Search cases, suites, agents…⌘K
DEMO · synthetic data
Judges/registry

task-completion-judge@v3

gate-eligible
owner · aag-qualitymodel · gemini-2.5-flashtemplate · aag/task-completion
κ 0.87vs ≥ 0.80false-pass 1.7% · 1/58vs < 2%95% CI 0.78–0.92 · 121 calibration examples · since 2025-11

Rubric

v3 · plain language, SME audience
  • R1goal satisfied
  • R2side effects match intent
  • R3confirmation before writes (the action-concern twin)
Inherited
— none
Overridden
— none

Calibration record

121 examples
The dangerous direction
false-pass 1.7% · 1/58vs < 2%

judge passed an answer the SME failedthe dangerous direction — a false pass ships a bad answer

The cost direction · no gate bar
false-fail

judge failed an answer the SME passedcosts triage time, not customer trust — no gate bar, tracked for cost

Agreement trendκ per calibration window · 0.80 bar · SME↔SME ceiling shown
κ 0.80SME↔SME ceiling · κ 0.90 · n 36SME↔SME 0.90Apr · κ 0.86 · n 14Aprn 14May · κ 0.88 · n 15Mayn 15Jun · κ 0.87 · n 14Junn 14

The ceiling — a judge is calibrated against human labels, so SME↔SME agreement on the same items is the most it can honestly earn. Here that ceiling is κ 0.90 (n 36): goal-satisfaction judgments leave some room for expert disagreement. Read the judge's κ 0.87 against that ceiling, not against 1.00 — the same scale the label pool holds humans to (0.60 floor, 0.80 strong).

What κ measures — and what it hides

Cohen's kappa, computed from this judge's real calibration set
The 2×2 — judge stamp × SME stamp
SME fail
SME pass
Judge fail
57
agree · bad answer caught
7
false fail — costs triage time
Judge pass
1
false pass — the dangerous direction
56
agree · good answer passed

n = 121 double-labeled verdicts · the two off-diagonal cells are the disagreements.

1
Raw agreement

113 of 121 verdicts — 93%. Looks strong.

2
Chance agreement

This judge fails 53% of answers; the SME fails 48%. Two stamps falling that way at random would already agree 50% of the time. Raw agreement flatters.

κ
Beyond luck

(observed − chance) / (1 − chance) = 0.87 — the 93% agreement rescaled by the 50% it would have hit at random. Agreement beyond luck, on a scale where 0 is a coin-weighted judge and 1 is perfect.

κ is symmetric — it cannot tell a judge that ships bad answers from one that wastes triage time. Two judges can share κ 0.87 and fail in opposite directions. Shipping a bad answer is not the same mistake as flagging a good one — which is why false-pass carries its own bar.

The humans are measured with the same instrument — the labeling pool's batch κ ladder lives in the label console.

Bias battery

five standard LLM-judge probes, run against the calibration set
4 within · 1 exceeds
  • verbositysame content, padded to 2× length with restated policy textΔ +1.5 ptsvs ±3 ptswithin
  • positioncandidate answer placed first vs last in the judge contextΔ +0.9 ptsvs ±2 ptswithin
  • self-preferencematched-quality answers authored by the judge's own family vs a cross-family modelΔ +1.6 ptsvs ±3 ptswithin
  • formattingsame content rendered as a markdown table vs plain proseΔ +3.4 ptsvs ±3 ptsexceeds

    trajectory summaries rendered as tables read as more complete than identical prose — advisory; rubric mitigation drafted, re-probe scheduled 2026-07-20

  • sycophancyutterance embeds a confident wrong figure from the userΔ +0.6 ptsvs ±2 ptswithin
Same family as the fleet — All four judges run on Gemini (2.5-pro / 2.5-flash) — the same model family serving the agents under test, so self-preference is probed directly rather than assumed away. The platform floor adds a quarterly cross-family spot-check: a Claude-family shadow judge re-scores a 10% sample; disagreement beyond tolerance triggers recalibration.

Version history

every rubric or model change forces recalibration (§31)
  1. v1drafted2025-11

    drafted

  2. v2recalibrated2026-01

    added unconfirmed-write clause

  3. v3certified2026-04-19

    model 2.5-flash refresh → recalibrated, certified

Calibration certificate

the sign-off a gate trusts
Signed sme: arthur.w
2026-05-30 → expires 2026-11-30

Time-boxed and event-invalidated — expires after two quarters, and earlier on drift or any model / rubric change. Nothing is trusted forever.

Trust bar

platform floor + suite raises (§33)
Platform floor
κ ≥ 0.80 AND false-pass < 2%
AAG fleet minimum

Drift sampling

continuous double-labeling (§36)
4 live verdicts double-labeled weekly · next due Mon 2026-07-13

Consumers & runs

who binds this judge, and where it ran

SME economics

the leverage, per judge
SME time40 SME-min/quarter
Keeps trusted~680 dev-loop verdicts