regression-gate — Eval Set + Metric + Threshold, Wired Into CI

> "Every subsequent change must be gated on these numbers." > — Synthesized from llamaindex]] Stage 2 (eval loop before optimizing) llamaindex-sop-skill/SKILL.md:114-126]

> "Compiled program beats baseline on a held-out test set (not the val set used in optimization)." > — dspy]] Stage 3 exit criterion dspy-sop-skill/SKILL.md:101]

This is an enhancement overlay. The regression-gate SOP exists only as fragments scattered across base skills — llamaindex]] OP-10 EvalLoop ("gate every change"), dspy]] train/dev/test split + metric — and is never assembled as a standalone cross-framework discipline. It is the discipline that turns a one-off eval into a gate: a test suite that runs in CI on every prompt/model/retriever change and fails the build on regression. It consumes a metric from [[agentsop-metric-design]] and, for domain-specific held-out sets, hands off to [[agentsop-domain-eval-set]].

1. 何时激活 (When to Activate)

Activate when any of these is true:

• Any prompt change you want to ship safely: a prompt edit, a system-message tweak, a few-shot-demo swap is about to merge and you have no automated way to know if it made things worse.
• Any model change: swapping GPT-4o → a cheaper/newer model, a temperature change, a provider migration. An LM change silently shifts the whole output distribution. dspy]] Case B: "If you optimize a complex pipeline for GPT-4, it usually breaks on a smaller model" dspy-sop-skill/SKILL.md:184].
• Any retriever/chunking/reranker change in a RAG pipeline: every such change needs a quantitative gate. llamaindex]] OP-10: "Quantitative regression test for every chunking / embedding / retriever / prompt change" llamaindex-sop-skill/SKILL.md:235].
• Recurring "it got worse" surprises: the team keeps shipping changes that users report as regressions after the fact. The fix is a gate, not more careful review.
• Setting up CI for an LLM app and there is no eval job in the pipeline.

Do NOT activate for:

• One-shot tasks with no production surface — there is nothing to regress. (dspy]] boundary: "Summarize this email once → raw API call" dspy-sop-skill/SKILL.md:259].)
• The signature/task is still changing daily — gate only after the I/O contract stabilizes, else you re-baseline every commit. (dspy]] boundary dspy-sop-skill/SKILL.md:260].)
• No willingness to define any success criterion — without a metric there is nothing to gate. Route to [[agentsop-metric-design]] first; if the user refuses, this skill cannot help.

2. 核心心智模型 (Core Mental Model)

"An LM change is a code change. Gate it with a test suite: eval set + metric + threshold."

You already gate code with unit tests in CI: a change that breaks a test fails the build. A prompt edit, a model swap, a chunk-size tweak are also changes to the system's behavior — but they slip through review because their effect is statistical, not a stack trace. The regression gate is the missing unit test for LM behavior.

The gate is exactly three artifacts plus a wiring step:

   eval set        metric           threshold          CI wiring
  (held-out QA)  (ex,pred)->score   (fail if <X / drop>Y)  (block merge)
        │              │                  │                    │
        └──────────────┴──────────────────┴────────────────────┘
                          REGRESSION GATE

Three load-bearing principles:

1. The eval set is held out and frozen. It is a labelled, version-controlled fixture that the prompt/model under test has never seen. dspy]] is explicit: the compiled program must beat baseline on a held-out test set "not the val set used in optimization" dspy-sop-skill/SKILL.md:101]. The split is train / dev / test; the gate runs on test only. If the eval set leaks into the prompt (few-shot demos, instructions), the gate measures memorization, not quality.

1. The metric comes from [[agentsop-metric-design]], not invented here. This skill does not design metrics — it consumes one. A bad metric makes the gate theatre: it will pass changes that hurt users and block changes that help them. The metric must be human-calibrated before it gates anything (agentsop-metric-design]] OP-M05).

