AML SAM QC — Auditor Diagnostic Cockpit

Model - Thresholds - Data - Built - Reviewed this session 0

📊 Component 1: Structured Risk & TreeSHAP

📈 Transaction Forensic Story

Customer Type

KYC Risk Tier

Account Tenure

Peer Cash 30D Ratio

Escalation Risk (P_struct)

0.00

NORMAL

TreeSHAP Typology Attribution Threshold ≥ 0.50

📝 Component 2: Narrative Intelligence & Documentation

📋 Documentation Audit & Note Critique

💡

Key Takeaway

D_proc (Narrative/Proc): 0.00 D_trans (Data/Narr): 0.00

ALIGNED

Conclusion Intent: -

Narrative Typology Attribution BACKEND AUTHORITATIVE

Evidence & Token Attribution (Φ)

Investigator Narrative (Masked)

Strategy:

Cutoff Floor: |φ| ≥ 0.010 (Fixed) P_narr Threshold: 0.500 (Baseline) D_sem Band: Tension ≥ 0.70

Top Risk Tokens (+φ)

Top Mitigating Tokens (-φ)

Typology & Rationalization Tags

Narrative Conclusion & Intent Audit CONCLUSION

⚖️ Component 3: QC Verdict & Plain-English Story

QC ACTION

DEFECT CODE

🔍 Executive Supervisory Finding & QA Guide

📐 3-Pole Case Triangulation & Synthesis

ALIGNED

Anchor Poles (Calibrated Baselines) • Click to explain

Data (P_struct)

0.00

Notes (P_narr)

0.00

Decision (Y)

D_global: 0.00 • D_trans: 0.00 • D_proc: 0.00

Intent: -

✅

Triangulated Harmony

Structured data, narrative evidence, and operational outcome are in complete accord.

📊 Level 2: Detailed Layman Breakdown

▼ Click to expand

📋 Deterministic Rule Path Trace

▼ expand

Status:

Select action below

OPTION 2 ACTIVE Engine: LightGBM-SAM-v2.4.1 • Threshold: 0.650 • Source: SAM_TX_STREAM_DAILY • Mode: Component 1 Expandable Split Drawer (Zero-Void Fluid Fr Grid)

Queue: 14 demo alerts

Hotkeys: 1 2 3 • Layout: M

🛡️ Triangulation HUD:

P_struct: --

D_sem: --

QC Verdict: --

Customer: -- • Alert: -- •

📊 Component 1: Financial Risk, Explainability & Identity Triangulation

ALERT: D-ALT-02 • HIGH RISK (97.9%)

ESCALATION RISK PROBABILITY ✓ RECONCILED

97.9%

HIGH RISK

Base: -1.1450 • Net Δ: +5.0015

🌲 TreeSHAP Typology Attribution LUNDBERG SHAP

Threshold ≥ 0.50

AI FORENSIC STORYLINE LIGHTGBM-SAM

👤 CUSTOMER TRIAGE SNAPSHOT

✓ STABLE (90D)

Customer ID:-

Customer Name:-

Entity / Industry:-

KYC Risk Tier: -

Peer Velocity Divergence: -

Detailed Limits & PII:

🔍 FORENSIC DEEP-DIVE DRAWERS (SEQUENCED BY INVESTIGATIVE LOGIC) Click to toggle / unhide

📝 Component 2: Narrative Intelligence & Documentation

Structured vs Narrative Dissonance

ALIGNED

D_proc (Narr/Proc): 0.00 D_trans (Data/Narr): 0.00

Intent: -

Narrative Typology Coverage BACKEND AUTHORITATIVE

📝 Documentation Audit & Note Critique

Linguistic Logic vs. Reality

💡

Key Takeaway

🔍 Narrative Details & Token Audit (Hidden by Default) Click to toggle / unhide

⚖️ Component 3: Supervisory Synthesis & QC Verdict

STANDARDIZED ALIGNMENT

Three-Way Diagnostic Triangulation TRIANGULATED

Data (P_struct)

0.000

Notes (P_narr)

0.000

Decision (Y)

CLOSE (0)

D_global: 0.00 • D_trans: 0.00 • D_proc: 0.00

Intent: -

✅

Triangulated Harmony

Structured data, narrative evidence, and operational outcome are in complete accord.

