DESIGN = PowerAnalysis.from_file("golden/design_v3.json")
Q = Qrels.from_file("golden/qrels_v3.trec")
gate = ReleaseGate()

# Preregistered design must be powered for the actual evaluation set.
gate.require("design", DESIGN.resolves_joint(n=len(Q.to_dict())))
gate.checkpoint()

# Latency gates first because a p99 regression makes the quality run pointless.
LAT = benchmark_pair(
    "prod_v17", "hybrid_bge_rerank",
    warmup=500, queries=10_000, paired=True,
)
p99_ok = (
    LAT.p99_ratio_ci95.high <= DESIGN.lat_ratio_max
    or LAT.p99_delta_ci95.high <= DESIGN.lat_abs_ms
)
gate.require("p99", p99_ok)
gate.checkpoint()

BASE = Run.from_file("runs/prod_v17.trec")
CAND = Run.from_file("runs/hybrid_bge_rerank.trec")

# All integrity checks report before the gate aborts, so one failure cannot mask another.
gate.require("provenance", manifest_matches(LAT.artifacts, "runs/"))
gate.require("paired_qids", same_qids(Q, BASE, CAND))
gate.require(
    "pool",
    validate_pool(Q, BASE, CAND, depth=50, max_unjudged_gap=DESIGN.pool_gap),
)
gate.checkpoint()

metrics = list(DESIGN.margins)
for run in (BASE, CAND):
    evaluate(Q, run, metrics, return_mean=False)

# Qrels define the paired set. A missing qid must fail loudly, not shrink an intersection.
qids = sorted(Q.to_dict())

# Every primary metric must clear its own margin (intersection union test).
# Alpha therefore needs no multiplicity adjustment.
for metric, margin in DESIGN.margins.items():
    cand_scores = [CAND.scores[metric][qid] for qid in qids]
    base_scores = [BASE.scores[metric][qid] for qid in qids]
    delta = paired_bca(cand_scores, base_scores, resamples=DESIGN.resamples, seed=DESIGN.seed)
    gate.require(f"primary:{metric}", delta.ci95.low > margin)

# Slices are guardrails with their own preregistered margins. Full NI proofs
# would be underpowered at slice size. Holm controls the familywise alarm rate.
gate_slices(
    gate, BASE, CAND,
    margins=DESIGN.slice_margins,
    correction="holm",
    min_n=DESIGN.slice_min_n,
)

raise SystemExit(gate.report())