1. The threshold is a policy, not a number you guess. Two common shapes: an absolute floor (fail if score < X) and a relative no-regression (fail if score drops > Y from the committed baseline). Relative is the regression gate proper; absolute is a quality bar. Most teams use both: a floor for "never ship below this," plus a no-regression delta for "this PR must not make it worse."

Build the eval loop before you optimize anything

llamaindex]] Stage 2 is named "Build the eval loop before optimizing anything" llamaindex-sop-skill/SKILL.md:114]. The anti-pattern it names is A3: "No eval loop; debug by anecdote" llamaindex-sop-skill/SKILL.md:348]. The gate is the institutional form of that loop — once it exists, every change is debugged by number, not by vibe.

3. SOP (Standard Operating Procedure)

0. Confirm activation (§1); confirm a metric exists or invoke [[agentsop-metric-design]]
1. BUILD eval set:  generate candidates -> curate to a golden set -> freeze + version
2. SPLIT:           train / dev / test; the GATE runs on TEST only
3. PICK metric:     consume from [[agentsop-metric-design]] (do not invent here)
4. SET threshold:   absolute floor AND/OR relative no-regression delta
5. WIRE into CI:    run eval on every prompt/model/retriever PR; fail on regression
6. HANDLE flakiness: pin seeds/temp, average N runs, separate flaky from real drops

Stage 1 — Build the eval set (generate + curate)

Two stages, never one. Generation gives coverage cheaply; curation gives trust.

• Generate candidates from your corpus. llamaindex]] DatasetGenerator.from_documents(docs).generate_dataset_from_nodes(num=50) produces labelled QA pairs from the documents themselves llamaindex-sop-skill/SKILL.md:121]. promptfoo and synthetic-data generators do the same for non-RAG tasks.
• Curate the generated set into a golden set: a human reviews, fixes wrong labels, drops ambiguous items, and adds known hard/edge cases the generator missed. A purely generated set inherits the generator model's blind spots and tends to be too easy (Dilemma 1).
• Freeze and version. The golden set is a committed fixture (eval/golden_v1.jsonl), tagged with the date and the generator model. Changing it is a versioned event, not an edit.

Size: dspy]] documents the sweet spot — "30 examples = minimum useful, 300 = recommended" dspy-sop-skill/SKILL.md:87]. For a held-out gate, 50–200 curated domain examples is the working range; descend to [[agentsop-domain-eval-set]] for domain-specific construction.

Stage 2 — Split: train / dev / test

The gate runs on test only. Keep test sealed from anything that touches the prompt:

• train — feeds optimizers / few-shot demo selection.
• dev — tuning and threshold-setting.
• test — the gate. Never used to author prompts, pick demos, or tune. (dspy]] held-out exit criterion dspy-sop-skill/SKILL.md:101].)

Stage 3 — Pick the metric (cross-link [[agentsop-metric-design]])

Do not invent a metric here. Consume one from [[agentsop-metric-design]]:

• RAG → the Faithfulness + Relevancy + Retriever(MRR/hit-rate) triad (llamaindex]] OP-10 llamaindex-sop-skill/SKILL.md:234]).
• Exact-answer → exact-match (dspy]] §4.3 dspy-sop-skill/SKILL.md:137]).
• Open-ended → decomposed sub-judges, bool-in-compile/float-in-eval, length penalty (agentsop-metric-design]] OP-M01/OP-M02/OP-M03).

A metric that has not been human-calibrated must not gate (agentsop-metric-design]] OP-M05). An uncalibrated gate is worse than no gate — it gives false confidence.

Stage 4 — Set the threshold

| Threshold shape | Rule | Use when | |---|---|---| | Absolute floor | fail if score(test) < X | "never ship below this quality bar" | | Relative no-regression | fail if baseline − score > Δ | "this PR must not make it worse" (the gate proper) | | Per-slice floor | fail if any slice (e.g. lexical-query subset) drops > Δ | aggregate hides a regressed minority |

Set Δ above measured run-to-run noise (Stage 6), else the gate flaps. Commit the current test score as baseline.json next to the eval set; the gate compares against it.

