# CSL-Meter

> CSL-Meter is a public observatory that measures whether personal AI
> subscription plans and pay-as-you-go API plans actually serve the model
> they claim. It probes plans with real upstream API calls and publishes
> a CSL index per plan, with every reading labeled by origin.

## What CSL means

CSL = Consistency, Substitution & Latency. For each plan CSL-Meter runs
a fixed probe battery (real outbound API requests), extracts a numeric
distribution fingerprint from the responses, and compares it against a
golden baseline for the claimed model. The published CSL index combines
identification consistency across harnesses, substitution risk (the
fingerprint drifting towards a different model's profile) and latency.

## Data labeling rules (read this before citing any number)

- Simulated: simulated baseline data, seeded for display. It is NEVER
  mixed into verified aggregates.
- Server-Verified: produced by a real probe battery executed and stored
  by the server against the provider's official endpoint.
- A plan only counts as verified once it carries at least one
  Server-Verified sample; the site-wide status turns verified-live only
  once EVERY listed plan has one.
- If a number has no label, it is not a measurement.

## Key links

- Live site: https://konspace.cc/
- Plans API (JSON): https://konspace.cc/api/v1/plans
- Methodology: https://github.com/konspace-cc/csl-meter/blob/main/docs/methodology.md
- Source code (Apache-2.0): https://github.com/konspace-cc/csl-meter

## Calibration

The golden baselines are still being crowd-calibrated: if you hold a
first-party API key for a baseline model, you can contribute a
calibration capture — see CONTRIBUTING.md in the GitHub repository.