kolm / insure
Every kolm-compiled .kolm artifact emits a tamper-evident receipt for every inference: CID of the model that ran, hash of the input, hash of the output, K-score gate decision, HMAC chain. When a claim hits, you have the audit trail.
Without a receipt, the carrier argues: was it the model, or was it the adjuster? With a kolm receipt, the underwriter sees the artifact CID, the input hash, the output hash, and the K-score at decision time. If K-score was below the carrier's stated gate, the carrier was operating an unsanctioned model. If above, the model performed as documented.
receipt.k_score: 0.91 / gate: 0.95 / decision: gate_bypassed_at_runtime
The artifact's manifest declares its verifier (constrained citation lookup). The receipt records whether the verifier was invoked and what it returned. If the verifier was bypassed (e.g., a fall-through path) the receipt names which code path. The carrier sees exactly where the chain broke.
receipt.verifier_invoked: false / fallthrough: prompt_too_long_truncated
The receipt anchors the artifact CID + redaction-pass count at the time of failure. The underwriter compares the failing artifact against the carrier's stated production CID. If they differ, the production stack drifted from the certified model — a finding the carrier owes.
declared_cid: cidv1:sha256:1bcf.. / actual_cid: cidv1:sha256:9a3f.. / drift: true
{
"artifact_cid": "cidv1:sha256:9a3f4e1b..",
"input_sha": "sha256:8c2d1fa0..",
"output_sha": "sha256:4e1a90c3..",
"k_score": 0.982,
"gate": 0.92,
"verifier_ok": true,
"ts": "2026-05-15T13:42:08Z",
"issuer_pubkey": "kolm-issuer-2026q2",
"hmac": "b7c41e87.."
}
Replayable months later. The carrier exports a JSONL of receipts; the underwriter spot-checks any of them against the artifact CID and re-runs the same input through their own kolm-verify install.
Your insureds use kolm Cloud or self-host. You request the receipt JSONL during claim review. No platform fee from us.
Hosted receipt vault scoped to your book. Bulk export, replay against historic artifacts, named-account SLA.
White-label deployment inside your insureds' kolm Cloud tenants. Co-branded verifier UI, custom K-score gates.
We do not underwrite the policy. We do not opine on whether a model is safe to deploy. We do not certify a carrier's risk model. We are the audit trail.
Your underwriters do the underwriting. Our receipt is the artifact they cite when explaining a claim outcome to a regulator, an arbitrator, or a court.
underwriting@kolm.ai for carrier conversations.