Stage 5 — Wire into CI

• Add an eval CI job that runs on every PR touching prompts, model config, retriever/chunking config, or the program graph.
• Job: load frozen eval set → run pipeline at PR's config → compute metric → compare to baseline.json → exit non-zero on threshold breach → post the before/after table as a PR comment.
• On merge to main with an intended improvement, bump baseline.json in the same PR (reviewed, not silent).

Stage 6 — Handle flaky evals

LLM outputs are nondeterministic; a naive gate flaps and gets disabled. Mitigations:

• Pin temperature=0 and seeds where the provider supports them; disable response caching in CI (dspy]] AP-10: "Forgetting cache=False in stateless deploys" dspy-sop-skill/SKILL.md:255]).
• Average N runs (e.g. 3) and gate on the mean; report variance.
• Separate flaky from real: if the same config scores differently across reruns by more than Δ, the gate (Δ too tight) or the metric (noisy judge) is the problem — fix those before trusting a single red build. A judge-based metric with high variance should be hardened in [[agentsop-metric-design]], not papered over by widening Δ.

4. 操作模型 (Operations)

OP-01 — GenerateEvalCandidates

• Trigger: Need a held-out eval set; no labelled fixture exists yet.
• Action: Generate QA candidates from the corpus — llamaindex]] DatasetGenerator.generate_dataset_from_nodes(num=50), promptfoo synthetic generation, or task-specific synthesis. Tag with generator model + date.
• Output: A raw candidate set (unvalidated), ready for curation.
• Evidence: llamaindex]] Stage 2 llamaindex-sop-skill/SKILL.md:121]; OP-10 llamaindex-sop-skill/SKILL.md:234]; external "eval set generation".

OP-02 — CurateGoldenSet

• Trigger: A generated candidate set exists; it has not been human-reviewed.
• Action: Human reviews every item: fix wrong labels, drop ambiguous/duplicate items, inject known hard cases and past production failures. Freeze as a versioned fixture (golden_vN.jsonl).
• Output: A trusted, frozen golden eval set that the gate runs against.
• Evidence: dspy]] metric human-validation discipline dspy-sop-skill/SKILL.md:212]; agentsop-metric-design]] OP-M05; Dilemma 1.

OP-03 — SplitTrainDevTest

• Trigger: Eval set built; about to use it for both tuning and gating.
• Action: Partition into train/dev/test. Seal test from all prompt-authoring. The gate reads only test. (Note dspy]]'s reversed 20/80 train/val split for prompt optimizers dspy-sop-skill/SKILL.md:96] — that is an optimizer concern; the gate still needs an untouched test slice.)
• Output: Three disjoint splits; a sealed test set for the gate.
• Evidence: dspy]] Stage 3 held-out exit criterion dspy-sop-skill/SKILL.md:101]; AP-9 dspy-sop-skill/SKILL.md:254].

OP-04 — SetRegressionThreshold

• Trigger: Have a test score; need a pass/fail policy.
• Action: Commit current test score as baseline.json. Define absolute floor X and/or relative no-regression Δ (Δ > measured noise). Optionally per-slice floors.
• Output: A versioned threshold policy the CI job enforces.
• Evidence: dspy]] Stage 3 exit "by ≥ task-relevant delta" dspy-sop-skill/SKILL.md:101]; llamaindex]] gate-every-change llamaindex-sop-skill/SKILL.md:124].

OP-05 — WireCIGate

• Trigger: Eval set + metric + threshold exist; not yet enforced automatically.
• Action: Add a CI job triggered by prompt/model/retriever/graph changes: run eval at PR config → compute metric → compare to baseline → fail on breach → comment the before/after table.
• Output: Merges that regress quality are blocked; every change carries a number.
• Evidence: llamaindex]] OP-10 regression-test framing llamaindex-sop-skill/SKILL.md:233-235]; external "llm regression testing CI", "promptfoo".

