Glass Box Philosophy

Every Conclusion Is Citable. Every Calculation Is Auditable.

Every AI output in AdjudiCLAIMS shows where it came from, how it was calculated, and what statute governs it. We don't ask you to trust the AI — we show you exactly what the AI found, where it found it, and what the law says about it.

100%

AI outputs cite their source document

60+

Audit event types logged immutably

7 yrs

Immutable audit retention (LC 3762)

57

Always-present education entries

Core Principles

The Glass Box Standard

Six commitments that define what transparent AI looks like in a CA Workers' Compensation claims operation.

Every Output Cites Its Source

AI responses include document name, page number, and extracted text. Examiners always know where a fact came from — not just what the AI said.

Every Calculation Shows Its Formula

TD rate: (AWE × 2/3) per LC 4653. Penalty: 10% of overdue payment per LC 4650(c). The formula, the inputs, and the statutory basis are always visible.

Every Deadline Cites Its Statute

Not "14-day deadline" — "LC 4650 requires first TD payment within 14 days of the date of temporary disability. Failure triggers an automatic 10% penalty per LC 4650(c)."

Every AI Decision Is Auditable

60+ audit event types. Every document upload, AI response, deadline trigger, and user action is logged with timestamp, user ID, and action type. Retained for 7 years per LC 3762.

Every Classification Shows Confidence

AI classifications include confidence scores. Low-confidence results surface for human review — never silently auto-accepted.

Every Regulatory Rule Is Explained

57 always-present education entries explain why each rule exists and what happens when it's violated. The law is not a black box to users of AdjudiCLAIMS.

What Transparency Looks Like

Every Output Cited. Every Formula Visible.

Benefit CalculationLC 4653

TD rate: AWE = $1,200/week. Rate = 2/3 × $1,200 = $800/week. 2024 statutory max = $1,619.15/week. $800 is within the statutory range. First payment due by 03/29/2024 (14 days from employer knowledge date of 03/15/2024 per LC 4650).

Medical Record ExtractionQME Report, p. 14

Dr. Rivera assessed 12% WPI for the lumbar spine (L4-L5 disc herniation with right-sided radiculopathy). Work restrictions: no lifting over 25 lbs, no prolonged standing beyond 30 minutes. MMI declared as of 01/15/2024. [Source: QME Report 01/15/2024, page 14, paragraph 2]

Deadline Tracking10 CCR 2695.7(b); LC 5402(b)

Proof of claim received: 03/01/2024. Accept/deny deadline: 04/10/2024 (40 calendar days). 90-day presumption of compensability: 05/30/2024. Delay notification due every 30 days if determination not made. Current status: 28 days elapsed — 12 days remaining.

The Immutable Audit Log

Every Action Logged. Nothing Deleted. 7-Year Retention.

Every action in AdjudiCLAIMS is recorded in an immutable audit log. Nothing can be deleted or modified. Retained for 7 years per California LC § 3762.

Document Events

document_uploaded
document_classified
document_extraction_complete
document_access_denied

AI Chat Events

upl_zone_classification
upl_output_blocked
upl_disclaimer_injected
upl_adversarial_detected

Claims Events

claim_created
deadline_triggered
deadline_missed
benefit_calculated

Auth & Access

user_login
mfa_verified
session_expired
permission_denied

Plus 44+ additional audit event types. All events include: timestamp, user ID, session ID, IP address, action, resource ID, before/after state, and result.

Explicit Commitments

What We Explicitly Do Not Do

Transparency is also about being clear about the limits of the system.

We do not give legal advice

AdjudiCLAIMS cannot and does not provide legal advice, legal analysis, or legal conclusions to non-attorney users. Cal. Bus. & Prof. Code § 6125 is a hard constraint.

We do not make claims decisions

The AI surfaces information. The examiner makes all substantive decisions. The system is designed to support — not replace — human judgment.

We do not hide confidence scores

All AI outputs include confidence indicators. Low-confidence outputs are flagged for human review rather than presented as certain facts.

We do not log PHI content

PHI is never logged. The audit log records document IDs and action types — not medical content, diagnoses, or personal health information.

Request the Full AI Explainability Documentation