AI Quality, Drift & Fairness: continuous evaluation

A provider quietly updated the base model. The API is formally the same, the version string is unchanged — but behavior shifted: Kovcheg started inventing loan terms not present in the documents slightly more often, and treating applications from one region slightly more strictly. No one changed anything in the code. A month later it surfaces as a spike in complaints and a question from compliance: "Why have approvals in region N dropped?" This is what silent degradation looks like — the most common way an AI system's quality erodes without a single deployment.

This chapter covers the quality plane: the guarantee that the model doesn't degrade over time, answers stay accurate through base-LLM updates, and decisions don't discriminate. For a credit-scoring system like Kovcheg, bias in scoring isn't a dashboard metric. It's a direct violation of the EU AI Act (Art. 10 — data quality, non-discrimination) and a legal claim.

The quality-plane engineer builds the evaluation infrastructure that makes these guarantees measurable and enforces them as a release gate.

The business goal

Kovcheg's decision quality and fairness must be measurable, continuously monitored, and wired into the release pipeline. Three promises to the business:

A model update never ships to production without passing regression against benchmarks.
Degradation and drift are caught by monitoring, not by complaints.
Absence of demographic bias in scoring is provable.

What drives this

Silent degradation: an under-the-hood model update tanks quality invisibly.
EU AI Act (high-risk): Art. 10 (data quality, non-discrimination), Art. 15 (accuracy, robustness), Art. 14 (human oversight).
Bias: systematically worse outcomes for a demographic group means a lawsuit, a fine, and reputational damage.
OWASP LLM: LLM09 (Misinformation) — hallucinations as a class of failure.

The architectural pattern

Continuous Evaluation & LLM-as-a-Judge Pipeline. Evaluation is a continuous process: regression against golden datasets in CI/CD, online RAG metrics, drift monitoring, and fairness tests — all backed by an honestly calibrated judge.

Engineering stack

Eval frameworks: Ragas (v0.2+ — no longer just RAG, agentic pipelines too), DeepEval, TruLens, MLflow (experiments and registry).
Drift and data quality: Evidently AI, Great Expectations.
Judge: LLM-as-a-judge, calibrated against human labeling.

Engineering implementation

Step 1. Golden datasets

Reference sets for regression-testing prompts and models in CI/CD. A model or prompt update can't ship without passing the run. The dataset includes "hard" cases and cases pulled from real incidents.

Step 2. Online RAG metrics

On a sample of production traffic: Faithfulness (grounding to context — a direct defense against the hallucinations from the opening scenario), Answer Relevance, Context Recall/Precision. This ties to the output guardrail from Chapter 3 — that's the real-time block; this is the trend.

Step 3. Drift monitoring

Anomalies in the distribution of incoming query embeddings (data drift) and in answer quality (concept drift). A shift triggers an alert. One subtlety: semantic caching (Chapter 5) masks drift — measure on the uncached sample.

Step 4. Fairness and bias

A regular run of synthetic tests: swap protected attributes (gender, age, region) while holding everything else equal, then compare decisions. A systematic difference is a bias signal. For Kovcheg, this is a mandatory gate for the credit module.

Step 5. Release eval gate

Metrics are part of the pipeline (policy-as-code, Chapter 6). Falling below a threshold blocks the deploy:

eval_gate:
  faithfulness:      {min: 0.92}
  answer_relevance:  {min: 0.85}
  fairness_delta:    {max: 0.03}   # max. difference in decisions on attribute swap
  block_release_on_fail: true

Where it breaks

LLM-as-judge is itself biased and non-deterministic. Documented distortions: verbosity bias (longer = "better"), self-preference (favors its own style), position bias (answer order). Mitigation: randomize order, use clear rubrics, ensemble judges, human golden labeling. Without calibration, you're "evaluating the evaluator."
Calibration drifts. Judge-human agreement that was solid six months ago goes stale — prompts, data, and model versions change. Recalibration is a recurring process, not a one-off.
The golden dataset goes stale. It doesn't cover new cases, so metrics stay green during real degradation. Refresh the dataset from incidents and production traffic.
Fairness is hard to measure. "Fairness" is context-dependent; several formal fairness metrics are mathematically incompatible at the same time — choosing a metric itself requires justification.
Cost. Continuous judge evaluation is expensive; sampling versus completeness is a real tradeoff.

Standards and mapping

EU AI Act: Art. 10 (data governance/quality, non-discrimination), Art. 15 (accuracy/robustness), Art. 14 (human oversight).
ISO/IEC 42001: performance monitoring, continual improvement.
NIST AI RMF: Measure (valid, reliable, fair, safe).
OWASP LLM: LLM09 (Misinformation).

Lab and artifact

Build a golden dataset for Kovcheg (including incident-derived cases), configure Ragas metrics online, set up Evidently for embedding drift on the uncached sample, implement an attribute-swap fairness test for the credit module, wire an eval gate into CI, calibrate the judge against human labeling, and measure agreement.

Artifact: an eval suite + a quality/drift dashboard + a fairness report + a judge-calibration report (evidence for the RMS, Chapter 6).

Maturity checklist

L1: manual quality checks on update.
L2: a golden dataset + online RAG metrics, drift alerts.
L3: an eval gate in CI, a calibrated judge with recalibration, regular fairness runs, drift measured on the uncached sample, tied into the RMS (Chapter 6).

Sources

The provider will update the base model again next quarter — the version string won't change, and the behavior will shift. The question is whether your eval gate catches it before compliance does.

AI Quality, Drift & Fairness: continuous evaluation and bias mitigation