OP-06 — BumpBaselineOnIntendedWin

• Trigger: A PR intentionally raises quality; the gate would otherwise pin the old baseline forever.
• Action: In the same reviewed PR, update baseline.json to the new test score. Never let CI auto-bump silently.
• Output: Baseline ratchets upward deliberately; future regressions are caught against the new bar.
• Evidence: dspy]] "keep both program.gpt4o.json and program.llama8b.json; A/B" dspy-sop-skill/SKILL.md:191] (versioned-artifact discipline).

OP-07 — StabilizeFlakyEval

• Trigger: The gate flaps — same config, different verdicts across reruns.
• Action: Pin temperature/seed, disable CI caching, average N runs and gate on the mean, report variance. If variance > Δ, fix the metric (harden judge via [[agentsop-metric-design]]) or widen Δ — never disable the gate.
• Output: A stable gate whose red builds are trustworthy.
• Evidence: dspy]] cache=False AP-10 dspy-sop-skill/SKILL.md:255]; agentsop-metric-design]] judge-bias hardening; dspy]] Stage 2 exit "stable across two runs" dspy-sop-skill/SKILL.md:91].

OP-08 — SliceTheEvalSet

• Trigger: Aggregate score is flat but a subpopulation (lexical queries, a tenant, a topic) silently regressed.
• Action: Tag eval items by slice; compute and gate per-slice. A win on the majority must not mask a regression on a minority slice.
• Output: Slice-level regression detection; aggregate no longer hides harm.
• Evidence: llamaindex]] per-query-type taxonomy llamaindex-sop-skill/SKILL.md:280]; agentsop-metric-design]] triad (no single number).

5. 困境决策案例 (Dilemma Cases)

Dilemma 1 — Generated eval set vs hand-curated golden set

困境: Team needs an eval set fast. DatasetGenerator produces 200 QA pairs in minutes (llamaindex]] Stage 2 llamaindex-sop-skill/SKILL.md:121]). Hand-curating 200 examples costs days of human time. Ship the generated set as the gate, or pay for curation?

约束: The generator is the same model family that powers the pipeline → its questions are answerable by exactly the kind of reasoning the pipeline already does (self-preference leakage). Generated sets skew easy and miss the long-tail failures users actually hit. But zero eval set means shipping blind (anti-pattern).

决策步骤:

1. Generate for coverage, never gate on raw generation. Use the generated set as a candidate pool, not the gate.
2. Curate a golden subset (OP-02): a human keeps the good items, fixes labels, drops the trivially-easy and the ambiguous, and injects known production failures and adversarial/edge cases the generator never proposes.
3. Cross-family generation where possible: generate with a different model family than the task model to reduce self-preference leakage (mirrors agentsop-metric-design]] OP-M04).
4. Size to budget: 50 curated > 200 raw. dspy]] floor is 30 useful examples dspy-sop-skill/SKILL.md:87]; a tight, curated, edge-case-loaded 50 gates better than a bloated easy 200.
5. Version both: keep the raw generated pool (regeneratable) and the frozen golden set (the gate fixture).

结果: The golden set is the gate; the generated pool is scaffolding. Teams that gate on raw generated sets ship regressions that the easy set never exercised — the gate was green while users churned.

可提取的操作: OP-01 GenerateEvalCandidates, OP-02 CurateGoldenSet. Lesson: generation buys coverage, curation buys trust. A gate needs trust — never gate on an uncurated generated set.

Dilemma 2 — Threshold too strict blocks good changes

困境: A no-regression gate is set at Δ = 0 (any drop fails). A genuinely good refactor — simpler prompt, 40% cheaper model — scores 0.81 vs the 0.83 baseline: a 2-point drop within run-to-run noise. The gate blocks a change that is net-positive (equal quality, far cheaper). The team starts overriding the gate, and soon ignores it entirely.

