Sit down. Type a buyer-language question. Every click runs against signed evidence.

Finding Dashboard · Cases · Findings · Lineage · Responsibility · Verification — for regulators, insurers, auditors, PE partners, and CISOs. Open the Workspace →

Evidence Cases · Three case types · one engine

Three different investigations. Three different buyers. Three different questions. Same generator.

Each case is an Independent Regulatory Submission. Each ships a submission.tar.gz a regulator can consume without H33. Each leads with a buyer-facing question — what caused this loss? · what changed without authorization? · what did we just buy? — answered by composing existing primitives (replay · responsibility · lineage · reality gaps · inherited risk) into a single signed package. No vendor required at any step.

The promise that runs this surface · Phase C

Can a regulator consume this submission without us?

Every submission ships eight entries: case_summary.json · verification_report.json · reconstruction_report.json · lineage_report.json · responsibility_report.json · finding_report.json · the embedded evidence_package.tar.gz · verifier_runbook.md. Open submission → verify package → review findings → drill into evidence. Regulators investigate findings — the evidence supports them.

Generated: each case is auto-produced from a case_descriptor.json by the h33-evidence-case-generator binary. Given (tenant_id, claim_id, asset_id, decision_id), the generator composes the case file + the canonical submission.tar.gz directly from canonical events. Same descriptor + same database state = same submission. The evidence_package.tar.gz becomes one component of the submission — not the submission itself.

Case File · Insurance Claim Defense · Insurer / Reinsurer

Claim #84711 — $4.2M credit-default loss · what caused this loss?

Claim IDclaim_84711

Assetloan_84711_borrower_principal

Loss$4,200,000 · credit_default

StatusFiled 2026-03-20

Primary decisiondecision_loan_84711_approval

Tenant dissolution2031-04-01 (dissolved)

Replay T2035-06-01 (~4 yrs post-death)

state_id at Te72d3c0e…3bee2

✓ Package Integrity ✓ Independent Replay ✓ Evidence Package Included ✓ Responsibility Chain Present ✓ Lineage Present ✓ Vendor Independence Verdict: Reviewable

Case File · Reality Gap Investigation · Auditor / CISO / Internal Audit

Reality Gap #001 — 12 gaps across 4 classes · what changed without authorization?

Investigation typeReality Gap

Total gaps12

Shadow Authority3

Undocumented Delegation4

Policy Bypass2

Responsibility Drift3

Replay T2026-09-21

state_id at Tbe607447…2376b7

✓ Package Integrity ✓ Independent Replay ✓ Evidence Package Included ✓ Responsibility Chain Present ✓ Lineage Present ✓ Vendor Independence Verdict: Reviewable

Case File · Inherited Risk Review · PE / M&A · Diligence

Acquisition #001 — $18.4M inherited risk exposure · what did we just buy?

Acquisitionacquisition_001

Inheritance policyfull

Aggregate exposure$18.4M

Inherited consequences5

Losses2 · $7.7M

Lawsuits1 · $3.2M

Regulatory actions1 · $7.0M

Operational failures1 · $450K

Replay T2027-12-31

state_id at T9581cc7b…918add

✓ Package Integrity ✓ Independent Replay ✓ Evidence Package Included ✓ Responsibility Chain Present ✓ Lineage Present ✓ Vendor Independence Verdict: Reviewable

Why three cases · proof that this is a system, not a story

Three tenants. Three questions. Three buyers. Three finding shapes. One engine.

The same h33-evidence-case-generator binary produced all three submissions from canonical events with no new schema. Case #1 — claim_84711 — leads with "what caused this loss?" · substantiated financial finding · insurer defense buyer. Case #2 — reality-gap-001 — leads with "what changed without authorization?" · four open governance findings · auditor/CISO buyer. Case #3 — acquisition-001 — leads with "what did we just buy?" · $18.4M inherited-risk finding + five per-consequence findings · PE/M&A buyer. The next move is the API — POST /generate-case. After that, every new case is data, not hand-curation.