技能 编程开发 规范驱动功能实现评估

规范驱动功能实现评估

v20260619
spec-driven-eval
该工具用于评估功能实现是否彻底满足了产品需求文档(PRD)或规范。它按单个案例评分,独立评估代码的实现程度和配套的测试覆盖率(单元测试和端到端测试)。适用于对多个开发成果进行基准测试、审计功能是否完全符合规范,或验证功能是否已完整实现。
获取技能
437 次下载
概览

Spec-Driven Implementation Evaluation

Evaluate a spec-driven-development (SDD) effort against a PRD case by case (acceptance criterion by acceptance criterion), scoring implementation and tests separately, and roll up to a single comparable final grade. Designed for benchmarking: the same PRD evaluated across different SDD frameworks — and the same effort evaluated twice — must yield comparable, reproducible numbers.

Two subjects, two questions

This eval answers two independent questions, and keeps their verdicts separate because they fail independently:

  1. How good is the framework at respecting and extracting requirements? — Did it honor the PRD and stay in bounds (respect: Final implementation side + Scope S), and did it surface the implicit requirements the PRD only implied, without noise (extract: Elicitation E)?
  2. How good is the harness at ensuring they are all implemented? — Does the test suite prove every sanctioned requirement is actually built (T + Engineering Gates G)?

The linkage that makes "all implemented" well-defined: the harness is held accountable for the full sanctioned requirement set = PRD acceptance criteria ∪ valid E-additions. A valid requirement the framework extracted but the harness never tests is a harness miss (T), not a framework miss (E keeps the extraction credit). Extraction defines the verification target.

Scoring is checklist-based: every criterion is decomposed into atomic binary (MET / UNMET) checks, each backed by file:line evidence. Binary decomposition is the design choice that makes the grade reproducible — graded/Likert scales (is this a 3 or a 4?) are the dominant source of evaluator-to-evaluator disagreement; binary checks raise inter-evaluator agreement to roughly human level. Partial credit is derived from the fraction of checks met, never judged on a sliding scale.

When to use

  • "Evaluate/score this PRD case by case and give a final grade"
  • "Was this story/feature implemented 100%?"
  • Benchmarking multiple spec-driven implementations of the same PRD
  • Auditing implementation and test coverage (unit + e2e) against acceptance criteria

Inputs required

  1. The PRD (ground-truth product intent) — the user story acceptance criteria are the unit of evaluation.
  2. The implementation (production code).
  3. The tests (unit + e2e).
  4. The SDD-derived artifactsspec.md and tasks.md (refined ACs, requirement IDs, derived requirements). Required for the Elicitation E and Scope S axes (they grade these against the PRD). When absent, Final/T still run, but report E/S as n/a — no derived spec. The PRD remains the source of truth for what counts as "expected"; the derived spec is what gets graded for respect and extraction.

Scoping the diff. Before scoring, use git to identify which files changed for this implementation. That diff surface is the primary search scope for all file:line evidence in steps 4 and 8:

git diff <base>..<head> --name-only   # branch or PR
git diff --name-only HEAD             # uncommitted changes
git status --short                    # include untracked new files

Record the diff surface in the report. Evidence outside it is still valid (e.g. a pre-existing file was modified), but note when a check relies on files not in the diff — that may indicate the wrong base was chosen.


Quick start

New to running this end-to-end? See quickstart.md for the 4-chat-session flow (freeze baseline → plan → implement → evaluate) with paste-ready prompts. Read it when you need the operational how-to; the rest of this file is the scoring methodology.


Core rules (read first)

These five rules govern every score. They exist to make the grade auditable and reproducible.

