Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 1 addition & 8 deletions .secrets.baseline

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

42 changes: 33 additions & 9 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,16 +7,28 @@

**Evolve code. Show the evidence.**

LibreEvolve is an open-source workbench for inspectable code-evolution
experiments. **Engineering preview: bounded Python bin-packing optimization
with Codex OAuth.** It starts from a small Python heuristic, searches a bounded
local space of changes, records candidate workspaces and evaluations, and lets
you inspect a saved run before a fresh export check.
LibreEvolve is an **open-source workbench for inspectable code-evolution
experiments**. **Engineering preview: bounded Python bin-packing optimization
with Codex OAuth.**

It starts from a deliberately simple Python bin-packing heuristic, searches a
bounded local space of changes, and saves every candidate workspace and
evaluation so you can read the evidence yourself — before a fresh export check
Comment on lines +15 to +16

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Qualify the guarantee that every candidate is saved

If a run is forcibly terminated or persistence fails after evaluation but before db.log()—the loop has separate evaluated and history_logging phases—the run may contain only partial state, as docs/run-artifacts.md explicitly acknowledges. Saying the tool saves “every” candidate workspace and evaluation therefore promises evidence that may not exist; qualify this as successfully persisted candidates or restore the previous non-universal wording.

AGENTS.md reference: AGENTS.md:L37-L39

Useful? React with 👍 / 👎.

re-runs the candidate you chose.

The product is an experiment you can examine, not a promise that every search
improves code. No improvement is a valid result. A successful named check is
not proof of universal correctness, optimality, security, or performance.

## What makes it worth inspecting

| | |
| --- | --- |
| **Bounded by design** | Local call, evaluation and runtime limits keep a run small enough to follow. They are not token or dollar caps, and subscription usage may stay unknown. |
| **Saved, not summarized** | Candidate workspaces, evaluations, archive events and LLM calls land in a run directory you can read. |
| **Honest by construction** | Run status, validity, quality, retention and usage stay separate, so a finished run is never mistaken for a win. |
| **Checked again before use** | Reports and exports re-run the selected candidate in fresh local subprocesses instead of trusting old metadata. |

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Distinguish report workspace from the selected export source

When best_workspace/ differs from the saved history candidate, alpha report always evaluates the on-disk workspace, whereas alpha export requires an explicit --source workspace or --source history choice. Saying both commands re-run “the selected candidate” hides this source divergence and can make the report sound like verification of the history-selected program; describe the report’s workspace source separately from export’s selected source.

AGENTS.md reference: AGENTS.md:L44-L45

Useful? React with 👍 / 👎.


## What is available today

- A source install for the public engineering-preview checkout.
Expand All @@ -33,10 +45,22 @@ An approved inspectable example viewer is not included in this branch. The
seed-only flow in the quickstart is an onboarding check, not an optimization
result or live-model demonstration.

The broader research platform, additional providers, plugins, examples, papers
and experimental tools live in the private
[experimental repository](https://github.com/LibreEvolve/libreevolve-experimental);
they are not capabilities of this public engineering preview.
Research features, additional providers, plugins, papers and experimental tools
are out of scope for this checkout; they are not capabilities of this public engineering preview.

## How a run moves

1. **Seed.** Initialize the bundled bin-packing task and inspect its
intentionally simple heuristic.
2. **Change.** Propose and evaluate bounded local changes — optionally through
the Codex OAuth lane, after you authorize your own subscription use.
3. **Inspect.** Read the saved run. Completion, validity, training quality,
retention and recorded usage are separate dimensions, so a finished run is
not the same as a retained win.
4. **Verify.** Freeze your selection, then let a fresh export check it on named
training and held-out cases and record its identity.

**Run the preview** through the [canonical quickstart](docs/alpha-quickstart.md).

## Start with the canonical quickstart

Expand Down
2 changes: 1 addition & 1 deletion tests/test_brand_docs.py
Original file line number Diff line number Diff line change
Expand Up @@ -168,7 +168,7 @@ def test_docs_keep_future_items_and_example_slot_unpublished():
readme = _read("README.md")
roadmap = _read("docs/roadmap.md")
assert "approved inspectable example viewer is not included" in readme
assert "private\n[experimental repository](https://github.com/LibreEvolve/libreevolve-experimental)" in readme
assert "libreevolve-experimental" not in readme
assert "not capabilities of this public engineering preview" in readme
assert "Planned or proposed work" in roadmap
assert "not current capabilities" in roadmap
Expand Down
Loading