gtkb-benchmarks

Run read-only GT-KB benchmarks and emit JSON and markdown summaries.

2|1|Updated Mar 30, 2026

Search Tags:#recall #benchmarking #governance #assertion quality #evidence quality #run comparison #gt cli

Ready to run

npx skills add https://github.com/Remaker-Digital/groundtruth-kb --skill gtkb-benchmarks

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: gtkb-benchmarks
Source: https://github.com/Remaker-Digital/groundtruth-kb/tree/main/.claude/skills/gtkb-benchmarks
Command: npx skills add https://github.com/Remaker-Digital/groundtruth-kb --skill gtkb-benchmarks

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

GT-KB needs repeatable, read-only measurement benchmarks to quantify how well governance and evidence practices are working over time.

Core Features & Use Cases

Run GT-KB read-only benchmarks to compute structured observations (headline scalars plus per-dimension breakdowns) and produce both JSON and human-readable markdown summaries.
Assess assertion and evidence quality such as linkage survival across artifacts, recall coverage in change reasoning, attribution presence, and semantic recall of deliberations.
Compare benchmark runs by diffing prior results to understand changes in idempotency and benchmark values under consistent window and commit inputs.

Quick Start

Run read-only benchmarks for the default one-year window by executing: python -m scripts.benchmarks.cli run --all

Frequently Asked Questions about gtkb-benchmarks

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I benchmark governance and evidence quality in a repository?▼

To benchmark governance and evidence quality, run read-only measurement benchmarks that evaluate artifact-linkage survival, recall coverage in change reasoning, and attribution presence across time-windowed repository mutations.

What is evidence quality benchmarking for artifact linkage and recall?▼

Evidence quality benchmarking is a read-only measurement process that computes structured observations across multiple dimensions, generating deterministic JSON and markdown summaries to quantify how well governance practices are working over time.

How do I compare benchmark runs to track evidence quality changes over time?▼

Compare benchmark runs by diffing prior results using run idempotency keys, which allows you to understand changes in benchmark values under consistent window and commit inputs via the compare subcommand.

Can I run evidence quality benchmarks on a specific time window of repository mutations?▼

Yes, you can run benchmarks on time-windowed repository mutations. The default configuration evaluates a one-year window by executing the run subcommand, generating structured observations and human-readable summaries.

What output formats are generated when assessing assertion and recall coverage?▼

Assessing assertion and recall coverage generates deterministic JSON and human-readable markdown summary outputs, which are emitted under a designated runs directory to ensure repeatable measurement.

Do I need write access to run governance quality benchmarks?▼

No, governance quality benchmarks are read-only measurements that evaluate evidence without modifying the repository, requiring only the ability to execute the benchmarking CLI and emit output files.