  1. Evidence or zero. Every MET check MUST cite evidence as file:line (or file:startLine-endLine). No located evidence ⇒ the check is UNMET. Never award credit from assumptions or from the PRD restating intent.
  2. Search-before-zero (anti false-negative). Before marking a check UNMET for "not found", record the search actually performed — starting with the diff-surface files from the scoping step, then the grep/glob terms tried and any additional files/dirs inspected. A check scored UNMET must confirm the behavior is absent from the diff surface, not just from an ad-hoc grep across the full repo. UNMET means searched and genuinely absent, not did not look. If no search is shown, the check is not yet scored, not UNMET.
  3. Read the path end-to-end — including the data shape. A check is MET only if you traced the real code/test path, not because a symbol name matches. For an emitted/returned/persisted artifact, inspect the constructed payload object itself, not just the call site — a present emit(...)/return ... does not prove the named field is in the payload (see the Conjunction rule). A test that exercises a behavior but does not assert it does not meet a verification check.
  4. Judge ≠ author for benchmarks. When grading to compare implementations, the evaluating model should differ from the model that wrote the code; LLM judges over-reward their own output (self-preference bias). If they are the same, flag it under Assumptions and treat borderline checks as UNMET.
  5. The evaluator is read-only over the subject. Never modify the code under evaluation — not to fix a failing gate, not to "help", not even for trivial type errors. A red gate stays red: record , apply the Adjusted Final, and put the required correction in the report's fix list. Modifying the subject mid-evaluation contaminates the benchmark (the grade no longer measures the framework's output) and invalidates the diff surface. If a correction was accidentally applied, revert it and score the original state.

Reproducibility rules

The grade must be stable: the same PRD + same commit must yield the same Final across runs and evaluators. Obey all of:

  1. Freeze the AC list AND its checklist (the single biggest drift source). AC enumeration and check granularity are decided once and pinned, because both I, T, and the Story_score denominator depend on them:
    • If spec.md with stable requirement IDs exists, the AC list and IDs are frozen — use them verbatim; do not re-split or merge.
    • Derive the binary checklist for each AC once and persist it to <spec-folder>/evaluations/_ac-baseline.md (IDs + verbatim AC text + the I-checks and T-checks). Every implementation of that PRD is scored against the identical checklist.
    • One AC = one testable assertion of intent; one check = one atomic, observable yes/no proposition. Do not collapse distinct behaviors into one check, nor split one behavior across several.
  2. Priority comes from the PRD, never inferred silently. Use the PRD's explicit P0/P1/P2 labels. If a story is unlabeled, mark its priority ASSUMED in the report and list it under Assumptions; never let an inferred priority silently move the 3/2/0 weights.
  3. Round once, at the end. Carry full precision through the arithmetic; round only the reported AC_score, Story_score, and Final to 2 decimals. Band assignment uses the rounded Final.
  4. Compute, don't calculate. The roll-up (Σw, every Story_score, Final, Adjusted Final) MUST be computed by executing a script (e.g. node -e / python3 -c) that takes the per-AC I/T fractions and the priority-weight table as input, with the script's output pasted into the report. Never do the arithmetic mentally or by hand — a hand-summed denominator (Σw) is a known failure mode that silently shifts the grade band. Σw must be derived inside the script from the same priority table shown in the report, not typed in as a literal.
  5. Self-consistency ensemble (k = 3). Evaluate the checklist three times independently at low/zero temperature and take the majority MET/UNMET per check before computing any number. Majority voting over k=3 removes most stochastic flips at low cost. If the three passes disagree on a check, that check is borderline — keep the majority verdict and note it; persistent disagreement means the check wording is ambiguous (sharpen it in the baseline, see Calibration).

Scoring model

Unit of scoring: the acceptance criterion (AC), decomposed into binary checks

For each in-scope AC, the frozen checklist holds two sets of atomic checks, each scored MET (1) / UNMET (0):

Implementation checks (I-checks) — one per distinct observable behavior the AC asserts. A check is MET only if production code performs that behavior, evidenced file:line. Include only behaviorally-observable clauses (a verb the AC states: creates / returns / rejects / persists / grants / defaults-to). Non-functional polish (wording, logging) is not a check — note it separately so it never moves the score.

When parsing an AC into checks (done once, at baseline-freeze time), apply both rules below so verbs aren't the only thing captured:

Conjunction / payload-field rule. When an AC enumerates multiple items — joined by "and", ",", "with [field]", "including"especially in the subject or payload of an emitted/returned/persisted artifact, treat each named field or entity as its own I-check. Do not stop at confirming the parent method is called: for each named field, open the actual data object constructed at file:line and verify the field is present in it (score against the payload shape, not the call site). Example: "emit a trigger associated with the user and the trial end date" → two I-checks: (1) trigger payload carries userId, (2) trigger payload carries trialEndsAt. Confirming emit(...) is reached does not satisfy check (2) if the payload object omits the field.

