SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsai-evals

ai-evals

Testing & quality

View repository

Updated 10-16 05:21

About this skill

AI Evaluation and Fine-Tuning Methodology Skill

Use this skill to calibrate graders, derive release gates, and compare evaluation or adaptation methods. Domain skills own what to measure; this skill owns the measurement controls. Training loss is telemetry; held-out behavior determines whether an adaptation improved the system.

Quick Reference

Task Read or Run Outcome
Build a (question, ideal-answer) set and tune it references/dataset-construction.md Sourcing, ideal-answer authoring, run→compare→tune loop, robustness slice by perturbing memorized cases
Compose an eval dataset references/eval-dataset-design.md Task distribution, difficulty, sampling, annotation, versioning, contamination, bias audits, golden vs dynamic sets
Stop a judge from rating its own output high references/llm-judge-bias.md Self-preference, position, length, verbosity controls
Pick / wire an eval framework references/framework-integration.md inspect-ai, lighteval, Ragas, DeepEval, promptfoo, Braintrust integration snippets + when to use each
Choose metrics and a pass threshold defensibly references/threshold-derivation.md Metric set by use case with direction and action-if-fails; derive thresholds from a labeled set; inter-rater agreement; gate design
Stop flaky runs reading as regressions references/flake-and-reproducibility.md pass@k vs pass^k, seeds, temperature, quarantine, contamination/leakage, system-benchmark hazards (hardware lottery, thermal/background load, unreported CIs)
Decide if "A beats B" is real, size the set references/eval-statistics.md Bootstrap CIs, McNemar, power/MDE sizing, FDR, variance reduction
Get maximum from an LLM references/llm-optimization-technique-map.md Technique ladder across prompts, data, RAG/tools, test-time compute, SFT, preference/RFT, PEFT, distillation; measured case where a confidence scorer scored below no scorer
Decide whether and how to fine-tune references/fine-tuning-eval-loop.md Prompt/RAG/tool baseline, SFT vs preference/RFT vs PEFT, split hygiene, promotion gates
Evaluate on live/production traffic references/online-production-eval.md Offline-online correlation, A/B+guardrails, shadow/canary, drift, regression replay, HITL
Turn user behavior into eval and preference data references/conversational-feedback-signals.md Implicit NL signals, action signals, edit→preference pairs, collection timing, feedback biases
Evaluate refusals, jailbreaks, harm references/safety-redteam-eval.md Over/under-refusal, ASR per attack family, injection, harm rubrics, robustness
Evaluate responsible or multimodal AI references/responsible-multimodal-evaluation.md Intersectional fairness, privacy/memorization/poisoning, oversight/appeals, provenance, grounding, diffusion, multimodal attacks, latency/cost
Go beyond one judge references/advanced-judging.md Juries, fine-tuned judges, CoT/probability scoring, calibration (kappa/ECE), agentic reward, human-vs-judge routing
Evaluate a retrieval, KB, or memory system references/retrieval-and-memory-eval.md Independent held-out set, slice floors, Wilson CIs, confident-wrong rate, per-stage tests, change-triggered regression gate, state-maintenance acceptance tests, cross-tenant canaries, learned-memory baselines, vendor benchmark claims
Measure whether memory skills improve an agent memory-skills-ablation Validate paired skills-on/off outcomes, complete safety slices, configuration parity and measured telemetry; export descriptive paired results
Measure hallucination references/hallucination-eval.md Taxonomy, claim-level scoring, abstention next to hallucination rate, reference-free triage

Scope Boundaries (Use These Skills for Depth)