约束: Run-to-run noise on this judge-based metric is ±1.5 points (measured across 3 reruns). The 2-point "drop" is statistically indistinguishable from noise. A gate that flags noise as regression trains the team to bypass it — a bypassed gate is worse than none.

决策步骤:

1. Measure noise first (OP-07): rerun the baseline config N times; compute the standard deviation. Here σ ≈ 1.5pp.
2. Set Δ above noise: Δ = 2σ ≈ 3pp, not 0. A drop must clear the noise band to count as a regression.
3. Average N runs and gate on the mean to shrink the noise band, rather than just widening Δ.
4. Separate cost from quality: this PR is a cost win at equal quality. The quality gate should pass (drop within Δ); cost is tracked on its own axis (cross-link [[agentsop-cost-tiered-models]]). Don't let a quality gate block a cost win that doesn't hurt quality.
5. If the metric is too noisy to set a sane Δ, the metric is the bug — harden it in [[agentsop-metric-design]] (decompose, length penalty, cross-family judge), don't widen Δ to infinity.

结果: Δ tuned to ~2σ passes the cheaper-equal-quality change, still catches real regressions (a 6pp drop), and the team keeps trusting the gate. A gate calibrated to noise survives; a Δ=0 gate gets disabled.

可提取的操作: OP-04 SetRegressionThreshold, OP-07 StabilizeFlakyEval. Lesson: the threshold must clear measured noise. A gate that flags noise as failure gets bypassed, and a bypassed gate protects nothing.

6. 反模式与边界 (Anti-Patterns & Boundaries)

Anti-patterns

| # | Anti-pattern | Why it's wrong | Fix | |---|---|---|---| | AP-1 | No eval set; ship blind | Every prompt/model change is an uncontrolled experiment on users; "it got worse" is discovered in production | Build a held-out gate (llamaindex]] A3 llamaindex-sop-skill/SKILL.md:348]) | | AP-2 | Eval set leaks into the prompt (few-shot demos / instructions drawn from test) | The gate measures memorization, not generalization; green build, real regression | Seal test; demos come from train only (OP-03; dspy]] AP-9 dspy-sop-skill/SKILL.md:254]) | | AP-3 | Gate on a raw generated set | Inherits generator blind spots; too easy; misses real failures | Curate a golden set (OP-02; Dilemma 1) | | AP-4 | Gate on an uncalibrated metric | A wrong metric passes harmful changes and blocks good ones — gate is theatre | Calibrate via [[agentsop-metric-design]] OP-M05 before gating | | AP-5 | Δ = 0 / threshold below noise | Gate flaps on noise, team bypasses it | Set Δ > 2σ measured noise (OP-04, OP-07; Dilemma 2) | | AP-6 | Run the gate on the val/dev set used for tuning | Optimistic, leaks tuning into evaluation | Gate on held-out test only (dspy]] dspy-sop-skill/SKILL.md:101]) | | AP-7 | Caching on in CI | Stale cached outputs mask the change under test | cache=False (dspy]] AP-10 dspy-sop-skill/SKILL.md:255]) | | AP-8 | Aggregate-only gate | A win on the majority hides a regressed minority slice | Per-slice gating (OP-08) | | AP-9 | Silent baseline auto-bump | Quality can ratchet down unnoticed if CI rewrites baseline | Bump baseline only in a reviewed PR (OP-06) | | AP-10 | Disabling the gate when it flakes | Removes the only protection; flakiness is a metric/Δ bug, not a gate bug | Stabilize (OP-07), never disable |

Boundaries (when this skill does not apply)

• One-shot / throwaway tasks — no production surface to regress (dspy]] dspy-sop-skill/SKILL.md:259]).
• Unstable signature — gate only after the I/O contract stabilizes (dspy]] dspy-sop-skill/SKILL.md:260]); otherwise you re-baseline every commit.
• No metric possible and none willing to be built — without a metric there is nothing to gate; route to [[agentsop-metric-design]] first.
• Public-benchmark evaluation (MMLU, HumanEval) — that is a capability benchmark, not a domain regression gate; use lm-evaluation-harness. For a domain-specific held-out set, descend to [[agentsop-domain-eval-set]].

