Observation Center

Phase 6

Before AI performs work, it must understand how experts work. Every connector is read-only, every action is append-only, and no automation is ever created directly from an observation.

0 automations created from observation

Sessions observed

trailing 90 days

Actions captured

1,284

append-only, evidence-linked

Patterns above threshold

6 of 7

≥5 observations, ≥2 coordinators

Candidates awaiting approval

routed to Assistant only

Automations created from observation

by design — human approval required

Live workflow sessions

812 Ridgeline Dr · TX-2041

OBS-1041 · M. Alvarez · Title Commitment Review

Live88% conf

Read-only CRM connector · started Today 08:14 · 27 actions · 2 questions

44 Marsh Ln · TX-2039

OBS-1042 · D. Okafor · Curative

Shadow74% conf

Email/SMS mirror · started Today 09:02 · 18 actions · 4 questions

1907 Belden Ave · TX-2033

OBS-1043 · R. Chen · Closing Scheduling

Live91% conf

Document event stream · started Today 10:31 · 12 actions · 0 questions

6 Quarry Rd · TX-2028

OBS-1044 · S. Patel · Earnest Money Verification

Shadow69% conf

Screen-share session · started Yesterday 15:20 · 41 actions · 6 questions

330 Sable Ct · TX-2012

OBS-1045 · M. Alvarez · Funded → Recorded

Replay93% conf

Read-only CRM connector · started Closed 12 Jul · 96 actions · 3 questions

Questions raised

Why was the servicer contacted by fax rather than portal?

Q-71 · OBS-1042 · TX-2039

Awaiting coordinator

Was skipping the buyer-agent confirmation intentional policy or file-specific?

Q-72 · OBS-1044 · TX-2028

Awaiting coordinator

Is the curative-first rule state-specific?

Q-73 · OBS-1041 · TX-2041

Answered — file-specific, easement only

Was the 5-day insurance binder gap avoidable?

Q-74 · OBS-1045 · TX-2012

Awaiting coordinator

Observed transactions

transaction	type	coordinator	stages observed	actions	exceptions	outcome
TX-2041	Traditional purchase	M. Alvarez	6/11	27	1	In flight
TX-2039	Subject-to	D. Okafor	8/11	63	3	In flight
TX-2033	Wholesale assignment	R. Chen	9/11	48	0	In flight
TX-2028	Double close	S. Patel	7/11	71	4	In flight
TX-2012	Seller financing	M. Alvarez	11/11	96	2	Closed
TX-1994	Novation	D. Okafor	11/11	88	5	Closed

Action timeline — append-only

source systems never modified

documents

Title commitment v2

communications

—

dependencies

Commitment received

decision

Escalate to curative before ordering survey

reason

Stated: survey cost is wasted if easement kills the lot line

result

Curative task opened

evidence

Commitment PDF page 4 · CRM task 90124

source

Read-only CRM connector

Expert patterns

≥5 obs · ≥2 coordinators

Resolve Schedule B-II exceptions before ordering survey.

Document ordering

recommendable

17 observations · 3 coordinators · 94% success · 2.4d median saving · 89% conf

Verify EM receipt at escrow source, never via forwarded copy.

Earnest money

recommendable

22 observations · 4 coordinators · 100% success · saving unknown · 93% conf

Send lender status request Tuesday morning, not Friday afternoon.

Communication timing

recommendable

31 observations · 4 coordinators · 81% success · 1.1d median saving · 84% conf

Reserve notary before clear-to-close on out-of-area signings.

Closing preparation

recommendable

9 observations · 2 coordinators · 89% success · 0.8d median saving · 72% conf

Escalate to loan-officer manager after 2 unanswered condition requests.

Lender follow-up

recommendable

12 observations · 3 coordinators · 75% success · 3.2d median saving · 70% conf

Schedule re-inspection while repair invoice is still pending.

Inspection coordination

sample too small

3 observations · 1 coordinators · 100% success · saving unknown · 41% conf

Open curative task at commitment review rather than at cure deadline.

Exception handling

recommendable

14 observations · 3 coordinators · 86% success · 4d median saving · 87% conf

Decision intelligence

Escalate curative before survey order

Commitment exception on lot-line easement

inputsCommitment v2 · Prior survey absent

reasonSurvey cost wasted if easement moves the lot line

outcomeSurvey deferred, $625 not spent

alternatives observedOrder survey immediately (observed 4×)

evidence ACT-88121 · 90% conf

Pad payoff good-through +4 days

Subject-to with slow servicer

inputsServicer SLA history · Scheduled close date

reasonServicer reissues take 3 business days

outcomeNo reissue needed

alternatives observedRequest exact-date payoff (observed 11×)

evidence ACT-88130 · 89% conf

Skip buyer-agent EM confirmation

Double close, escrow already responsive

inputsEscrow direct receipt

reasonnot stated — not inferred

outcomeNo downstream issue

alternatives observedRun playbook step 4 (observed 19×)

evidence ACT-88155 · 71% conf

Exception intelligence

trigger	reason	action taken	outcome	freq	playbook?
Payoff expires before scheduled closing	Servicer reissue latency	Request +4 day good-through window	88% of expirations eliminated	29	Pending approval
Seller is a trust with successor trustee	Trust certification not in file	Request certification at intake, not at signing	Avg 3.1 days saved	11	Approved as module
Buyer-agent forwarded EM receipt	Chain-of-custody concern	Verify direct with escrow	Zero fraud incidents in observed window	22	Pending approval
HOA estoppel delayed past 10 days	not stated	Escalate to HOA management company principal	Mixed — 6 of 9 resolved within 48h	9	Pending approval
Municipal lien search returns open permit	not stated	Coordinate seller-side permit closeout	Observed twice, outcome varies	2	Insufficient sample

Knowledge candidates — approval queue

4 awaiting human approval

Insert 'curative before survey' gate at commitment review.