Workflow

  1. Transform the vague ask into a verifiable goal. "Is it good?" is not gradeable. Ask: which case would fail first if the requirement reverted?
  2. Build the dataset before the grader. Source real questions, author ideal answers from the system's allowed context, and plan the run→compare→tune loop — see references/dataset-construction.md. No dataset, no eval.
  3. Pick the cheapest grader that works. Deterministic check > LLM judge > human. Reserve the LLM judge for what code cannot decide (Rule 5: use the model only for judgment calls).
  4. If using an LLM judge, control its bias before trusting any number — see references/llm-judge-bias.md. Judge bias can confound quality with surface features. When one judge isn't enough (high stakes, weak agreement, open-ended), escalate to juries / fine-tuned judges / calibrated scoring — see references/advanced-judging.md.
  5. Control flake and leakage with repeated trials (pass^k for reliability gates, pass@k only for capability), low judge temperature, seed pinning, and held-out testsets — see references/flake-and-reproducibility.md.
  6. Derive thresholds from a labeled calibration set, not intuition or copied targets — see references/threshold-derivation.md. Size the gating set and judge "A beats B" with statistics (bootstrap CIs, McNemar, power/MDE, FDR) — see references/eval-statistics.md. For frozen paired results, run python3 scripts/analyze_paired_results.py results.csv --estimand unit_mean; declare whether units or clusters define the target population first. A population-level superiority claim needs design-compatible uncertainty; exact fixed-suite differences remain descriptive evidence.
  7. Only fine-tune after the baseline has earned it. Compare prompt/RAG/tool fixes first, then choose SFT for imitation/style/format/tool-call behavior, preference/RFT for rubric-scored reasoning or tradeoffs, and PEFT/LoRA/QLoRA when adapting an open model under compute or deployment constraints — see references/fine-tuning-eval-loop.md. Training loss is telemetry; held-out behavior is the verdict.
  8. Apply the full optimization ladder, not one pet method. For maximum LLM performance, evaluate cheap prompt/context/tool fixes, then inference-time methods (self-consistency, best-of-N, rerank/verify/refine), then data/SFT, preference/RFT/RLVR, PEFT, and distillation as the evidence warrants — see references/llm-optimization-technique-map.md. Each technique gets its own failure mode and gate.
  9. Gate loudly. A gate that passes while silently skipping cases is a failure dressed as success (Rule 12: fail loud). Report skipped/quarantined cases in the gate output.
  10. Extend past the offline gate where the system warrants it. Add safety/red-team evaluation (refusal precision/recall, jailbreak ASR, injection, harm rubrics) — see references/safety-redteam-eval.md — and, once in production, online evaluation (offline-online correlation, A/B with guardrails, drift, regression replay) — see references/online-production-eval.md. The offline gate is a filter; production is the verdict.

Core Principles

  • The judge is a model with failure modes. Treat its scores as one calibrated input, never as ground truth.

  • Judge independence needs evidence. Prefer a different judge model and calibrate against human labels and planted failures. A different family reduces one self-preference risk; it does not establish an unbiased grader.

  • Behavior, not plausibility. Rubrics that reward "looks good" reward length and confidence. Pin rubrics to verifiable behavior.

  • No threshold without a labeled set. A copied target (">95%") is a guess until validated on your own distribution.

  • No fine-tune without a baseline and a holdout. A tuned model that beats no prompt/RAG/tool baseline, or only wins on the training/dev set, has not earned release.

  • No "maximum performance" without a technique ladder. The best result often comes from composition: cleaner data + stronger retrieval/tool contracts + calibrated judge + small test-time search + selective post-training. Test the cheapest credible lift before moving weights.

  • Optimize behavior, not hidden knowledge. Fine-tune for stable formatting, domain style, tool-use patterns, rubric-following, or compact specialized behavior. Use retrieval/context for facts that change or must be cited.

  • Locate variability before changing the gate. Regrade frozen outputs to separate judge instability from system unreliability; keep real stochastic failures in the reliability denominator.

  • Held-out or it's contaminated. If tuning ever saw the eval cases, the scores are inflated.

  • Check that the proxy still tracks the outcome. Tuning to a metric can weaken its relationship with user outcomes. Re-anchor it against fresh human labels or production evidence when the model, grader, or traffic changes.

  • Match uncertainty to the claim. Population-level pass rates, win rates, and judge-human agreement need design-compatible uncertainty. Exact counts over a fixed deterministic suite describe that suite and need their denominator and limits; do not invent sampling intervals for them. See references/eval-statistics.md.

  • Eval framework APIs (inspect-ai, lighteval, Ragas, DeepEval, promptfoo, Braintrust) change across releases. Verify current API and version against official docs before recommending a specific call or flag.

