Independent Colorado Insurance Review Record

We compared a defined set of Colorado insurance filings and published files with what was publicly observable after filing. Some aligned. Some did not. This page shows exactly what was checked, what remains unresolved, and what the Division could verify next.

• Thirteen of the 128 rate-file links on the Division’s own January 2026 submission tab have been desk-checked — all thirteen returned HTTP 403 on every dated request since July 8, while the sheet lists them as valid through 2027. the availability record →
• The other 115 remain outside this point-in-time review. We have not identified a complete, single-vintage, dated review of the remaining 115 — from this record or elsewhere. A gap in what has been observed, never a claim about anyone’s compliance.
• Sixteen of the twenty Colorado Option plans filed in SERFF match the rate files their carriers actually published; four are filed but not observed — each stated with its SERFF anchor, reproducible from the underlying public record. the CO Option packet →
• Nothing here determines compliance; the Division alone does that. Postures on this record are coverage states of the evidence, never findings about any carrier.

The important question is not whether one broken link matters — it is whether the Division can know, on a dated basis, what was filed, what remained observable, and what changed.

Together, these findings show that the same public filing estate can be turned into a dated exception record: what aligned, what did not, what changed, and what still needs a reviewer’s judgment.

The register publishes 128 rate-file URLs; this record has desk-checked 13; 115 remain outside the current review. The natural next question is whether the same pattern exists across the complete register — and, separately, whether filed-vs-observed differences persist across the PY2027 population the Division is currently reviewing.

Walk through the current record with us — pick one finding, and every observation behind it can be recomputed from the public filing while we are on the call.

Walk through this record →

Opens a pre-addressed email. Nothing is transmitted until you send it. Do not include PHI, privileged matter details, or sensitive personal information.

Everything above is meant to be understood quickly. Everything below exists so you do not have to trust us.

Six issuers publish the same mandated product in incompatible formats — this record makes 7 of 11 books comparable, and says plainly why the rest are not.

Counts differ by layer, by design: the negotiated-rate corpus spans every filing carrier; directory-accuracy cross-checks are materialized for a subset of them so far; the DOI transparency list holds 12 entities as 11 evidence books; the Colorado Option product has 6 issuers. Each measures a different thing — the record keeps them distinct rather than collapsing them into one number.

materialized — rate dataset 2026-07-09 · record registry 2026-07-10 · dated computations 2026-07-13 (same-provider cross-issuer · per-issuer roster slice) · current record, version-ready architecture

cadence receipts — 48 state links receipted per-URL and dated (Anthem 32 · UHC family 16), all 48 HTTP 403 at the last full-register test 2026-08-25 · the thirteen UnitedHealthcare Insurance Company + Surest links carry ten dated tests each (2026-07-08 → 2026-08-31) · day-one retest committed at the July publication · every recompute locked against the sealed 2026-07-09 baseline before a new test runs

cross-issuer grid + identification method + rate checks vs regulation: the Colorado Option packet →

additional carriers assemble from the same registry — one entry per carrier, no new build.

After the Network Integrity Record withdrawal, no edition of a finding-bearing record publishes here without passing Gate 6 — eight preconditions, run per edition, ending in a recorded human determination and a served-content parity proof. The status below is derived from the append-only edition-lifecycle log that /verify itself is exported from; this page cannot assert a pass the log has not declared.

1. M1 — claim–evidence contract replay. Every finding's rendered sentence is re-checked against its own payload AND by direct re-read of the source tables its claimed absences quantify over. Zero violations, at both layers.
2. M2 — governed templates + the slot contract. Every reader-facing sentence comes verbatim from a versioned template; the class token printed for a finding must be derivable from that finding's own evidence. Hand-written subject prose is a refusal.
3. M3 — severity by rubric, never by hand. The displayed severity must equal what a versioned, content-hashed rubric derives from the finding's own payload. Missing inputs are an explicit abstention — never a defaulted tier. Manual overrides do not exist.
4. M4 — audience permission, then posture. Every load-bearing source must permit the intended audience; the most restrictive source governs, and no disclaimer can cure a prohibited use. Where publication is review-only, the prescribed posture wording must render verbatim.
5. M5 — staged lifecycle + ledger registration. The edition must be declared in the append-only lifecycle log (draft → mechanically admissible → human admitted → candidate sealed → canary served → production pinned) and registered in /verify. An edition invisible to the log cannot proceed.
6. Human determination, bound to the exact candidate. A named reviewer records three separate answers — claim admissible, severity admissible, audience admissible — against the candidate's content digest. Any drift in evidence, templates, rubric, or audience policy invalidates the admission automatically.
7. Gate 6B — served-content parity at the canary. The candidate is served at a nonpublic preview origin first; the served semantic content envelope must be digest-identical to the admitted candidate before the public pin may advance.
8. Production parity, before and after the pin. The same parity check runs against production as a precondition of declaring the pin, and again after activation. Any mismatch blocks or withdraws the pin.

# Gate 6 — Edition Five, public aggregate edition

- edition_id: `ghost-provider-audit-batch:5518bb79b3955c751391dd22236ff2fac0b9b641dc57c7426d99d591ce05273a:e5:public-aggregate`
- audit_batch_id (evidence root): `ghost-provider-audit-batch:5518bb79b3955c751391dd22236ff2fac0b9b641dc57c7426d99d591ce05273a`
- audience: `public_aggregate`
- method versions: policy `publication-release-gates/v2` · severity rubric `gate6-severity-rubric/2.3.0` (content sha256 `b6f1bdb9dd152e975afb92d20c2ad4d7b25097e370830a5e2eac40ea4a538293`) · audience policy `gate6-audience-policy/2.0.0`

## What this edition is

Edition Five passed mechanical Gate 6 review for the public-aggregate audience and was submitted for human determination. The public edition contains no subject-level finding, provider identifier, adverse class, or severity. A separate review-only edition is governed independently for an authorized regulatory audience.

Edition Five of the Network Integrity Record was split into two independently admitted audience editions under the adopted Gate 6 ceremony review (decision (b)). A pass for one edition never implies a pass for the other, and this page speaks only for the public aggregate edition.

## Process history

An earlier Edition Five candidate was mechanically admitted under the wrong audience and was invalidated before any public serving (reason: audience mismatch and post-admission artifact change). The invalidation record is preserved in the append-only lifecycle log; the replacement editions were evaluated from the start of the sequence. The invalidation is part of the evidence that the gate works.

## Small-cell rule

Subject-level class and severity counts below 11 are suppressed or reported as "fewer than 11" on public aggregate surfaces — a count of one never becomes a public adverse descriptor.

## Source-derived counts: deliberately not included

Source-derived aggregate counts are not included in this edition: the Colorado licensing source's machine-readable manifest grants review-only publication and does not affirm a public-aggregate audience, so this panel limits itself to TrustNav lifecycle and gate metadata. Permission never transfers between audience classes by assumption, and neither a disclaimer nor a template can cure a prohibited use.

## Non-conclusions

This edition does not establish anything about any provider, carrier, network, or person. Gate status describes TrustNav's own publication controls — which editions passed which mechanical checks and human determinations — never a determination about the subjects of the underlying evidence. Subject-level evidence is available only through the authorized review workspace, after its own independent admission.

Complaints and enforcement history (operator-gated source, not loaded) · essential community provider data (loader dormant) · APCD claims (walled source). A declared void is a statement about this record's coverage — absence of a loaded source is never evidence about any carrier.

The review chips above are the queue: every packet carries an append-only review log, and anything without a recorded state is open work. Queue views and assignments compose onto these same objects — same record, same proof.