Goldset
← goldset.ai
REDACTED SAMPLE
Pilot deliverable · QA pack

Gold scorecard & inter-annotator agreement report

This is the exact QA pack that ships with every Goldset batch, shown on a representative code-review eval pilot. Client identity, rubric text and batch ID are redacted; the data below is illustrative.

Batch summary
Batch
GS-PILOT-[REDACTED]
Task
Code-review evals — verdicts on model-written Python
Items delivered
250 (dual-labelled: 500 labels)
Gold seeded
25 items (10%)
Pool
5 certified annotators + founder review
Turnaround
8 business days
Acceptance vs. agreed thresholds
✓ PASS

Thresholds agreed before kickoff: aggregate gold accuracy ≥ 90% and Cohen's κ ≥ 0.70. Delivered: 95.0% gold accuracy and κ = 0.80. Redo guarantee not triggered.

Gold scorecard — per annotator
AnnotatorLabelsGold checksGold accuracyStatus
A-0110019 / 2095%CLEAR
A-0210020 / 20100%CLEAR
A-0310018 / 2090%RECALIBRATED D4
A-0410019 / 2095%CLEAR
A-0510019 / 2095%CLEAR

Aggregate: 95 / 100 gold checks correct (95.0%) against the 90% acceptance threshold. Gold items are seeded blind — annotators cannot distinguish them from production items.

Agreement report
Raw inter-annotator agreement (221 / 250 items)88.4%
Cohen's κ, pairwise average (4 verdict classes)0.80
Items adjudicated by reviewer29 (11.6%)

Most common disagreement: Correct vs Correct, minor issues (17 of 29) — an expected boundary; adjudications applied rubric §3.1 consistently and are marked in the dataset. One drift alert (day 4, A-03, minor/incorrect boundary) → 30-minute recalibration against the rubric; A-03's subsequent gold checks: 9/9.

Verdict distribution (250 items)
Correct96 · 38.4%
Correct, minor issues62 · 24.8%
Incorrect58 · 23.2%
Unsafe / vulnerable34 · 13.6%
Reviewer notes (founder sign-off)

R-01 · The rubric's minor issues boundary needed one clarification (over-broad exception handling, §3.1) — pressure-tested with the client on day 2, applied consistently from day 3. Recommend folding the clarified wording into rubric v1.4 before any production batch.

R-01 · Security verdicts (34 items) were the strongest dimension: every Unsafe call names the concrete attack vector in its rationale, per the rubric's evidence requirement.

R-01 · All 29 adjudications are flagged in labelled_data.json with the deciding rationale, so downstream filtering by agreement level is possible.

About this sample: the batch above is illustrative — the numbers, annotator pseudonyms and reviewer notes are representative, fictitious data prepared so buyers can inspect the deliverable format before engaging. No client data appears anywhere in this document. In a real engagement you receive this pack generated from your batch: labelled_data.json (every label, rationale, adjudication and gold flag) · agreement_report (the metrics above) · gold_scorecard (per-annotator accountability). NDA-first; production runs in Goldset's access-controlled workspace.