Our results.

We are building Excellent with Excellent.

This page shows what changed, how we tested it, what failed, and what we still do not know.

Current release.

Release                         Not published
Date                            Not published
Source commit                   Not published
Build digest                    Not published
Agent version before            Not published
Agent version after             Not published
Task distribution version       Not published
Held-out set digest             Not published
Verification contract version   Not published
Receipt ID                      Not published

No signed public relaunch release result has been published yet. The page is wired to the public results manifest and will render the release fields when they exist.

Before and after.

                                Before        Current

Verified result rate             Not published Not published
Cost per verified task           Not published Not published
Human help per task              Not published Not published
p50 elapsed time                 Not published Not published
p95 elapsed time                 Not published Not published
Known outcome regressions        Not published 0

Unknown work stays in the denominator. The primary comparison will show an uncertainty interval when a signed result includes one.

What changed.

Model use       Not published
Instructions    Not published
Context         Not published
Tools           Not published
Checking        Not published
Retrying        Not published
Stopping        Not published

Each real change will link to the experiment that supported it. No change is listed as promoted until a receipt exists.

What was tested.

Tasks used for baseline          0
Tasks used for candidate work    Not published
Tasks used for validation        Not published
Held-out tasks                   0
Hidden from candidate proposer   Not published
Mechanical checks                Git, tests
Model or human review            Not published
Outcome observation window       0 days
Environments tested              macOS local
Skipped                          Release manifest, public receipt, confidence interval

Models and providers.

Excellent is designed to sit beside the coding agent and model you choose. It checks the output against repo evidence, so the comparison can survive a model change.

OpenAI: Supported
Anthropic: Supported
Google: Supported
xAI: Supported
Meta: Supported
Mistral: Supported
DeepSeek: Supported
Groq: Supported

Candidates rejected.

Candidate records              Not published
Rejected candidate count       0
Invalid candidate count        Not published

Rejected candidates are part of the result. This section should show meaningful failures, not only the successful path.

Known limits.

  • No signed public relaunch result has been published yet.

Verify it yourself.

excellent receipt verify release-Not published.json
  • Download receipt: Not published
  • View signed payload: Not published
  • View public key: Not published
  • Receipt format: Read the receipt format

Release history.

Version   Date   Verified result   Cost/result   Status       Receipt
current   -      Not published     Not published  In progress  Not published
Install ExcellentView current release