Disjunction / product-chosen rule. When an AC presents alternatives ("A or B", "pause or cancel", "immediately or at period end"), decide the reading once and freeze it in the baseline (the reading itself must not vary per run):

  • Independent paths — both are separately reachable behaviors (e.g. "cancel immediately or at period end"): one I-check per path, each independently MET/UNMET.
  • Product-controlled / configurable — the AC frames the choice as product-driven (e.g. "apply the product-chosen behavior: pause or cancel"): two I-checks — (1) the recommended/default behavior is implemented, and (2) the alternative is reachable without a code change (config key / flag / env). Hard-coding one option with no switch to the other ⇒ check (2) UNMET. A different feature that happens to reach the other option (e.g. user-initiated cancel) does not satisfy (2) — it must be the same product-controlled decision point.
  • If "product-chosen" is itself ambiguous between runtime-configurable and design-time chosen, resolve it against the PRD/spec wording and record the resolved reading in the baseline; only require reachability of the non-default path when the AC frames the choice as runtime/product-controlled.

Wiring / ingress rule (asynchronously-delivered side-effects). When an observable side-effect is delivered asynchronously through an inbound event (e.g. a Stripe webhook → handler → persisted status), add a dedicated wiring I-check in addition to the behavior I-check: the entry endpoint actually receives, verifies, and dispatches the inbound event — by type — to the handler. This is distinct from the T-outcome test that the handler produces the right state: a fully-tested handler sitting behind a dead, unregistered, or mis-routed endpoint still fails the wiring check. Without it, a correct-but-unreachable handler would earn full credit (the blind spot the T-outcome entry-point-neutrality opens up — letting the test bypass HTTP means something else must prove the route exists). Score the wiring I-check against the controller/route that maps the event type to the handler (signature/authenticity verification + dispatch), not against the handler internals. Freeze a wiring I-check on every AC whose observable effect arrives via an inbound async event.

I = (# I-checks MET) / (# I-checks)

Test checks (T-checks) — verification checks at the required level(s). Required levels are fixed by policy, not judged per-AC:

  • pure / business logic ⇒ unit required
  • observable HTTP / contract / persistence side-effect ⇒ e2e / integration required
  • an AC with both ⇒ both required

T-outcome — persistence / async side-effects (outcome-based, entry-point-neutral). When the AC asserts a resulting state — phrased "results in / persisted / displayed status / reflects the change / updates the status" — the verification check is on the real resulting state, not on a call. Label these checks T-outcome (distinct from a plain synchronous-response T-e2e). The policy has four parts:

  • Assert the real outcome. MET requires asserting the actual persisted state (a real DB row) or the actually-returned payload, against real infra. A test that only exercises the path without asserting the resulting state is UNMET (Core rule 3).
  • Entry point is the implementation's choice (entry-point-neutral). Drive the real component via any entry point and assert the resulting state: the HTTP endpoint, OR by invoking the handler / queue-consumer / service method directly against the real DB. An async (ack-fast-then-process) design need not be tested through the full HTTP→queue→worker path — awaiting the handler and asserting the persisted row is sufficient and equivalent. Do not penalize a queue/async architecture for not being driven end-to-end through HTTP; the question is whether the real resulting state is asserted, not which door the test entered.
  • Mock-only exclusion (scoped precisely). Asserting that a method was called on a mocked repository / state-machine (expect(save).toHaveBeenCalled() / expect(transition).toHaveBeenCalledWith(...)) proves a call, not a persisted outcome ⇒ UNMET for any "results in / persisted / displayed status" check (it may still count toward Robustness R). The exclusion is scoped to outcome checks only — it must not over-fire: a check whose proposition is "invokes the external API" (e.g. "immediate cancel calls StripeClient.cancel", "creates the subscription in Stripe") is correctly verified by a spy/mock, because the asserted proposition is the call. Mock-only is fine there; the exclusion does not apply.
  • Minimum-level floor. The required level is a floor, not an exact match. A stronger-level test (an e2e/integration test covering a unit-required proposition) satisfies that proposition, provided the specific proposition (e.g. a named payload field, a default value) is actually asserted on the real artifact. Do not mark a unit-required check UNMET solely because the only assertion of it lives in an e2e/integration test.

