SME workbench

Your week

One expert's week on the bench: the judges earning and holding trust, the sign-offs that gate promotions, and the disagreements that become permanent calibration assets.

sme: padma.pWeek of 2026-07-06

dueShared label pool

9 items assigned to you · batch 68% labeled · human κ 0.74

sme: padma.p (you)

Judge work

4 items

duecomp-answer-judge@v1 — calibration queue

5 shadow and live verdicts to double-label — the judge is earning gate-eligibility

sme: padma.p (you)

5 verdicts · ~12 min

this week

duepolicy-answer-judge@v4 — weekly drift sample

6 live gate verdicts double-labeled to keep the certificate honest (§36)

sme: hermione.g

6 verdicts · ~9 min

Mon 2026-07-13

duerefusal-judge@v2 — weekly drift sample

3 live gate verdicts double-labeled — the §36 drift cadence

sme: remus.l

3 verdicts · ~5 min

Mon 2026-07-13

duetask-completion-judge@v3 — weekly drift sample

4 live dev-loop verdicts double-labeled — the §36 drift cadence

sme: arthur.w

4 verdicts · ~6 min

Mon 2026-07-13

Sign-offs & renewals

2 items

dueTrace promotion tr_c14f0 — awaiting SME sign-off

“Why is my overtime lower than I expected this period?” — scrubbed, overlay proposed (employees.split_shift_overtime_tx); your sign-off gates the PR

sme: padma.p (you)

this week

upcomingrefusal-judge@v2 — certificate renewal

certificate expires 2026-10-02; renewal window opens 2026-09 (§37: nothing is trusted forever)

sme: remus.l

opens 2026-09

The bench

24 SMEs across Payroll · WFM · HR compliance — median 14 yrs in domain

Payroll8WFM9HR compliance7

~32 min this week, across the whole bench

That is the entire weekly cost of keeping every gate-trusted judge honest.

Label pool — live

PN
Priya N.
Pay
27
today
MT
Marcus T.
HCM
19
today
SF
Sofia F.
Frontline
15
today
DC
Derek C.
HR Policy
12
today
AN
Aisha N.
Pay
21
today
JI
Jordan I.
HCM
9
today

7 online now · 1 more not shown · 4 product teams, routed blind to items they haven't seen.

Cross-team batch · 300 items204/300 · 68%

204 of 300 items have 2 or more labels

Knowledge captured

Expertise this deep used to live in people's heads. Every judgment call here becomes a permanent, versioned asset.

118 golden double-labels seeded policy-answer's calibration set
61 disagreements promoted into permanent calibration examples across policy-answer v1→v4
Rubric R4 (citation tightening) exists because SMEs kept failing paraphrased citations — shipped in v2, 2025-11

Understand the numbersWhat κ measures