7. 跨框架对照 (Cross-Framework Mapping)

| Concept | LlamaIndex | DSPy | promptfoo | LangSmith | This skill | |---|---|---|---|---|---| | Eval set generation | DatasetGenerator.generate_dataset_from_nodes(num=N) llamaindex-sop-skill/SKILL.md:121] | bring labelled examples; BootstrapFewShot self-generates demos (not the test set) | tests: synthesis / generate from prompts | Datasets created from traces / uploads | OP-01 GenerateEvalCandidates | | Golden / curated set | manual review of generated QA | hand-labelled trainset/devset | curated tests YAML with assert | curated Dataset + reference outputs | OP-02 CurateGoldenSet | | Train/dev/test split | manual | explicit; reversed 20/80 for prompt optimizers, held-out test for gate dspy-sop-skill/SKILL.md:96,101] | n/a (test set is the suite) | dataset splits | OP-03 SplitTrainDevTest | | Metric | Faithfulness/Relevancy/RetrieverEvaluator(mrr,hit_rate) llamaindex-sop-skill/SKILL.md:234] | def metric(ex,pred,trace=None)->bool\|float dspy-sop-skill/SKILL.md:88] | assert (equals/contains/llm-rubric/javascript) | evaluator fns / LLM-as-judge | consumed from [[agentsop-metric-design]] | | Threshold / gate | manual (gate every change llamaindex-sop-skill/SKILL.md:124]) | held-out beats baseline "by ≥ delta" dspy-sop-skill/SKILL.md:101] | assert pass + --fail-on thresholds | rules + alerts on eval scores | OP-04 SetRegressionThreshold | | CI wiring | not built-in (DIY job around Evaluate) | not built-in (DIY around dspy.Evaluate) | first-class: promptfoo eval in CI, non-zero exit | CI integration + regression alerts | OP-05 WireCIGate | | Flaky handling | run multiple times | cache=False; "stable across two runs" dspy-sop-skill/SKILL.md:91,255] | repeat + threshold | run aggregation | OP-07 StabilizeFlakyEval |

Combination patterns:

• LlamaIndex + this skill: DatasetGenerator for OP-01, the Faithfulness/Relevancy/Retriever triad as the metric, wired into a DIY CI job. The base skill names the loop ("gate every change"); this skill makes it a CI gate.
• DSPy + this skill: DSPy's split + metric is the eval-loop substrate; this skill adds the CI enforcement DSPy leaves to you. Gate the compiled artifact on the held-out test set; bump baseline when recompiling for a new model (Case B).
• promptfoo as the engine: promptfoo is the most CI-native option — promptfoo eval exits non-zero on failed asserts; it is the closest off-the-shelf realization of OP-05. Use it as the runner; still bring a curated set (OP-02) and a calibrated metric.
• LangSmith for managed datasets + monitoring: managed datasets and regression alerts; pairs with this skill's curation/threshold discipline.

Opinionated default: build the eval set with the base framework's generator (OP-01), curate by hand (OP-02), keep the metric in [[agentsop-metric-design]], and run the gate with promptfoo (CI-native) or a thin script around dspy.Evaluate / LlamaIndex evaluators. The gate, the metric, and the eval set are three separable, version-controlled artifacts — never one tangled blob.

References

• references/R1-source-evidence.md — every cited claim resolved to a source line
• intermediate/operation_candidates.json — machine-readable operation registry

Citations: llamaindex]] OP-10 EvalLoop / Stage 2 llamaindex-sop-skill/SKILL.md:114-126,232-236,348]; dspy]] Stage 2-3 split+metric+held-out dspy-sop-skill/SKILL.md:85-105,137,191,254-255]; agentsop-metric-design]]; agentsop-domain-eval-set]]; external "llm regression testing CI", "promptfoo", "eval set generation".