For each required level, derive binary checks such as: primary behavior asserted at this level (MET only if asserted, not merely exercised), relevant negative/error case asserted, each secondary clause asserted. Mocked external systems (e.g. Stripe via nock) cannot prove the external side — write the check against the observable local contract and note the limitation.

The conjunction rule applies to T-checks too: each named payload field gets its own verification check. A test that asserts the artifact was emitted/returned but does not assert the field's presence/value is UNMET for that field's check (asserting emit was called ≠ asserting the payload carries trialEndsAt).

T = (# T-checks MET across all required levels) / (# T-checks)

Both I and T land on a fine-grained value, but every underlying atom is a binary, evidence-backed yes/no — that is what makes them reproducible. There is no "minor vs meaningful" judgment: a clause either has its own check or it does not, and that decision was frozen in the baseline.

Per-AC, per-story, final

AC_score    = 0.6 * I + 0.4 * T
Story_score = mean(AC_score over the story's ACs)      # ACs equally weighted by default
Final       = Σ(w_s * Story_score) / Σ(w_s)            # weighted by story priority

Story priority weights w_s:

Priority Weight Rationale
P0 3 Core MVP
P1 2 MVP-adjacent / launch
P2 / out-of-scope 0 Excluded from the grade; report separately as roadmap readiness

The 0.6/0.4 implementation/test split reflects "working software first, proven by tests". Keep these weights — and the P0/P1/P2 weights — fixed across the benchmark; changing them breaks comparability.

Grade bands

Final Band
≥ 0.90 Spec-complete
0.75–0.89 Strong (minor gaps)
0.60–0.74 Partial (meaningful gaps)
0.40–0.59 Weak
< 0.40 Inadequate

Bias controls

LLM judges have well-documented biases; hold these explicitly:

  • Verbosity / quantity bias — more code or more tests never raises I or T. Only checks that map to an AC clause count. Surplus tests go to R/D (below), never into the grade.
  • Self-preference bias — see Core rule 4 (judge ≠ author).
  • Anchoring — score against the frozen baseline and the calibration anchors in references/reference.md, not against the previous implementation you happened to grade.

Reported beside the grade (NOT folded into Final — keep comparable)

  • Robustness Index R — extra tests beyond PRD cases, weighted High=1.0 / Med=0.5 / Low=0.25, summed. Signals defensive quality. Never inflates Final.
  • Scope Adherence Spass / partial / fail, governed by the traceability principle: every built behavior must trace to a sanctioned source (a PRD acceptance criterion → Final, or a valid E-addition). Three checks, each per built behavior:
    • PRD-boundary — built something on the PRD's explicit out-of-scope list ⇒ fail.
    • Rogue build — built something that traces to neither a PRD AC nor a valid E-addition (untraceable / invented) ⇒ fail.
    • Plan driftspec.md/tasks.md sanctioned a behavior the code didn't build (or left half-built / inconsistent) ⇒ partial. Everything traces cleanly ⇒ pass. Report the failing/partial behaviors with file:line. A valid requirement the framework derived into the spec but correctly did NOT build (because it is out of scope) is good discipline — it is not an S penalty; record it as deferred-valid under E.
  • Engineering Gates Gbuild, lint, unit, e2e. Each gate must be an actually-executed / or an explicit not-run (with reason, e.g. e2e infra unavailable). Never report a gate as passing without running it. The build gate is pinned mechanically to remove evaluator discretion: pick the project's canonical build command once per benchmark (the build script the repo documents) and define the gate as that command exiting 0 — nothing else decides it. Record the exact command in the report so every run uses the identical gate. A pre-existing, unrelated toolchain failure in code the change didn't touch (e.g. a known typecheck error in an untouched migration path) is a documented non-graded NOTE, not a : it does not gate build and does not trigger Adjusted Final. Record it in the report as a known note, but the build verdict follows only the pinned build command's exit code. Probe-before-not-run: a gate may only be reported not-run after a recorded real attempt — the command executed (or the prerequisite check, e.g. docker ps for e2e infra) and its error output pasted as evidence. A reason without an attempt is not valid; the gate stays not yet scored, exactly like an UNMET without a search (Core rule 2 applied to gates). Only a confirmed-red gate () triggers Adjusted Final = Final × 0.5; a not-run gate cannot grant or deduct credit — it is reported as a known blind spot. The unadjusted Final stays reported for comparison. A red gate is never fixed by the evaluator (Core rule 5) — record , adjust, and list the fix.
  • Test Distribution D — every feature test classified into one of three tiers, reported as counts and % of the suite. Shows where the testing effort went; never inflates Final. See below.

