---
title: Baselines
description: Save a JSON report on the main branch and fail the build when any score gets worse.
---

For an existing codebase, save a JSON report on the main branch:

```sh
poly-crap --format json --output poly-crap.json
```

Then compare a change against it:

```sh
poly-crap --baseline poly-crap.json --fail-regression --fail-above
```

Each current function is matched to the baseline by file, name, and position,
with allowances for a baseline written from a different root. Every function
gets one status:

- `regressed`: the score rose by more than `--epsilon`, default `0.01`.
- `improved`: the score fell by more than `--epsilon`.
- `unchanged`: neither. The table hides these rows; JSON keeps them.
- `new`: nothing in the baseline matched.
- `moved`: the same function, found in another file.
- `removed`: a baseline function nothing matched.

```text
Changes since baseline
  Status     Score  Delta  Language  Symbol  Location    Uncovered
  regressed    6.0  +5.00  python    run     ./app.py:4  6-8
  new          1.0    N/A  python    helper  ./app.py:9
1 regression found.
1 of 3 functions exceeds CRAP threshold 5.0.
```

:::tip
`--fail-regression` fails the run on any `regressed` function. It says nothing
about a `new` one, because there is no earlier score to rise from, so pair it
with `--fail-above` to hold new functions to the threshold.
:::

A baseline written by an older poly-crap may not parse; regenerate it.

`--allow` applies to the baseline too, so an allowed function never appears as
`removed`.