Cross-layer

Process Control System

Cross-layer statistical process control for governed inference quality.

Process Control System monitors governance quality over time, separates common-cause variation from special-cause degradation, and triggers remediation before failures become systemic.

Without this

Without statistical process control, governance quality is invisible over time. You can’t distinguish a bad day from a degrading system. By the time failures become obvious, they’re systemic.

Position in the platform

System	Layer	Function
Oracle Foundry	Foundation	Supplies retrieval and corpus quality signals
SIRE Crosswalk	Post-Foundry	Supplies gap/conflict counts and relationship-drift metrics for library SPC
Prompt Compiler	L0	Produces structural risk and rewrite-pass telemetry
Claim Ledger	L1-L4	Produces per-claim scores, gate outcomes, and loss metrics
Process Control System	Cross-layer	Computes SPC signals, capability indices, and operational state
Forensics Lab	L5	Consumes SPC special-cause signals for root-cause investigation

Statistical methods

Shewhart control charts

I-MR, p-chart, and c-chart monitoring for request-level governance characteristics.

Western Electric rules

Pattern-based special-cause detection beyond single-point control-limit breaches.

EWMA and CUSUM

High-sensitivity drift and shift detection for gradual degradation.

Process capability (Cpk)

Capability scoring against risk-tier specification limits to prove margin, not just pass/fail.

Monitored quality characteristics

Metric	Source	Chart
Calibrated confidence	Claim Ledger L1	I-MR
Worst-claim entailment	Claim Ledger L2	I-MR
Stability score	Claim Ledger L3	I-MR
Representation uncertainty	Claim Ledger L4	I-MR
Retrieval recall	Oracle Foundry	I-MR
Gate pass rate	Gate outcomes	p-chart
Uncited claim count	Claim Ledger L2	c-chart

Operational posture states

in_control, Cpk >= 1.33

Stable and capable process with operating margin

Action: Normal operations and continuous improvement

in_control, Cpk < 1.33

Stable but marginally capable process

Action: Targeted system improvement to increase margin

out_of_control

Special-cause variation detected

Action: Trigger investigation and Forensics escalation

Deployment profile

• Runs across Studio, Refinery, and Clean Room environments (runtime profiles: Lite, Standard, Full)
• Requires governance-envelope telemetry, not hidden-state access
• Feeds asynchronous triggers to Forensics Lab through canonical trigger keys

What operators and auditors see

• Control charts with UCL/LCL, rule annotations, and drift overlays
• Weekly capability indices by characteristic and risk tier
• Special-cause alerts with probable-cause categories and investigation status
• Before/after baseline comparisons following remediation

Live SPC snapshot

Weekly SpcSnapshot showing two out-of-control charts with special-cause signals triggering Forensics Lab.

{  "snapshot_id": "spc_20250228_0800",  "window": "2025-02-21T00:00:00Z / 2025-02-28T00:00:00Z",  "sample_size": 312,  "posture": "out_of_control",  "charts": [    {      "chart_id": "uq.entailment_pass_rate",      "type": "p-chart",      "characteristic": "Gate pass rate",      "source": "Gate outcomes",      "centerline": 0.936,      "ucl": 0.978,      "lcl": 0.894,      "last_value": 0.871,      "in_control": false,      "violated_rules": ["below_lcl", "we_rule_2"],      "cpk": 0.91    },    {      "chart_id": "uq.worst_entailment",      "type": "I-MR",      "characteristic": "Worst-claim entailment",      "source": "Claim Ledger L2",      "centerline": 0.812,      "ucl": 0.943,      "lcl": 0.681,      "last_value": 0.754,      "in_control": true,      "violated_rules": [],      "cpk": 1.38    },    {      "chart_id": "uq.uncited_claims",      "type": "c-chart",      "characteristic": "Uncited claim count",      "source": "Claim Ledger L2",      "centerline": 1.4,      "ucl": 4.9,      "lcl": 0,      "last_value": 5,      "in_control": false,      "violated_rules": ["above_ucl"],      "cpk": null    }  ],  "special_cause_signals": [    {      "chart_id": "uq.entailment_pass_rate",      "rule_id": "we_rule_2",      "signal_value": 0.871,      "consecutive_count": 3    },    {      "chart_id": "uq.uncited_claims",      "rule_id": "above_ucl",      "signal_value": 5,      "consecutive_count": 1    }  ],  "forensics_triggered": true,  "trigger_ids": ["spc_special_cause"]}

Run AI quality like a controlled process

Process Control System turns governance telemetry into statistical evidence of stability, capability, and degradation response.

Architecture Claim Ledger SIRE Crosswalk Forensics Lab The Clean Room Check Your Risk

Who uses this

Operator

Quality engineers

Operations teams who monitor control charts, respond to special-cause alerts, and track capability indices.

Consumer

Governance operators and Forensics Lab

Operators use dashboards for real-time quality visibility. Forensics Lab receives special-cause triggers for root-cause investigation.