Test distribution by tier D (reported beside)

Classify every added feature test (each it/test case, counting it.each rows individually) into exactly one tier. The classification reuses work already done in the test-check and robustness steps, so it is cheap and reproducible:

Tier Definition Maps to
Necessary (primary happy path) Asserts the primary success path of a P0/MVP AC, or a dependency required for that path to work. A T-check MET on a P0 main flow
Secondary (important) Maps to an AC but is not the P0 primary path: P1 ACs, edge/negative/error paths, secondary clauses, security/authz, duplicate/idempotency/dedup, status-trigger emission. A T-check MET that is not Necessary
Nice-to-have Maps to no AC: defensive/exhaustive coverage (enum mapping, unknown-status, repo/infra internals, helper methods) or unit coverage redundant with an e2e-proven path. The Robustness R extras

Rule of thumb: AC-mapped tests split into Necessary + Secondary; the Robustness R inventory is the Nice-to-have tier. A test belongs to exactly one tier — do not double-count.

Report D as a table of tier → count → % over total feature tests, plus a one-line shape read. Healthy suites cover every P0 primary path with at least one Necessary test; a suite thin on Necessary but heavy on Nice-to-have is a smell (robustness without proven core), and the opposite (no defensive tests) is fragile. D is descriptive — it never changes Final.


Elicitation E — how well the framework extracts requirements (reported beside; never folded into Final)

E grades the SDD-derived spec.md/tasks.md against the PRD: did the framework surface the implicit requirements a senior engineer would insist on, and did it do so cleanly (no hallucinated or contradictory additions)? This is the "value beyond transcription" of an SDD framework. It evaluates the spec artifacts, not the code. Like R/S/D, it is reported beside Final so the fidelity grade stays comparable.

Three binary-checklist sub-metrics:

E_recall — did it find what it should have?

Run the frozen implicit-requirement category rubric below. For each category, mark Addressed (cite spec.md:line), Missed, or N/A (the PRD/domain makes it irrelevant). The rubric is PRD-agnostic and reusable, so recall is reproducible without per-PRD gold labeling.

E_recall = Addressed / (Addressed + Missed)        # N/A excluded
# Category A requirement is expected if the feature…
1 Input validation & bounds accepts user/external input (ranges, formats, required fields)
2 Error taxonomy & messaging can fail in ways the caller must distinguish (typed errors, codes)
3 AuthN / AuthZ exposes an endpoint or touches user-owned data
4 Idempotency & dedup can be retried or replayed (payments, create-once resources)
5 Concurrency & race conditions has interleaving writes / check-then-act on shared state
6 Data lifecycle & consistency persists state (transactions, partial-failure rollback, retention)
7 Observability runs in production (logs, metrics, traceability of failures)
8 Limits, pagination & rate returns lists or is callable at volume
9 External-dependency failure calls a third party (timeouts, retries, circuit-breaking, fallback)
10 State-transition integrity has a lifecycle/status machine (illegal-transition guards)

Extend the rubric per domain, but freeze it per benchmark and apply the identical list to every framework. For head-to-head benchmarks, also report pooled recall: union the valid additions across all frameworks into one adjudicated master list, then score each framework's recall against that pool.

E_precision — is the signal clean, or gold-plating?

Build the added-requirement ledger: every requirement in spec.md/tasks.md not traceable to a PRD line. Adjudicate each, binary, with a one-line warrant + spec.md:line:

Verdict Meaning
Valid-necessary a real gap the PRD implied (e.g. "reject trialDays > 30")
Valid-defensive reasonable hardening (observability, retries, typed errors)
Invalid hallucinated, contradicts the PRD, or scope-creep into the explicit out-of-scope list
E_precision = (Valid-necessary + Valid-defensive) / total additions

The valid additions become the valid E-additions set referenced everywhere else (the harness denominator, and S traceability). Mark each valid addition built or deferred (deferred-valid = good discipline, see S).