  • Optimization-method papers from arXiv are often preprint-only and benchmark-sensitive. Treat unreplicated methods as validate, not promote, until they beat a strong local baseline with cost/latency/safety gates.

  • Judge-bias findings (position, length, self-preference) are well-replicated, but their magnitudes (and for length, the direction) are judge- and prompt-dependent — re-measure on your own setup; do not quote a fixed number as universal.

  • Fine-tuned-judge models (Prometheus, JudgeLM, and successors) and jailbreak attack/defense results move fast — verify the current model, license, and benchmark-agreement claims before recommending a specific judge or asserting a model is robust to a given attack family.

  • Indirect prompt injection is a leading agentic attack class; treat any "the agent is safe against injection" claim as requiring fresh adaptive testing.

  • Statistics methods (bootstrap, McNemar, FDR, power/MDE) are stable, but verify the exact API (scipy/statsmodels) before copying a call.

Release Decision Gate

Select foundations only when they change the eval decision

  • If a proxy score is being interpreted as quality, trust, fairness, or another construct, or compared across populations or grader versions, use measurement theory. Return the construct, observed proxy, interpretation evidence, comparability limits, and missing validation. Skip this audit for a routine execution of an already documented instrument with unchanged interpretation and population.
  • If uncertainty, dependent repeats/clusters, multiple comparisons, or repeated stopping can change a conclusion, use statistical inference. Return the estimand, independent unit, interval method and assumptions, and the multiplicity/stopping rule. Skip inference for reporting exact outcomes of a fixed deterministic fixture suite without a population claim; report its denominator and scope instead. Neither foundation substitutes for this skill's dataset, grader controls, or release decision.

Write the release rule before scoring: required tasks and slices, minimum acceptable effect, uncertainty method, regression budget, and handling for grader errors or missing results. A candidate passes only when every blocking slice is present and the predeclared decision rule clears. Learned, model, heuristic, and human graders must pass planted-good and planted-bad controls that detect calibration drift; deterministic exact-match, schema, executable-test, and invariant graders instead need versioned implementations with relevant positive and negative fixtures. Missing denominators, stale fingerprints, failed grader controls, or an unavailable judge produce inconclusive, never an implicit pass. For memory-skill effectiveness, use the paired ablation procedure; public development fixtures validate the harness, while an independently frozen set is required for a held-out claim. Keep measured results separate from the release recommendation.

Known Traps

  • Grading an agent with the same model that produced the output (self-preference)
  • Comparing two candidates in fixed order and trusting the winner (position bias)
  • Copying a >95% threshold from a blog without validating it on your data
  • One judge call per request with no cheap deterministic pre-filter (cost blowup)
  • Quarantining genuine stochastic system failures until the gate becomes green
  • Generating a synthetic testset from the same docs used to tune the system
  • Reporting a builder-written eval as held-out for a retrieval/KB/memory system, or letting the eval fall back silently when the index is stale (see references/retrieval-and-memory-eval.md)
  • Reporting "all passed" when some cases were skipped or errored (silent success)
  • Claiming "A beats B" from a point estimate with no confidence interval or test
  • Gating a small regression on a set far too small to detect it (no power check)
  • Tuning safety to block harm without a benign set, so the model over-refuses
  • Trusting a seed for reproducibility through a hosted API that isn't deterministic
  • Fine-tuning because the prompt is messy, the retrieval is broken, or the tool contract is ambiguous
  • Declaring the fine-tune better from training loss, validation loss, or one cherry-picked demo instead of a paired held-out eval with CIs
  • Letting the training set, grader calibration set, and release gate share cases
  • Training a judge or reward model on labels produced only by the same model family it will later grade

Navigation

Resources:

Related skills:

Learnings Loop

Consult learnings.consolidated.md for relevant prior decisions or pitfalls; open learnings.md only when their history is needed. Skip both when unrelated to the task. After applying it, append one dated bullet to learnings.md via agents-skills-feedback-loop/scripts/append_learning.py if you hit a pattern, mistake, or surprising fact. Do not modify SKILL.md itself.