Skip to content

Tags

Tags

v0.7.1

Toggle v0.7.1's commit message
### Development

- *(ci)* Move the self pins to v0.7.0

v0.7.0

Toggle v0.7.0's commit message
### Features

- *(package)* Let the sequential test take its error rates
- *(action)* Add an action at the root that pools games already on the runner

### Bug Fixes

- *(action)* Resolve any branch in a private repository

### Documentation

- *(docs)* Add a quickstart and the example files it copies
- *(package)* Say what mache is in the heading and link the quickstart from PyPI
- *(docs)* Publish the documentation as a GitHub Pages site

### Development

- *(ci)* Let strength.yml take its sequential test in normalized elo
- *(deps)* Bump git-cliff

v0.6.0

Toggle v0.6.0's commit message
### Features

- [**breaking**] *(action)* Ship a default book table
- *(package)* Put the difference in normalized elo in the estimate
- *(package)* Let the sequential test take its bounds in normalized elo
- *(package)* Let the rating estimate name the list its ladder came from

### Bug Fixes

- *(package)* Read the movetext from the blank line that ends the tags
- *(package)* Say when the node rate is not a speed comparison
- *(package)* Read a one-sided match rather than refuse its fit
- *(package)* Act on the statistical review of the normalized test

### Documentation

- *(docs)* Lead with what mache is for and what it can measure
- *(docs)* Correct and tighten the new README sections

### Development

- *(ci)* Guard each ladder stage on the batches asked for and on the verdict
- *(ci)* Move the self pins to v0.5.0
- *(ci)* Hold the self pins to the version being released

v0.5.0

Toggle v0.5.0's commit message
### Features

- *(package)* Put the clocks and the node counts in the report

### Development

- *(ci)* Move the self pins to v0.4.0 and give back the commit pin

v0.4.0

Toggle v0.4.0's commit message
### Features

- *(package)* Reserve the whole sequential test in the book, not one batch
- *(action)* Hand back what the sequential test decided

### Development

- *(ci)* Move the reusable workflows' self pins to v0.3.0
- *(ci)* Stop a sequential test as soon as a batch settles it
- *(ci)* Pin the batch workflow's actions at a commit, and test the pins

v0.3.0

Toggle v0.3.0's commit message
### Development

- *(ci)* Add reusable workflows for a whole match in one call
- *(ci)* Play a real match in this repository's own ci
- *(ci)* Stop the release trying to push a workflow file

v0.2.0

Toggle v0.2.0's commit message
### Features

- *(package)* Name the version in the report of each estimate
- *(action)* Move the match orchestration into six composite actions

### Bug Fixes

- *(package)* Say what an sprt verdict means
- *(package)* Read a pgn as utf-8 and not as the locale

### Documentation

- *(docs)* Finish the tools table and show the command lines
- *(docs)* Say that the sequential test keeps no state

### Development

- *(ci)* Pin what the workflows install from PyPI
- *(ci)* Take the persisted credential off every checkout
- *(ci)* Release from the actions tab

v0.1.0

Toggle v0.1.0's commit message
The first release.

The tools come from the chess engine they were written for, where they had
grown into something no longer specific to it.

Four command line tools read a sharded match: a pooled elo estimate over every
shard rather than the one a single process saw, a pentanomial sequential test
judged at batch boundaries so that shards can play at once without spending the
error rates, a count of how the games ended, and the opening slices that keep
two shards off the same position. A composite action builds fastchess at a
pinned commit, fetches an opening book and checks it against a recorded hash,
and puts the package on the path.

The json output is format 1 and is a contract from this release.