KC-601 · Traditional purchase

Awaiting human approval

17 observations · 94% success · 1 exceptions · 89% conf

coverage 17/21 · routed to Assistant only

EM verification must originate from escrow of record.

KC-602 · All deal types

Awaiting human approval

22 observations · 100% success · 0 exceptions · 93% conf

coverage 22/24 · routed to Assistant only

Payoff requests default to good-through +4 business days.

KC-603 · Subject-to

Approved

29 observations · 88% success · 3 exceptions · 90% conf

coverage 29/31 · routed to Assistant only

Lender condition requests batched Tuesday 09:00.

KC-604 · Traditional purchase

Awaiting human approval

31 observations · 81% success · 6 exceptions · 84% conf

coverage 31/44 · routed to Assistant only

Reserve notary before clear-to-close on out-of-area signings.

KC-605 · Double close

Awaiting human approval

9 observations · 89% success · 1 exceptions · 72% conf

coverage 9/15 · routed to Assistant only

Insufficient sample — no candidate generated for re-inspection timing.

KC-606 · Novation

Rejected

3 observations · 100% success · 0 exceptions · 41% conf

coverage 3/12 · routed to Assistant only

Shadow learning

observation only

Observation only — AI predicts the next action and never performs it.

412 predictions logged

Next-action prediction accuracy79%

412 predictions across 6 transactions · 86% conf

Timing accuracy (±2h)64%

318 timed predictions · 77% conf

Human corrections applied21%

87 corrections logged · 90% conf

Missing-context rate14%

58 predictions lacked an observable input · 81% conf

Exception-path accuracyunknown

Only 9 exception predictions — below threshold · 30% conf

human corrections

predicted Order survey → actual Open curative task

M. Alvarez · Model missed easement dependency.

predicted Email buyer agent for EM → actual Email escrow directly

S. Patel · Source-of-truth rule not yet learned.

predicted Schedule signing → actual Reserve notary first

R. Chen · Sequencing preference, out-of-area file.

CRM connector framework

read-only

Follow Up Boss

Connected

observes: Transaction changes · Task completion · Status changes

access Read-only · writes None · last event Today 10:44

Dotloop

Connected

observes: Document events · Signature events

access Read-only · writes None · last event Today 10:31

Google Workspace mirror

Connected

observes: Communications · Calendar deadlines

access Read-only · writes None · last event Today 10:52

Qualia

Pending approval

observes: Title milestones · Closing documents

access Read-only · writes None · last event —

SkySlope

Not configured

observes: —

access Read-only · writes None · last event —

No writes. No automation. No mutation of external systems.

Expert comparison

anonymized · advisory

coordinator	workflow difference	median stage hrs	exception handling	communication	playbook usage	observed
Coordinator A	Curative-first on title exceptions	26	Escalates early, documents reason	Written, batched	91%	21
Coordinator B	Parallel lender + title tracks	31	Waits for servicer SLA, then escalates	Phone-first, logged after	78%	18
Coordinator C	Signing logistics front-loaded	24	Rare deviations	SMS for scheduling, email for record	95%	16
Coordinator D	Source-verification heavy	29	Deviates when source of truth is available	Written, immediate	72%	14

Comparisons are organizational-playbook inputs only. Individuals are anonymized and never publicly ranked.

Knowledge coverage

Transaction types6/10 · 60%

Lease option, land, commercial, hybrid unobserved.

Pipeline stages9/11 · 82%

Recorded and post-close stages thinly observed.

Exception classes12/21 · 57%

Bankruptcy and probate exceptions unobserved.

Document types34/51 · 67%

Commercial and HOA packages missing.

Communication channels4/5 · 80%

Fax observed indirectly only.

Playbooks7/10 · 70%

Three playbooks have no observed execution.

knowledge gaps — unknown remains unknown

No observed commercial purchase execution — playbook remains unvalidated.
Probate and bankruptcy exception handling has zero observations.
Post-recording follow-through observed on 2 files only.
Timing data for fax-based servicers is inferred from document events, not observed directly.

Replay learning

candidates only

Seller-financing note reviewed before escrow opening.

Critical decision

TX-2012 · evidence ACT-77120 · note draft v1 · learning candidate generated

Insurance binder waited 5 days on a single unanswered email.

Delay

TX-2012 · evidence Email thread 3102 · learning candidate generated

Novation consent obtained in writing before marketing.

Best practice

TX-1994 · evidence Consent PDF · doc event 71002 · learning candidate generated

Payoff could have been requested 6 days earlier without cost.

Missed opportunity

TX-1994 · evidence Payoff timestamps vs. escrow open · learning candidate generated

Trust certification obtained at signing; intake path was available.

Exception path

TX-2012 · evidence EXC-52 precedent · no candidate — insufficient basis

TC Pros owns

Workflow observationExpert knowledge captureTransaction pattern recognitionReplay analysisKnowledge candidate generation

TC Pros does not own

CRM controlExternal system automationCross-app orchestrationLearning publication

All connectors are read-only and append-only. No automation is created directly from observation; Assistant remains the sole orchestrator.