⚖️ Executive Supervisory Finding & QA Guide

Institutional Audit Standard

🚨 What Happened: High statistical risk detected.

💡 Why it's a Defect: Investigator disregarded material indicators.

📋 QA Action: Return for rework and full CTR reconciliation.

DEFECT FLAGGED BY SECOND-LINE QC

DEF_C1_DISCORDANCE

📊 Level 2: Detailed Layman Breakdown

▼ Click to expand

⚙️ Deterministic Rule Path Trace

▼ expand

Status:

Select action below

👤

Authoritative Customer Profile Audit Workspace

ADR-CP-0002 Compliance

Subject Entity: Arthur Pendelton • Cust ID: CUST-AML-07 • Active Alert: ALT-0007

Review Stage:

🔒

Zero Hindsight Audit Baseline • Frozen at T_decision

Evaluation Cutoff: 2024-04-20 16:45:00 UTC • Sealed Manifest: RFC 8785 SHA-256: 8a7f92e4...b41d

✓ Temporal Isolation Enforced

⚡ 1) Profile Mutation Audit (T_decision) LOCK AT T_DECISION

Decision-time facts and subsequent profile changes are separated below; later events are non-penalizing.

📈

2) Expected vs. Observed Engine ✓ Within Normal Tolerances

Velocity < 90%, Spike < 3.0x, KYC Limit < 1.5x

⏱️ Authoritative Bitemporal Profile Mutation Timeline Bitemporal Fact Log

Valid-Time × Assertion-Time Replay

🛡️

Bitemporal Audit Standard: Reconstructs valid-time validity intervals against system assertion timestamps. Validates whether the investigator had access to updated profile facts at decision time (T_decision).

Reconstruct As-Of:

Attribute Fact	Effective Value (T_decision)	Valid-Time Interval	System Record Time (T_assert)	Source System	Knowable at T_decision	Defect Attribution

🆔 1. Entity Identity & Demographics

Legal Name: -

Customer ID: -

Entity Type: -

Formation / DOB: 1948-02-10

Tax ID / SSN: ***-**-6789

Jurisdiction: United States (USA)

Occupation/Industry: Real Estate Consulting

Customer Since: 2019-05-14 (5.1 yrs)

Status: ACTIVE

📍 2. Authoritative Contact Facts Bitemporal

PRIMARY PHYSICAL ADDRESS:

1400 Ocean Drive, Apt 4B, Miami, FL 33139

Source: BRANCH_PORTAL ⚠️ OMITTED IN NARRATIVE (D2)

PRIMARY PHONE NUMBER:

+1-305-555-0199 (Miami VoIP)

Source: DIGITAL_PORTAL ⚠️ OMITTED IN NARRATIVE (D2)

EMAIL ADDRESS:

arthur.pendelton@oceanproperties.com

📋 3. KYC / CDD Baseline Version 2 Active

Expected Wire Limit: $150,000 / mo

Expected Cash Limit: $10,000 / mo

Stated Annual Revenue: $1,200,000

Source of Funds: Real Estate Consulting Fees

Source of Wealth: Commercial Property Holdings

PEP Status: NO (CLEAR)

Adverse Media: NO ADVERSE MEDIA

Authorized Corridors: US, GB, CH, KY

🏦 4. Accounts & Banking Footprint

Operating Checking (ACC-789012)

Opened 2019-05-14 • Status: ACTIVE

$142,500.00

USD

Wire Settlement (ACC-456789)

Opened 2021-08-20 • Status: ACTIVE

$230,000.00

USD

🛡️ 5. Sanctions & Watchlist Screening Cleared

Candidate: Arthur Pendelton Score: 18/100

Watchlist: OFAC_SDN_AND_NON_SDN_V2 • Match Type: FUZZY

✓ Cleared: False match due to distinct DOB and nationality

🌐 6. Beneficial Ownership & Network 2 Links

Ocean Drive Realty LLC

Role: 100% Beneficial Owner • Formed 2022

UBO

Sarah Pendelton

Role: Authorized Signer (Spouse)

Signer

⚖️ Stage 1: Decision-Time Profile Consideration Audit (T_decision) DEFECT D2: EVIDENCE OMISSION

What Investigator Stated in Narrative at T_decision:

"Customer transactions reviewed. Activity aligns with seasonal operational expenditure. No suspicious adverse media or typologies identified."