E_justified — are additions warranted?

E_justified = additions carrying an explicit rationale/traceability / total additions

Unjustified additions are a smell even when individually plausible. Report all three: E_recall / E_precision / E_justified. A strong framework reads high on all three — found the implicit requirements, added little noise, justified what it added.


Calibration (do this once per PRD, before trusting a benchmark)

The checklist is a hypothesis about what each AC means; calibrate it so different evaluators agree:

  1. Anchor. references/reference.md holds worked checklist anchors — at least one clearly-MET, one clearly-UNMET, and one borderline check with the verdict and reasoning. Read them before scoring; they fix where the MET/UNMET boundary sits.
  2. Agreement check. For a high-stakes benchmark, have the checklist reviewed against a human-labeled reference (or a second judge model). If verdicts disagree on more than ~20% of checks, the checklist wording is too vague — sharpen the ambiguous checks in _ac-baseline.md and re-run. Do not "average away" disagreement; fix the operational definition.
  3. Borderline cases are the valuable ones. When you find a check humans/judges split on, add it to the anchor pool in references/reference.md with the resolved verdict.

Process

Copy this checklist and track progress:

FRAMEWORK — respect & extract requirements
- [ ] 1. Locate PRD, spec.md/tasks.md. Run git diff to identify changed files (diff surface). Use the diff surface as the primary search scope for implementation and test evidence throughout steps 4 and 8
- [ ] 2. Enumerate stories + ACs; tag priority; mark out-of-scope items
- [ ] 3. Build/load the frozen binary checklist per AC (I-checks + T-checks) → _ac-baseline.md
- [ ] 4. Score I-checks MET/UNMET with file:line evidence (record searches before any UNMET)
- [ ] 5. Elicitation E: run the category rubric (E_recall) + added-requirement ledger (E_precision, E_justified) → freeze the `valid E-additions` set
- [ ] 6. Scope S: trace every built behavior (PRD-boundary / rogue-build / plan-drift)

HARNESS — ensure they are all implemented
- [ ] 7. Define the sanctioned set = PRD ACs ∪ valid E-additions; ensure each has a T-check
- [ ] 8. Score T-checks MET/UNMET with file:line evidence (unit + e2e per fixed level policy)
- [ ] 9. Inventory extra tests → Robustness Index R
- [ ] 10. Classify every feature test into tiers → Test Distribution D (counts + %)
- [ ] 11. Run/record Engineering Gates G (✓/✗/not-run)

ROLL-UP
- [ ] 12. Repeat the scoring steps three times (k=3); take majority verdict per check
- [ ] 13. Compute I, T, AC/Story/Final; apply bands; Adjusted Final only if a gate is ✗
- [ ] 14. Emit the report (see references/reference.md template): Final + E + S + R/G/D + ranked gaps + fixes

Step rules:

  • Step 1 — Identify the diff surface before any evidence search. Pick the base that matches what you are evaluating (branch tip, last commit, or working tree):
git diff <base>..<head> --name-only   # for a branch/PR
git diff --name-only HEAD~1           # for the last commit
git diff --name-only HEAD             # for uncommitted changes
git status --short                    # for untracked new files

Record the file list in the report. Steps 4 and 8 use it as the first set of paths to inspect.

  • Step 2 — Out-of-scope items from the PRD's explicit out-of-scope / non-goals section (whatever heading the PRD uses) get w=0. Absence is not a defect. Do not score them.
  • Step 3 — Reuse the existing _ac-baseline.md if present; otherwise create it and treat it as the contract for all future runs of this PRD. Include only observable-behavior clauses as checks.
  • Step 5E grades spec.md/tasks.md, not code. The ledger's valid additions define the valid E-additions set used by Steps 6 and 7. Mark each valid addition built/deferred.
  • Step 6 — Traceability: a built behavior with no PRD AC and no valid E-addition is a rogue build (fail). A deferred-valid out-of-scope addition is not a penalty.
  • Step 7 — The harness denominator is the sanctioned set (PRD ACs ∪ valid E-additions). A valid extracted requirement with no test is a T miss, not an E miss.
  • Step 8 — Binary only: MET (evidence) or UNMET (searched, absent). No partial per-check values.
  • Step 9 — Extra tests are a positive signal but never substitute for missing coverage. A high R with low T is still a low grade.
  • Step 10 — Each test lands in exactly one tier; counts must sum to the total feature-test count. Exclude pre-existing tests not added by the feature (note them separately).
  • Step 11 — A gate is / only if actually executed; otherwise not-run with the attempt evidence (command + error output — see Probe-before-not-run). Do not assume infra is unavailable: check it (e.g. docker ps, or whatever the project's own docs say the e2e suite needs running). If a gate is red, do NOT fix the subject code (Core rule 5).
  • Step 12 — Majority over k=3. Note any check where the three passes disagreed.
  • Step 13 — Show the arithmetic: list MET/total per check set so I and T are recomputable. Execute the roll-up as a script and paste its output (Reproducibility rule 4) — never hand-compute Σw or Final.

