diff --git a/README.md b/README.md index bfdf5325..13038eff 100644 --- a/README.md +++ b/README.md @@ -45,8 +45,8 @@ cases with known totals and a local program as the agent, then runs the real loop — only the model is replaced — and ends with: ``` -held-out reward: 1.000 outcomes: {'committed': 1} -what it learned: rules.md -> 'COLUMN: amount' +held-out reward: 0.000 -> 1.000 outcomes: {'committed': 1} +what it learned: rules.md -> 'COLUMN: amount' ``` Skill evolution here means a **multi-file skill library**, not a single prompt: diff --git a/changelog.d/1146-first-run-transcript.md b/changelog.d/1146-first-run-transcript.md new file mode 100644 index 00000000..6534df00 --- /dev/null +++ b/changelog.d/1146-first-run-transcript.md @@ -0,0 +1,5 @@ +Fixed +The first-screen demo transcripts (`README.md`, `docs/start.md`, +`docs/plugin-quickstart.md`, `docs/testing-guide.md`) now show the seed baseline +the CLI prints, `held-out reward: 0.000 -> 1.000`, instead of a bare `1.000`, and +a docs guard keeps those four blocks on the command's output (#1146). diff --git a/docs/plugin-quickstart.md b/docs/plugin-quickstart.md index 792b9729..ff678fb0 100644 --- a/docs/plugin-quickstart.md +++ b/docs/plugin-quickstart.md @@ -78,7 +78,7 @@ plan: 12 tasks, 4 rounds x 2 workers, up to 68 agent calls running (20260905-134121-1c695b)... round 0 reward=1.000 +1/-0 -held-out reward: 1.000 outcomes: {'committed': 1} +held-out reward: 0.000 -> 1.000 outcomes: {'committed': 1} what it learned: rules.md -> 'COLUMN: amount' ``` diff --git a/docs/start.md b/docs/start.md index 996b1fab..78cc7666 100644 --- a/docs/start.md +++ b/docs/start.md @@ -23,8 +23,8 @@ cases with known totals and a local program as the agent, then runs the real loop — only the model is replaced — and ends with: ``` -held-out reward: 1.000 outcomes: {'committed': 1} -what it learned: rules.md -> 'COLUMN: amount' +held-out reward: 0.000 -> 1.000 outcomes: {'committed': 1} +what it learned: rules.md -> 'COLUMN: amount' ``` Skill evolution means a **multi-file skill library**, not a single prompt: diff --git a/docs/testing-guide.md b/docs/testing-guide.md index a1c93572..0bba1cc2 100644 --- a/docs/testing-guide.md +++ b/docs/testing-guide.md @@ -77,7 +77,7 @@ plan: 12 tasks, 4 rounds x 2 workers, up to 68 agent calls running (20260905-134121-1c695b)... round 0 reward=1.000 +1/-0 -held-out reward: 1.000 outcomes: {'committed': 1} +held-out reward: 0.000 -> 1.000 outcomes: {'committed': 1} what it learned: rules.md -> 'COLUMN: amount' ``` diff --git a/tests/test_docs_examples.py b/tests/test_docs_examples.py index b7e81829..746f6d6e 100644 --- a/tests/test_docs_examples.py +++ b/tests/test_docs_examples.py @@ -75,3 +75,25 @@ def test_snapshot_docs_spec_examples_are_json(): for name in re.findall(r'load_spec\("([^"]+)"\)', text): assert name.endswith(".json"), ( f'docs/snapshot.md load_spec("{name}") is not a file load_spec can read') + + +FIRST_SCREEN_DEMO_PAGES = ( + "README.md", + "docs/start.md", + "docs/plugin-quickstart.md", + "docs/testing-guide.md", +) + + +def test_first_screen_demo_transcript_shows_the_seed_baseline(): + """#1146: the CLI prints the seed and the final reward on one line. + + `agentdescent demo` ends with `held-out reward: 0.000 -> 1.000`, and these + four pages are the transcript a new user compares a working install against; + a bare `1.000` makes the first screen disagree with the command it documents. + """ + for page in FIRST_SCREEN_DEMO_PAGES: + text = (ROOT / page).read_text() + assert "held-out reward: 0.000 -> 1.000" in text, page + assert "held-out reward: 1.000 outcomes" not in text, page + assert "held-out reward: 1.000 outcomes" not in text, page