Trust Intelligence Center
Validation proved the AI understands the work. This measures whether it can assist humans without reducing trust. Every value is observed evidence — unmeasured steps stay unknown, and nothing here expands autonomy on its own.
AI trust score
+0.04 vs prior 30d
Score reflects observed behavior only. Coverage below 100% means some workflow steps are unmeasured and remain Unknown.
Trust score inputs
weight 0.18 · 288 comparisons
tc.expert_agreement.v1
weight 0.16 · 412 shadow predictions
tc.shadow_accuracy.v1
weight 0.15 · 1,206 suggestions
review log (append-only)
weight 0.12 · 318 edits
review log (append-only)
weight 0.12 · 94 replayed files
tc.validation_replay.v1
weight 0.11 · 38 playbook revisions / 90d
tc.knowledge_stability.v1
weight 0.09 · 241 exception episodes
tc.exception_library.v1
weight 0.07 · 54 calibration bins
calibration study
Delegation readiness — every workflow step
| step | workflow | obs | val | approval | replay | exc stability | playbook | override | conf | verdict |
|---|---|---|---|---|---|---|---|---|---|---|
| Escrow opening packet | Traditional purchase | 94% | 91% | 93% | 90% | 89% | 92% | 6% | 90% | READY FOR AUTONOMY REVIEW |
| Payoff request | Traditional purchase | 90% | 88% | 91% | 87% | 87% | 88% | 8% | 87% | READY FOR AUTONOMY REVIEW |
| Title commitment review | All | 86% | 79% | 71% | 74% | 66% | 78% | 22% | 76% | HUMAN APPROVAL REQUIRED |
| Curative escalation | All | 71% | 62% | 58% | 60% | 49% | 61% | 34% | 63% | ASSIST ONLY |
| Assignment fee disclosure | Wholesale assignment | 68% | 55% | 52% | 57% | 51% | 54% | 38% | 58% | ASSIST ONLY |
| Double-close funding sequence | Double close | 52% | 41% | 44% | 40% | 38% | 42% | 46% | 47% | OBSERVE ONLY |
| Subject-to lender notification | Subject-to | 44% | 31% | unknown | unknown | unknown | 36% | unknown | 34% | OBSERVE ONLY |
| Novation consent tracking | Novation | unknown | unknown | unknown | unknown | unknown | unknown | unknown | unknown | OBSERVE ONLY |
| Signing scheduling | All | 92% | 84% | 86% | 83% | 80% | 85% | 12% | 84% | HUMAN APPROVAL REQUIRED |
| Funds release instruction | All | 88% | 82% | 80% | 79% | 78% | 81% | 14% | 82% | HUMAN APPROVAL REQUIRED |
Funds movement, contract terms, and legal advice are permanent human-only gates regardless of verdict. Promotion out of any tier requires explicit human approval routed through Assistant.
Human confidence analytics
612
accepted
318
edited
141
rejected
135
ignored
1206 suggestions · median review 3.4m · Acceptance +6pts over 60d · confidence 79%
reason for override
Overrides are treated as signal, never as user error. No coordinator is scored or ranked on override behavior.
Knowledge stability
Workflow trust map
Open & order
Title & curative
Funding & close
Expert consensus
consensus 91% · 6 experts · confidence 86%
variation: 1 coordinator orders in parallel
consensus 88% · 5 experts · confidence 84%
variation: range 3–7 days
consensus 62% · 6 experts · confidence 66%
variation: day 3 to day 7
consensus 54% · 4 experts · confidence 58%
variation: 2 wait for title
consensus 100% · 6 experts · confidence 94%
variation: none
consensus 47% · 4 experts · confidence 52%
variation: at contract vs. at title order
consensus unknown · 2 experts · confidence unknown
variation: sample too small
Consensus is descriptive. No single coordinator's practice becomes the organizational standard without human approval through Assistant.
Operational confidence
Transaction types
Playbooks
Document types
Exception categories
Coordinator roles
Recommendations only. Low-coverage dimensions are reported as unknown rather than estimated.
Executive readiness dashboard
most trusted workflows
148 files · 0.06 override
121 files · 0.08 override
163 files · 0.07 override
least trusted workflows
22 files · 0.46 override
38 files · 0.38 override
74 files · 0.34 override
knowledge gaps
Novation consent tracking
3 observed · needs ≥12 observations across ≥2 coordinators
Trust / probate document handling
5 observed · needs ≥12 observations, ≥3 files
Recording confirmation variance by county
9 observed · needs ≥20 observations
Commercial purchase close-out
2 observed · needs ≥12 observations
high override areas
Expert consensus is itself split (0.62)
Conflicting practice across 4 experts
Office-specific endorsements missing
emerging best practices
Pre-clear HOA estoppel at acceptance
4 coordinators · 21 files · confidence 58%
Payoff good-through +4 as default
5 coordinators · 63 files · confidence 84%
Batch title follow-ups at 9am local
3 coordinators · 44 files · confidence 61%
delegation opportunities
Advance to autonomy review
0.92 trust · 0.06 override · 148 files
Advance to autonomy review
0.89 trust · 0.08 override · 121 files
Keep approval-gated; expand draft scope
0.86 trust · 0.12 override
AI assistance opportunities
Escrow opening packet assembly + send
148 files · 0.93 approval · confidence 90%
Payoff request generation with good-through +4
121 files · 0.91 approval · confidence 87%
Add office-specific endorsement checklist to commitment review
41 repeated edits · confidence 77%
Curative escalation day threshold — experts disagree
consensus 0.62 across 6 experts · confidence 66%
Novation consent tracking
3 observations · confidence unknown
Double-close funding sequence
22 files · validation 0.41 · confidence 47%
Opportunities are routed to Assistant for orchestration. TC Pros never executes, publishes, or promotes anything from this list on its own.