✕ Failed to mention the customer's out-of-state relocation from Phoenix to Miami occurring 4 days before the $188,100 wire surge.

QC Auditor Finding & Root Cause:

Authoritative profile mutation fact was active and knowable at T_decision. Failure to document or reconcile the address change prior to high-velocity wire spike constitutes Defect D2 (Evidence Omission).

💬 Grounded Audit Copilot

🛡️ Strict Grounding Active

📖 Copilot Capability Guide & Prompt Playbook (Click any blue prompt chip to run it directly)

✅ What Works (Supported Capabilities) GROUNDED

1. Defect Summaries & Real Data Examples:
⚡ summarize different types of defects in a table with examples from data

2. Defect Comparisons & Tables:
⚡ difference between D1 and D3 in a table ⚡ compare D1 vs D6

3. Multi-Turn Follow-Up Formatting:
Ask any question first, then follow up with:
⚡ Can you put the response into a table?

4. Regulatory Standards & Model Governance:
⚡ 31 CFR SAR Rules & Deadlines ⚡ SR 11-7 Model Governance

5. Active Alert Forensics (Check [x] Include Alert Data):
⚡ Analyze selected alert

🚫 What Won't Work (System Boundaries) GUARDRAILS

1. Non-AML / Out-of-Domain General Queries:
❌ "What is the weather in London?" / "Write Python code"
Strictly grounded to uploaded AML compliance manuals to prevent corporate hallucination.

2. Speculation without Grounded Evidence:
❌ "Guess what the customer will do next month"
Every statement requires cryptographic citation attribution to indexed documents.

3. Mutating Audit Decisions via Chat:
❌ "Overturn ALT_0008088 to agree and close"
Decisions must be logged using the action buttons in the Processing tab with SHA-256 ledgers.

4. Unchecked Alert Context:
If [ ] Include Alert Data is unchecked and prompt does not mention the alert, responses remain strictly policy-level.

👋 Welcome to the AML SAM QC Institutional Copilot.

I provide precise, audit-grounded answers derived strictly from your uploaded compliance records, SAR filing manuals, model governance policies (SR 11-7), and alert dossiers.

💡 Need Help? Click [📖 Copilot User Guide & Playbook] in the header to see what works, what won't work, and one-click prompt templates.

System Copilot · Zero-Hallucination Guardrail

⚡ Explain D1 vs D6

📋 31 CFR SAR Rules

⚖️ Dissonance Thresholds

🔍 Analyze Selected Alert

📊 Include Alert Data

🔍 Citation Inspector Click citation badge to inspect

Click on any citation chip [Doc §Section] in the conversation to view the exact verified text passage and chunk hash.

⚡ FAQ Synthesis Studio 0 FAQs

📡 DAEMON WATCHER ACTIVE Latest: Candidate v1.1.0 evaluated in autonomous shadow benchmark • 36 CMS alerts indexed.

                    Continuous Shadow Monitoring
                

Champion (In-Production) ACTIVE

100.0%

Defect Recall @ 11.3% Review Workload (5.3 FTEs)

Latest Challenger VALIDATED

93.4%

CFN 50x @ 9.8% Review Workload (4.6 FTEs)

SR 11-7 Gate Verdict PASSED (4/4)

TIER-1 PASS

Wilson 95% CI: [88.2%, 96.5%] • Zero critical leakage

Pareto Efficiency OPTIMAL

9.53x

Defect capture ratio per 1% audit queue

🎛️ Interactive Capacity Sizer & Threshold Calibrator

Move slider to model operational workload vs. defect recall trade-offs

Target Review Workload (% of alerts routed for audit): 10.0%

5.0% (Aggressive Filtration) 10.0% (Balanced Baseline) 15.0% (Conservative) 25.0% (Exhaustive Regulatory Audit)

