Illustrative walkthrough using seeded data. Product capabilities are marked Built or Planned.
NOTARY AI
Screen 03[BUILT]

Evidence Readiness

Can Company XYZ account for the LoanCore lending decisions today, and is the evidence sufficient to verify candidate 3.2.0? Readiness distinguishes evidence completeness from severity, verification result, and the release decision. Assurance Signals appear below as an advisory planned capability — they never block a release or declare an incident. See .
Company XYZ decision accountability · LoanCore Underwriting 3.2.0Lending · Applicant A-1027— Representative lending boundary case
  1. ACCOUNT1–3→
  2. VERIFY4–7→
  3. DEFEND8–9
  1. ✓Connect→
  2. ✓Landscape→
  3. 3Readiness→
  4. 4Scenarios→
  5. 5Authorize→
  6. 6Verify→
  7. 7Release→
  8. 8Package→
  9. 9Hand-off
ACCOUNTScreen 03
Question
Which Company XYZ decisions are ready for review, verification, or external response?
Answer
LoanCore lending decisions are substantially ready; several other families remain partial because authority, policy, context, or outcome evidence is missing.
Output
A prioritized work queue for closing evidence gaps before they become audit, release, or dispute blockers.

Prioritized work queue — close these before they block audit, release, or dispute

Ranked opportunities to close evidence gaps, drawn from the same family readiness data shown on the Decision Landscape.

  1. 01Repeated decisions lacking policy bindings
  2. 02Decisions using a model version without an approved Scenario suite
  3. 03High-impact decisions missing human-review evidence
  4. 04Increasing numbers of decisions that cannot be replayed
  5. 05A policy change not yet tested against historical decisions
  6. 06Repeated overrides suggesting a missing Scenario
  7. 07A cohort with insufficient evidence coverage

Evidence Readiness is not

Business severity, model performance, a verification result, a compliance determination, or a release approval. It measures only whether a decision has enough preserved evidence to be reviewed, verified, or defended when asked — nothing more.

Assurance Signals (planned, advisory)

[PLANNED]seeded illustration

Signals help prioritize where to inspect. They do not establish that a decision was wrong. Illustrative opportunities: Repeated decisions lacking policy bindings · Decisions using a model version without an approved Scenario suite · High-impact decisions missing human-review evidence · Increasing numbers of decisions that cannot be replayed · A policy change not yet tested against historical decisions · Repeated overrides suggesting a missing Scenario · A cohort with insufficient evidence coverage.

Lending

142 lending denials without a policy binding

Rule failure · A rigid deterministic gate denied a case the underlying model would have passed.

Advisory
cohort_size
142
window
last 90 days
method
Denials at 640–699 with DTI<30 and no delinquencies
evidence_coverage
86%
suggested_action
Preserve evidence for the next occurrence
supporting_evidence
cassette sha256:9f2c…a17b
limitations
Excludes decisions missing bureau pull; advisory only.
Advisory — not an IncidentOpen candidate →
Healthcare Prior-Auth

96 auto-determinations missing human-review evidence

Rule failure · A rigid deterministic gate denied a case the underlying model would have passed.

Advisory
cohort_size
96
window
last 30 days
method
Auto-denials where high_risk_note=true and no clinician touch
evidence_coverage
72%
suggested_action
Preserve evidence for the next occurrence
supporting_evidence
cassette sha256:6b41…c02e
limitations
Does not include appeals data; advisory only.
Advisory — not an IncidentOpen candidate →
Customer Support

37 policy statements with no retrieval binding

Model hallucination · The LLM produced a policy statement with zero retrieved supporting documents.

Advisory
cohort_size
37
window
last 60 days
method
Bot responses containing policy claims with 0 retrieved documents
evidence_coverage
64%
suggested_action
Preserve evidence for the next occurrence
supporting_evidence
cassette sha256:d81a…4471
limitations
Detection limited to responses with structured retrieval logs.
Advisory — not an IncidentOpen candidate →
Contact Center

218 conversations with escalation asks and no agent join

Classifier gap · An intent classifier used keyword-only matching and missed vernacular escalation phrasing.

Advisory
cohort_size
218
window
last 30 days
method
Conversations with ≥2 handoff phrases and no agent join event
evidence_coverage
58%
suggested_action
Preserve evidence for the next occurrence
supporting_evidence
cassette sha256:2a70…9c33
limitations
Advisory; no CSAT causality established.
Advisory — not an IncidentOpen candidate →
Hiring

54 screening rejections weighted by an age-proxy feature

Biased feature · A proxy feature (age via graduation year) dominated the screening score.

Advisory
cohort_size
54
window
last 90 days
method
Screen scores where grad_year features contributed >30% of decision weight
evidence_coverage
31%
suggested_action
Preserve evidence for the next occurrence
supporting_evidence
cassette sha256:c14e…f0b8
limitations
Source not fully connected — coverage partial; advisory only.
Advisory — not an IncidentOpen candidate →