Validation replay — expert vs AI
| file | step | expert | ai | disposition | outcome | conf |
|---|---|---|---|---|---|---|
| TX-2041 | Commitment review | Flagged B-II #4 + endorsement T-19 | Flagged B-II #4 | Corrected | Closed on time | 77% |
| TX-2088 | Payoff request | Good-through +4 | Good-through +4 | Approved | No re-request | 91% |
| TX-2103 | Curative escalation | Escalated day 3 | Recommended day 5 | Rejected | 4-day delay avoided | 62% |
| TX-2117 | Signing scheduling | Mobile notary | Office signing | Corrected | Closed on time | 70% |
| TX-2129 | Escrow opening | Standard packet | Standard packet | Approved | Escrow opened same day | 93% |
Trust & accuracy trends
Trust trend (90d)
0.66 → 0.73
confidence 79%
Accuracy trend (90d)
0.71 → 0.79
confidence 82%
Correction trend (90d)
0.34 → 0.26
confidence 75%
Replay consistency
0.78 → 0.81
confidence 80%
Learning candidates derived from replay are submitted to Assistant, which owns publication.
Organization trust benchmarking (anonymous)
| org | validation coverage | trust | knowledge stability | approval | override | readiness |
|---|---|---|---|---|---|---|
| org_a | 82% | 78% | 74% | 79% | 15% | HUMAN APPROVAL REQUIRED |
| org_b | 69% | 71% | 68% | 73% | 21% | HUMAN APPROVAL REQUIRED |
| org_c | 54% | 60% | 55% | 64% | 29% | ASSIST ONLY |
| org_d | 31% | unknown | 40% | 51% | 36% | OBSERVE ONLY |
Organizations are anonymized and never re-identifiable. Comparisons are aggregate-only and advisory.
Executive scorecards
| scope | subject | observed evidence | coverage | validation | trust | delegation | confidence |
|---|---|---|---|---|---|---|---|
| Organization | org_a | 148 files | 82% | 80% | 78% | HUMAN APPROVAL REQUIRED | 83% |
| Organization | org_d | 27 files | 31% | 29% | unknown | OBSERVE ONLY | unknown |
| Playbook | Traditional close v4 | 163 files | 90% | 87% | 86% | READY FOR AUTONOMY REVIEW | 88% |
| Playbook | Assignment close v2 | 38 files | 69% | 60% | 58% | ASSIST ONLY | 62% |
| Workflow type | Document collection | 412 episodes | 88% | 84% | 85% | READY FOR AUTONOMY REVIEW | 86% |
| Workflow type | Curative | 74 episodes | 66% | 62% | 58% | ASSIST ONLY | 63% |
| Transaction type | Traditional purchase | 163 files | 92% | 88% | 87% | HUMAN APPROVAL REQUIRED | 88% |
| Transaction type | Double close | 22 files | 58% | 41% | 44% | OBSERVE ONLY | 47% |
| Coordinator team | Team North (anon) | 96 files | 79% | 76% | 75% | HUMAN APPROVAL REQUIRED | 78% |
| Coordinator team | Team West (anon) | 31 files | 42% | 38% | unknown | OBSERVE ONLY | unknown |
Every scorecard exposes observed evidence, coverage, and confidence. Unknown remains unknown.
Cross-app consumption
Attribute pre-contract delays without reading CRM rows
Published contract only
Align transaction start with deal acceptance
Published contract only
Confirm outbound delivery state for follow-ups
Published contract only
Confirm funding/disbursement outcome events
Published contract only
TC Pros never queries another application's internal tables. Assistant resolves all orchestration between apps.
Learning candidates → Assistant
Steps with override <0.10 across ≥100 files sustain trust ≥0.85
3 steps, 432 files · route Assistant
Replay accuracy is the strongest predictor of post-promotion stability
94 replayed files · route Assistant
Conflicting expert practice predicts elevated override within 30d
2 practices, 112 files · route Assistant
Office-specific requirements drive 30% of all edits
96 of 318 edits · route Assistant
Playbook revision rate falling while coverage rises
90d window · route Assistant
Confidence drops sharply below 25 observed files per step
31 steps · route Assistant
Governance
TC Pros owns
TC Pros does not own
- · All trust and readiness output is recommendation-only.
- · No workflow advances autonomy without explicit human approval.
- · Unknown remains unknown — unmeasured steps are never scored.
- · Coordinators are never penalized or ranked for overriding AI.
- · Organizational comparisons stay anonymous and non-identifiable.
- · Assistant is the sole orchestration and learning-publication layer.