Inspect what was actually delivered.

LeanCTX is the Context Gateway for AI systems. It selects context, controls what reaches the model, learns how your work connects and proves what was delivered. The delivery records on this page belong to the Prove step.

Ask three separate questions

Question	Evidence
What did LeanCTX return?	The tool result, source references and available operation metadata
How much input changed?	Before/after counts with a declared tokenizer and measurement scope
Did the task improve?	A task-specific quality check under comparable conditions

A smaller result answers only the second question.

Inspect local usage

lean-ctx gain --json
lean-ctx spend --json

gain reports observed or estimated local context usage. spend reports provider usage where the integration captures it. Missing provider measurements must not be replaced with a claim about an actual bill.

Open lean-ctx cockpit for local activity, context and project views. The Cockpit guide explains setup, each area and empty-state diagnosis. Its coverage depends on observed operations; it does not see every action outside LeanCTX.

Choose the right report

Report	Purpose
lean-ctx gain --by-tool --json	Local context usage grouped by tool
lean-ctx token-report --json	Project/session context and memory signals
lean-ctx output-savings --json	Output-token estimates; not quality proof
lean-ctx value-report --format json	Recorded task outcomes and associated cost
lean-ctx cep	Context-processing and cache metrics
lean-ctx savings	Representation-change ledger; verification does not establish task quality

gain --svg and gain --share write local share artifacts. Inspect their contents before publishing them yourself. Hosted publication and public rankings remain Research. For advanced comparison, shadow, verification and evidence commands, use the evaluation directory; a signed artifact authenticates its recorded contents, not the success of an undeclared quality test.

Follow evidence back to a source

Keep source identity, representation and available receipt references together. When an output is archived, inspect the archive before treating a summary as complete evidence.

Not every tool or integration emits the same receipt fields. Verify the actual artifact contract rather than relying on a mock dashboard.

Evaluate a change fairly

Use the same task, source revision, model configuration and success criterion for a comparison. Include retries, recovered context and failed attempts. Separate estimated token reduction from measured billing.

For a reproducible research example, see the Wavect case study. For local measurement semantics, use token measurements.

Sources & versions3 references · Reviewed 17 Sept 2026

• pipeline.rsrust/src/server/call_tool/pipeline.rsReviewed checkout
• keyed.rsrust/src/core/context_admission/receipt_store/keyed.rsReviewed checkout
• 11-analytics-and-insights.mddocs/reference/11-analytics-and-insights.mdReviewed checkout

Core checkout

0ce2207ee4

Reviewed 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