ↗