Decision support. ChironAI drafts the AOE/COE causation and apportionment analysis for the treating physician. The physician evaluates the reasoning, edits, and signs the determination. ChironAI does not make a regulatory clearance claim; see Disclosures.
AOE/COE causation and apportionment are where a California WC claim is won or lost. Labor Code §3600 asks whether the injury arose out of and in the course of employment; §§4663–4664 ask how much of the disability is industrial. Both questions get torn apart at QME and WCAB when the reasoning is thin. ChironAI™ OM runs both with Chain-of-Verification self-critique, multi-lens Digital-Employee consultation, and case law cited from what was actually retrieved — never invented. The physician reviews and signs a narrative built to hold up.
Apportionment computed the way a WCJ actually checks it — three prongs, audited, not asserted.
Must review before final
Decision-support output. Clinician review and attestation required before this content is signed into the chart.
Injury
42-year-old warehouse worker reports acute low back pain after lifting a 65-lb box from floor level on 2025-11-04. No prior lumbar imaging on file. History of mechanical low back pain managed in primary care 2019–2022 with periods of full work capacity in between.
Causation conclusion
Rationale
Before a causation analysis reaches the treating physician, Eve-Healthcare™ F5/reasoner runs the draft back through a verification pass: an initial determination, a set of adversarial self-critique questions, and a final pass that resolves what the verification surfaced. This runs on the same case shown above.
Initial
Draft conclusion: AOE/COE — yes, reasoned from the mechanism of injury, the imaging comparison, and the absence of an alternative precipitating event.
Verification questions
“Did I invent anything not in the record?”
Every rationale line traces to a source field: the 2025-11-04 incident note, the 2025-11-12 MRI, and the 2019–2022 primary-care episodes. No statement lacks a source.
“What would opposing counsel argue?”
That the 2019–2022 low-back episodes establish a pre-existing condition warranting apportionment. Rebuttal on file: those episodes were transient, resolved to full work capacity each time, and no imaging before 2025-11-12 showed the L4–L5 herniation now present.
“Would this survive cross-examination?”
The two hardest facts — no alternative non-industrial precipitating event in the 90 days prior, and a herniation absent on the only prior lumbar imaging — are both explicit findings with dates and sources, not inferences.
Final
Conclusion confirmed at likely confidence, with the counter-argument and its rebuttal now carried inside the analysis rather than left for opposing counsel to raise first. The verification pass itself is logged to the audit trail.
Chiron leads the medical determination. The system then decides, case by case, which specialist Digital Employees the record actually needs — Justine for apportionment and legal-evidence framing, Issac on every case for denial risk, Theo when Chiron’s own confidence drops below the review threshold, Eli only when a billing dispute is present. The routing is itself part of the reasoning.
Consulted because the case raises apportionment.
Consulted because lead confidence is below the 70% review threshold.
Consulted on every case.
Consulted only when a billing dispute is present - not triggered here.
Causation-based apportionment under §4663 asks what share of the current disability is industrial versus pre-existing, grounded in the Escobedo substantial-evidence standard — not a default split.
Must review before final
Decision-support output. Clinician review and attestation required before this content is signed into the chart.
15% apportioned to pre-existing degenerative disc disease documented on prior imaging; 85% to the industrial lifting injury, per the treating-physician analysis.
Labor Code §4663 (causation-based apportionment); Escobedo substantial-evidence standard.
When a prior permanent-disability award does exist, the §4664(b) offset does not run as a model guess. It runs as a deterministic engine that checks the same three prongs a workers’ compensation judge checks — final award, body-part overlap, PD-to-PD comparison — and writes its derivation to the case record. A second, illustrative case demonstrates the mechanism where an offset actually applies.
Must review before final
Decision-support output. Clinician review and attestation required before this content is signed into the chart.
Is the prior permanent-disability award final — no pending petition for reconsideration or appeal?
Prior award CT-2022-04410 issued 2023-08-02. No petition for reconsideration on file; the appeal window closed with no action taken. Final.
Does the prior award cover the same body part as the current claim?
Prior award rated the right shoulder (rotator-cuff pathology). Current claim: a right shoulder labral tear from a fall. Same body part — overlap confirmed.
Per Brodie, is the offset computed rating-to-rating — not dollar-to-dollar?
Current PD rating 22% − prior award 12% = 10% final apportioned PD. Percentage subtracted from percentage, per Brodie — never a dollar figure.
Final apportioned PD
10%
Engine: apportionment-4664-offset v2.3 · deterministic · same inputs reproduce the same output on replay.
Audit trace
Substantial-evidence standard
Every causation and apportionment conclusion has to rest on substantial medical evidence — specific, reasoned fact, not a bare percentage or a hunch. This is the standard Chiron’s rationale is written to meet, line by line.
§4664(b) offset mechanics
The prior-award offset is computed rating-to-rating — PD percentage against PD percentage — never dollar-to-dollar. This is exactly the arithmetic Prong 3 of the deterministic offset engine runs.
Compensable-consequence doctrine
A new impairment caused by treatment for the industrial injury itself is not apportionable, even where the condition being treated had a non-industrial component. The engine flags compensable-consequence findings and blocks apportionment against them.
AMA Guides deviation
A physician may reference other chapters or methods within the AMA Guides when the assigned table does not accurately reflect the impairment — with the deviation rationale carried in the report, never a silent substitution.
Citation validation
Every case-law citation, statutory citation, and record citation in a causation or apportionment analysis is checked against what was actually retrieved for that case. A citation the system cannot trace back to a real extraction it was given is not flagged for review — it is deleted before the draft reaches the physician. The same guardrail rejects a fabricated ICD-10 code or an apportionment citation without a retrieval behind it.
Before the causation-and-apportionment draft ever reaches the treating physician, Justine — the legal-evidence specialist Digital Employee — runs her own defensibility audit against the analysis Chiron just produced, scoring the dimensions a physician skimming for medical accuracy would not necessarily catch. The score rides alongside the draft; it never replaces the physician’s review, and nothing is signed without it.
Every conclusion traces to a specific source field. Meets the Escobedo substantial-evidence standard.
Every case-law and record citation checked against what was actually retrieved for this case; nothing unverifiable reached this draft.
The strongest opposing-counsel argument was raised and rebutted with dated evidence during Chain-of-Verification, not left for deposition to surface first.
§4663/§4664 math reproducible on replay; no compensable-consequence finding was apportioned against.
Every causation and apportionment analysis is versioned and audit-trailed. Amendments are recorded as new versions; the original analysis stays verifiable. The audit chain stamps each step — the Chain-of-Verification pass, the consultation record, the §4664 offset computation — with the actor, timestamp, and HMAC signature.
The must-review-before-final gate at the data layer means no analysis is treated as authoritative until the treating physician signs it. The signature is cryptographically verifiable via SHA-256 hash at signature time, and can be confirmed against a public verification endpoint — anyone holding the document can prove it has not been altered since attestation.
When the analysis ends up cited in deposition, at QME, or before the WCAB, the trail back to the underlying consultation, the underlying source data, and the underlying reasoning chain stays intact.