Bounded two-source reconciliation exercise on an immutable fictional fixture. No backend, no network call, no analytics, no browser storage. Lab state and session events are not persisted automatically; choosing Export writes one JSON file to the operator's device. Theme preference is outside the lab data model. Every action is simulated and requires explicit confirmation.
- Fixture
- FID-A · v1.0.0
- Matter / claim
- DEMO-FID-BR-001 · C-017
- Session (ephemeral)
- S-R5PAQ-0
- SHA-256 checksum (canonical JSON)
- computing…
Source A — claimant filing
Source B — debtor ledger
Observation prompts — no legal conclusion is proposed
Select what you observe. These are prompts for professional judgement, not determinations.
Reviewer controls — bounded, no free text
Dry-run readiness
- OPENFixture schema valid
- OPENBoth sources opened
- OPENAll three observations selected (0/3)
- OPENUncertainty set
- OPENProfessional review required — checked
- OPENBounded next action confirmed
Copy the exact candidate SHA from the release board and verify it against that entry before exporting. A build cannot embed its own final SHA, so this binding is operator-verified (OPERATOR_VERIFIED_RELEASE_BOARD). This field is evidence metadata only and never enters the event recorder.
Terminal event not yet recorded.
Export is manual only and unlocks once exactly one TASK_COMPLETED exists and a valid candidate commit SHA is entered. Choosing Export writes one JSON file to this device. Allowlisted fields: session_id, fixture_id, fixture_version, fixture_checksum, task_id, event_type, step_id, uncertainty, abstention, professional_review, outcome, error_code, timestamp, elapsed_ms. Schema 1.0.0.
Session event trail — in memory only
No events recorded yet.
No private data feeds public search, another tenant, cross-client training or benchmarking. This prototype contains synthetic records only; it makes no claim of tenant isolation, production security, court integration, AI accuracy or real ingestion. It demonstrates UX and workflow only.