Output

Produce a markdown report following the template in reference.md. It must separate the two subjects and contain: per-AC implementation checklist table (I-checks, MET/UNMET, evidence); the Elicitation E block (category-rubric recall table + added-requirement ledger with verdicts → E_recall / E_precision / E_justified); the Scope S traceability verdict; per-requirement unit/e2e test checklist table (T-checks over the sanctioned set = PRD ACs ∪ valid E-additions); extra-tests inventory; the test-distribution D table (tiers with counts + %); the computed Final (+ Adjusted Final if a gate is ✗); R / G / D; a ranked gap list; and concrete fixes to reach 1.00.

Save the report to <spec-folder>/evaluations/<priority>-<story-slug>-<timestamp>.md when a .specs/spec folder exists; otherwise present it inline. The <timestamp> is a UTC instant in YYYYMMDDTHHMMSSZ form (e.g. 20260606T143012Z), obtained from the system clock at write time (date -u +%Y%m%dT%H%M%SZ). The timestamp makes every report filename unique so multiple agents evaluating the same story in parallel never overwrite each other — each run produces its own file. If a same-second collision is still possible (many parallel runs), append a short run id: <priority>-<story-slug>-<timestamp>-<run-id>.md.

Note: only the report filename is timestamped. The frozen _ac-baseline.md is deliberately not timestamped — it is a single shared artifact that every run reads from to stay comparable; never fork it per run.

Worked example

A complete applied evaluation (an example billing service's P0 "Start Free Trial Without a Card") showing the binary checklist per AC with file:line evidence, plus calibration anchors, is in reference.md. Mirror its structure and rigor.

Anti-patterns

  • Scoring I/T on a sliding "feels like a 0.75" judgment instead of counting binary checks.
  • Re-deriving the AC list or checklist per implementation (breaks comparability) instead of reusing the frozen baseline.
  • Marking a check UNMET without showing the search performed; or MET without file:line evidence.
  • Searching the full codebase for evidence without first running git diff to bound the diff surface — this produces unreliable searches and inflates "not found" confidence.
  • Awarding implementation credit without reading the production code path end-to-end.
  • Counting extra/defensive tests toward an AC's required coverage, or letting more code/tests raise the grade.
  • Reporting an Engineering Gate as passing without running it; or applying Adjusted Final for a not-run gate.
  • Marking a gate not-run without a recorded attempt (command + error output). "Infra not available" without a probe is fabricated evidence.
  • Fixing the code under evaluation to turn a red gate green — the evaluator is read-only over the subject; a red gate is scored and the fix goes in the report.
  • Hand-computing the roll-up (Σw, Final) instead of executing it as a script and pasting the output.
  • Penalizing missing out-of-scope features.
  • Folding E or S into Final — the headline grade must mean only "fidelity to the given PRD" or comparability breaks.
  • Penalizing a valid requirement the framework extracted but correctly deferred as out-of-scope (that is good discipline, not a scope failure).
  • Scoring harness completeness (T) against the PRD alone — the denominator is the sanctioned set (PRD ACs ∪ valid E-additions).
  • Rewarding requirement volume in E: hallucinated/scope-creep additions lower E_precision, they don't raise the score.
  • Changing the weights (0.6/0.4, P0/P1/P2) between implementations of the same PRD.
信息
Category 编程开发
Name spec-driven-eval
版本 v20260619
大小 25.34KB
更新时间 2026-06-21
语言