> ## Documentation Index
> Fetch the complete documentation index at: https://cxity.kairaxis.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Metrics and Proof: Understand Contexity Value Estimates

> Contexity shows heuristic token-saved and time-saved estimates after relevant work. Learn what the numbers mean and when A/B benchmarks are required.

Contexity can display heuristic value metrics after relevant work completes. These estimates help you understand where context retrieval is saving time and tokens — but they come with clearly stated limits, and they are not a substitute for rigorous benchmarking when public claims are involved.

## Example Metrics Output

After a task closes, you may see a summary like this:

```text theme={null}
---
Contexity metrics, heuristic:
- Tokens saved: ~1,400 (93.4% of estimated 1,499-token total)
- Time saved: ~7m 12s (100% of estimated 7m 13s total)
- Contexity runtime: ~35ms observed
- Proof level: heuristic estimate; use paired A/B for measured proof.
```

## What the Metrics Mean

| Metric        | Meaning                                                                    |
| ------------- | -------------------------------------------------------------------------- |
| Tokens saved  | Estimated context or graph-discovery tokens avoided by using Contexity     |
| Saved percent | Saved amount compared with the estimated total token cost                  |
| Time saved    | Heuristic time avoided through context reuse or graph lookup               |
| Runtime       | Measured Contexity tool runtime for the completed task                     |
| Proof level   | Whether the number is a heuristic estimate or backed by a paired benchmark |

## Heuristic vs. Benchmark Proof

Heuristic metrics serve a specific and useful purpose. Use them for:

* Product feedback and iteration
* Internal diagnosis of retrieval quality
* Explaining why Contexity helped during a particular run
* Spotting patterns where retrieval consistently saves work

They are **not** sufficient for public marketing claims. If you need to make a verifiable, public statement about Contexity's impact, you need paired A/B benchmarks structured as follows:

* Same task, identical problem statement
* Same repo state at the start of both runs
* One Contexity-assisted run and one baseline run without Contexity
* Multiple valid trials to account for variance
* Clear exclusion rules for incomplete or invalid runs
* Confidence intervals where statistical rigor is required

Heuristic output is a useful internal signal. Benchmark output is evidence you can stand behind publicly.

## Controlling Metrics Display

You can disable visible metric summaries for a project while keeping structured metrics available for logging and tooling:

```bash theme={null}
contexity metrics disable --project /path/to/project
```

To re-enable visible metrics:

```bash theme={null}
contexity metrics enable --project /path/to/project
```

For full details on configuring and managing metrics output, see the [manage metrics guide](/guides/manage-metrics).