evaluation-anchor-checker

Solid

Audit and rewrite evaluation/numeric claims to ensure they carry minimal protocol context (task + metric + constraint) and avoid underspecified model naming. **Trigger**: evaluation anchor checker, numeric claim hygiene, underspecified numbers, protocol context, 评测锚点检查, 数字断言, 指标上下文.

AI & Automation 498 stars 39 forks Updated today

Install

View on GitHub

Quality Score: 81/100

Stars 20%
90
Recency 20%
100
Frontmatter 20%
70
Documentation 15%
100
Issue Health 10%
50
License 10%
0
Description 5%
100

Skill Content

# Evaluation Anchor Checker (make numbers reviewer-safe) Purpose: fix a reviewer-magnet failure mode in agent surveys: - strong numeric/performance statements appear - but the minimal evaluation context is missing This skill treats numeric claims as *contracts*: - if a number stays, the same sentence must contain enough protocol context to interpret it - if that context is not in evidence, the claim must be downgraded (no guessing) ## Inputs Preferred (pre-merge, keeps anchoring intact): - the affected `sections/*.md` files Optional context (read-only; helps you avoid guessing): - `outline/writer_context_packs.jsonl` (look for `evaluation_anchor_minimal`, `evaluation_protocol`, `anchor_facts`) - `outline/evidence_drafts.jsonl` / `outline/anchor_sheet.jsonl` - `citations/ref.bib` ## Outputs - Updated `sections/*.md` (or `output/DRAFT.md` if you are post-merge), with safer evaluation anchoring - `output/EVAL_ANCHOR_REPORT.md` (always; short report with files checked / changed / weakened sentences) - Optional completion marker: `output/eval_anchors_checked.refined.ok` ## Recommended slot in the survey pipeline Use this as the **last section-level numeric hygiene sweep before merge**: - after `style-harmonizer`, `opener-variator`, `section-logic-polisher`, and `paragraph-curator` - immediately before the final `argument-selfloop` snapshot and merge Reason: - earlier section-level rewrite passes can legitimately rephrase or fuse numeric sentences - if you only wait for...

Details

Author
WILLOSCAR
Repository
WILLOSCAR/research-units-pipeline-skills
Created
7 months ago
Last Updated
today
Language
Python
License
None

Similar Skills

Semantically similar based on skill content — not just same category

AI & Automation Listed

double-check

Exhaustively validate every claim, number, calculation, code-fact, and command in a document or outbound communication. Extracts all claims into a prioritized ledger (flagging what is NOT covered), validates each by the right method (run the math, run the command, cite the file:line, quote the primary source), then calibrates certainty with CIA Words of Estimative Probability and a provenance tier — and proposes corrected, properly-hedged rewrites for any overclaim. Use before sending a message, filing an issue, committing analysis, or whenever you state exact numbers/assertions others will rely on.

0 Updated 3 weeks ago
DazzleML
AI & Automation Listed

ai-eval-review

Audit the evaluation layer of an AI product against design-completeness questions: offline criteria, ground-truth quality, online signal, cohort/disparate-impact, adversarial + robustness coverage, and drift detection. Eval-side companion to `ai-ux-review`. Produces an editable Markdown artifact plus a self-contained HTML report under `docs/ai-ux/`. Use when the user asks to "review my AI eval setup", "audit my eval design", "is my AI eval rigorous enough", "responsible-AI eval review", "fairness eval check", or "drift detection design", or ships an LLM/ML feature and wants an eval-rigor check. Invoke even if only one block is named — the others stress-test it. Does NOT trigger for: human-AI UX review (`ai-ux-review`); lean-canvas work (`validation-canvas`); adversarial pre-mortem with a verdict (`startup-grill`); SKILL.md audits (`skill-evaluator`); implementing eval pipelines, writing eval code, or labeling datasets (this skill names gaps, it does not build them).

1 Updated 6 days ago
sorawit-w
AI & Automation Featured

eval-design-forensics

Audit whether a paper's EVALUATION DESIGN actually measures what it claims and whether its reporting is complete — the validity layer family D (experiment-forensics) cannot reach. Three patterns: train/test leakage means the reported score may not measure generalization (HP-EVAL-LEAKAGE — adopts the Kapoor & Narayanan 8-type / 3-category leakage taxonomy; the illegitimate-proxy / sampling-bias / pretraining-contamination subtypes hand off as needs_external_check, naming but NEVER running Oren-2023 exchangeability / Shi-2023 Min-K% / Golchin-2023 Time-Travel / BIG-bench canary); a load-bearing LLM judge is conflicted (same model/family as a compared system) or unvalidated (no human-agreement, no bias control) (HP-JUDGE-VALIDITY); a declared condition/metric is dropped or switched to favor the method, or 'best' is chosen with no held-out set (HP-SELECTIVE-REPORTING). Verdict-bearing at L0/L1 from the DESCRIBED protocol — NOT repo-gated like experiment-forensics; L2 only CONFIRMS against split/preprocessing/resu

139 Updated 3 days ago
wanshuiyin