Reference

Analysis tools

Inspect output, measurements and local evidence.

Status by featureReviewed September 2026

Choose the question first

Use representation comparisons to inspect what compression changes. Use usage reports to inspect recorded operations. Use verification tools for the specific integrity properties they check.

Do not turn a diagnostic score into a claim about general model quality.

Keep measurement scope explicit

Identify the source, delivered output, tokenizer and cache state. Include recovery and retries when comparing complete tasks.

Provider billing needs provider usage evidence. An estimate calculated from token counts and prices should remain labelled as an estimate.

Verify outcomes independently

A useful evaluation states the expected result before running the task. Check code behavior, required source coverage or another task-specific success criterion under comparable conditions.

The observability guide explains how to connect delivery and outcome without conflating them.

Find a tool

Analysis tools

Inspect output, measurements and local evidence.

ctx_benchmark3 parameters ctx_cost3 parameters ctx_feedback9 parameters ctx_gain4 parameters ctx_heatmap2 parameters ctx_metrics0 parameters ctx_perf0 parameters ctx_proof7 parameters ctx_quality_lab4 parameters ctx_radar1 parameters ctx_verify2 parameters ctx_discover1 parameters

Sources & versions2 references Reviewed 17 Sept 2026

11-analytics-and-insights.mddocs/reference/11-analytics-and-insights.mdReviewed checkout
16-signed-savings-ledger.mddocs/reference/16-signed-savings-ledger.mdGitHub

Core checkout: 0ce2207ee4
Installed runtime: 3.10.2
SDK release: 1.1.0

Separate baselines for source, CLI/configuration and SDK contracts. Review does not certify every platform or integration.

Versions & compatibility