nl2sql · baseline

← demo

Pick baseline

Production reference v21_cr · thinking-off batch · promoted 2026-06-11 PM

STABLE FLOOR 38/40 baseline_phase3_ev01_ds_v21_cr (pin 33c6e0b, 2026-06-11 PM) is the production EV-01 reference and Phase 8's Step-1 no-regression comparator. Promoted over v20_cr as the paired read for the thinking-off batch (thinking:{type:'disabled'} on gap_detect + decompose + pass2_decompose + diagnostic narrative): zero flips on the touched surface — all 4 verdict flips are documented fast-path rollers / refusal-class churn.
class:standard (honest metric)
95.0%
38 / 40 row-set Equal — honest floor 37–39/40 depending on the avtwo_040 / avtwo_043 rollers
headline row-set Equal
66.7%
40 / 60 — gap is class design, not wrong answers
latency (the batch's win)
p50 11.9s
−28% vs v20 · p95 68.7s (−26%) · max halved 182.9→92.0s · orchestrator path p50 32.4→27.1s
compile_route_v1
22 / 60
tied v20 · grounding 0.0% · gen_failed 0 · cost $0.521 (−2.3%)

The 2 standard misses (= the documented floor composition):
avtwo_057 — governance-fenced by its own LOCK (wrong-DAU-source value bug, deliberately not relitigated); routes fast-path since v20 (extractor label coining) — signature changed, same miss.
avtwo_040 — known LLM-variable compile-route outer-WHERE-clamp defect (fix open by choice). avtwo_043 (FTD-formula SQL-gen roller, the second documented roller) rolled Equal this cut — hence floor 37–39/40 depending on the 040/043 rolls.
Why the 66.7% headline is not a correctness ceiling: the 20 non-equal rows = 9 diagnostic (narrative answers — row-set Equal is the wrong lens; tracked by Track-W fan-out 78%) + 5 refusal_expected (proxy references — refusal is the correct behavior) + 4 blocked_infra (references fail by construction — need DB-owner DDL / grants) + the 2 above. None are wrong answers from the pipeline.
Arms: v16+ cuts are compile-route-only (_cr = production-representative; no ablation arm). Picker is latest-only per phase (2026-06-12): the named priors v20_cr (its memo documents the v19 dual-period catch + the 043 roller) and v18_cr (V1 closing reference), plus v17_cr / v14_cr, all live on disk under the direct-URL retention contract. Memo: eval/results/baseline_phase3_ev01_ds_v21.md · ⚠ pipeline_sha does not hash src/orchestrator (carried — live :3010 probes are the only valid verification for orchestrator-stage changes).