Monthly Audit Queue: 1,000 Alerts Out of 10,000 enterprise alerts/mo

Required Auditor Capacity: 4.7 FTEs @ 45 min/review, 160 hrs/mo

Projected Defect Recall: 91.2% Wilson 95% CI: [86.4%, 94.8%]

Calibrated Closed Risk Floor: 0.470 closed_alert_risk_min threshold

📐 Threshold Reconciliation (Champion vs. Calibrated)

Parameter	Champion	Challenger (50x)	Simulated	Delta
closed_alert_risk_min	0.450	0.485	0.470	+0.020
acute_shap	0.550	0.540	0.540	-0.010
over_escalation_struct_max	0.450	0.420	0.420	-0.030
narrative_defect_word_count	15	14	14	-1
narrative_defect_probability_max	0.250	0.265	0.265	+0.015
evidence_omission_word_count	24	22	22	-2
unsupported_conclusion_narrative_max	0.350	0.340	0.340	-0.010

📈 Evaluated Cost-Ratio Pareto Frontier

Scenario	Workload	Recall	Precision	FTE	Pareto
CFN 10x	7.2%	85.4%	74.1%	3.4	OPTIMAL
CFN 25x	8.6%	89.8%	71.5%	4.0	OPTIMAL
CFN 50x (Staged)	9.8%	93.4%	68.2%	4.6	OPTIMAL
CFN 100x	13.5%	97.8%	59.4%	6.3	OPTIMAL
Champion Baseline	11.3%	100.0%	70.6%	5.3	BENCHMARK

🛡️ SR 11-7 / OCC 2011-12 Model Risk Validation Scorecard

Mandatory Decision Gate	Regulatory Target	Observed Value	Wilson 95% Confidence Interval	Status
GATE 1: Minimum Defect Recall Floor	≥ 85.0%	93.4%	[88.2%, 96.5%]	PASSED
GATE 2: Operational Workload Ceiling	≤ 25.0%	9.8%	—	PASSED
GATE 3: Precision Floor (Alert Fatigue Protection)	≥ 30.0%	68.2%	[61.4%, 74.3%]	PASSED
GATE 4: Chronological OOT Stability	Δ Recall ≤ 8.0%	Δ = 2.1%	In-Sample: 93.4% • OOT: 91.3%	PASSED

💬 NLPLab Narrative Token Budget & Truncation Studio BENCHMARK ACTIVE

Model LLM token cutoffs (64 - 1024 tokens) and truncation strategies (Head, Tail, Extractive) vs. defect recall retention and API cost economics

★ RECOMMENDED: 64 TOKENS (HEAD)

Optimal Budget

64 Tokens

Head Truncation Policy

Defect Recall Retention

100.0%

Zero Recall Degradation

Est. Latency / Alert

~127 ms

-45% vs. Full Context

SR 11-7 Recall Floor

PASSED

Threshold ≥ 85.0% Floor

Truncation Strategy	Token Cutoff	Avg Tokens / Alert	% Token Reduction	Monthly API Cost	Defect Recall	Pareto Status
head (Recommended)	`64 tokens`	17.1	0.0%	$0.26	100.0%	★ PARETO
tail	`64 tokens`	17.1	0.0%	$0.26	100.0%	OPTIMAL
extractive_typology	`64 tokens`	17.1	0.0%	$0.26	100.0%	OPTIMAL
head	`128 tokens`	17.1	0.0%	$0.26	100.0%	OPTIMAL
extractive_typology	`256 tokens`	17.1	0.0%	$0.26	100.0%	OPTIMAL
full_unconstrained	`unconstrained`	17.1	0.0%	$0.26	100.0%	BASELINE

Authoritative Customer Profile Audit Workspace

🧪 QMS Challenger Lab & Pareto Studio

🎛️ Interactive Capacity Sizer & Threshold Calibrator

📐 Threshold Reconciliation (Champion vs. Calibrated)

📈 Evaluated Cost-Ratio Pareto Frontier

🛡️ SR 11-7 / OCC 2011-12 Model Risk Validation Scorecard

💬 NLPLab Narrative Token Budget & Truncation Studio BENCHMARK ACTIVE