Patch Report · HN readiness surgery Date: 2026-04-25
Scope: self-improving-loop only. No changes were made to zhuge-skill, taiji, or taijios-bundle.
Item
Before
After
Evidence
Overhead badge
Badge said generic <1% while README table says ~10 ms calls can be +3.0%.
Badge now says LLM overhead <1%, and README explicitly says to measure sub-10ms calls before wrapping.
README Performance section still preserves measured fixed-cost caveat.
Test count
Show HN draft had stale counts after adding CLI, alias, and Yijing entrypoint tests.
Draft says 40/40.
python3 -m pytest --collect-only -q reports 40 tests collected; python3 -m pytest -q reports 40 passed.
Ising hook
README wording implied this package itself survived the Ising experiment.
Wording now says it is extracted from TaijiOS, which includes the Ising experiment and production-scale workloads.
README top section changed.
Hard-coded day count
Show HN draft and README had date deltas that age badly.
Copy now anchors on Chinese New Year 2026-02-17 without hard-coded 60 days deltas.
README Background and SHOW_HN_DRAFT body changed.
Item
Before
After
Evidence
Author metadata
authors = [{ name = "TaijiOS" }].TaijiOS maintainers plus GitHub noreply-style email.pyproject.toml.
Development status
4 - Beta.3 - Alpha.pyproject.toml.
License metadata
license = { file = "LICENSE" } plus old license classifier.license = "MIT" and removed superseded classifier.Editable install initially failed until classifier was removed; now build passes.
Package discovery
Hard-coded package list.
tool.setuptools.packages.find with include = ["self_improving_loop*"].Build includes self_improving_loop.yijing and CLI.
CLI
No console script.
self-improving-loop --version.Clean venv and wheel install both print the installed package version.
Keywords/topics
PyPI keywords and GitHub topics could drift.
PyPI keywords expanded to include ai-agents, rollback, autonomous-agents.
pyproject.toml; existing GitHub topics already include the same core terms.
Extra high-ROI fixes
Item
Change
CI badge
Added dynamic GitHub Actions badge to README.
PyPI badge
Replaced hard-coded 0.1.0 badge with dynamic PyPI version badge.
Framework-war prevention
Moved Not a... section to the second screen, immediately after install.
Threshold clarity
Added exec_count_24h explanation and manual override pointer.
Short API alias
Added loop.track(...) as an alias for execute_with_improvement(...).
HN operating checklist
Added first-24h comment-response checklist.
Changelog
Added CHANGELOG.md with 0.1.0 and 0.1.1 entries.
Source package materials
Added MANIFEST.in so sdist includes CHANGELOG.md and repo examples.
Check
Result
Unit tests
40 passed in 2.55s
Test collection
40 tests collected in 0.20s
Compile
python3 -m compileall -q self_improving_loop examples passed
Editable install
python3 -m pip install -e ".[dev]" passed
CLI
self-improving-loop --version prints the installed package version
Clean venv install
pip install -e . passed; pip freeze showed only self-improving-loop
Wheel install
Built wheel installed in a clean venv and imported SelfImprovingLoop + YijingEvolutionStrategy
Build
python -m build produced both sdist and wheel
Twine
twine check dist/* passed
Follow-up · local release verification
Check
Result
Isolated build tooling
Created a temporary venv and installed build + twine, because the host python3 did not include build.
Build artifacts
self_improving_loop-0.1.1-py3-none-any.whl and self_improving_loop-0.1.1.tar.gz generated under /tmp/self_improving_loop_release_verify_011_20260425_192922/dist.
Metadata validation
twine check passed for both wheel and sdist.
Clean wheel install
Installed the wheel into /tmp/self_improving_loop_install_verify_011_20260425_192949/venv.
Installed CLI
self-improving-loop --version printed self-improving-loop 0.1.1.
Installed import/API
SelfImprovingLoop(strategy=YijingEvolutionStrategy()) imported and executed successfully; installed object stores the hook as improvement_strategy.
TestPyPI credentials
TWINE_USERNAME, TWINE_PASSWORD, TWINE_REPOSITORY_URL, TEST_PYPI_API_TOKEN, and PYPI_TOKEN were all missing, so remote TestPyPI upload remains blocked.
Follow-up · PyPI version drift
Check
Result
Public PyPI
self-improving-loop 0.1.0 already exists on PyPI.
Public 0.1.0 install
Core import, YijingEvolutionStrategy, and SQLiteTraceStore import successfully, but the self-improving-loop console script is missing.
Public 0.1.1 availability
self-improving-loop 0.1.1 currently returns 404 on PyPI, so the version is available for the next publish.
Release implication
Current release-ready state cannot reuse 0.1.0, because PyPI versions are immutable.
Version action
Bumped local package metadata and __version__ to 0.1.1; 0.1.1 is the next publishable package version.
Item
Status
TestPyPI upload
Blocked: no PYPI / TWINE credentials are present in the environment, and no token was used from chat history.
Demo GIF/asciinema
Done as local demo artifacts: SVG terminal card, asciinema-compatible .cast, and transcript under assets/demo/.
LoopResult dataclassNot done. The current dict contract remains stable; changing result type before HN would be higher API risk than adding the safe track() alias.
Follow-up · Yijing positioning/API alignment
Item
Before
After
Evidence
Public positioning
Generic rollback-first reliability loop.
Hexagram-guided reliability loop with the rollback guard still central.
README top line and pyproject.toml description updated.
Constructor API
Only improvement_strategy= was documented as the strategy hook.
Added preferred strategy= alias while keeping improvement_strategy= compatible.
tests/test_smoke.py::test_strategy_alias_is_preferred_entrypoint.
Duplicate strategy args
Passing both names was ambiguous.
Raises ValueError.
tests/test_smoke.py::test_strategy_alias_rejects_duplicate_strategy_args.
Yijing example
Used legacy improvement_strategy=.
Uses strategy=YijingEvolutionStrategy(...).
examples/04_yijing_strategy.py.
Follow-up · bilingual launch material
Item
Before
After
Evidence
Chinese launch copy
No dedicated Chinese or bilingual material.
Added LAUNCH_COPY_BILINGUAL.md with English and Chinese positioning, demo script, short posts, HN boundary, safe claims, and claims to avoid.
README links to LAUNCH_COPY_BILINGUAL.md.
Overclaim guard
Launch wording could drift between English and Chinese channels.
Both languages explicitly say the Yijing layer is experimental, deterministic, and not full 64-hexagram coverage yet.
LAUNCH_COPY_BILINGUAL.md sections 6 and 7.
Follow-up · persistence evidence
Item
Before
After
Evidence
Test framing
tests/test_smoke.py still described the suite as minimal and partially unported.Updated the header to match current safety-path coverage.
tests/test_smoke.py.
Restart persistence
SQLite trace and loop state persistence were implied but not explicitly asserted in restart-style tests.
Added assertions that SQLite traces and improvement backup state survive a new SelfImprovingLoop instance.
tests/test_smoke.py; test count remains 40.
Follow-up · demo recording
Item
Before
After
Evidence
README visual proof
README checklist asked for screenshot/asciinema, but no recording artifact existed.
Added a terminal SVG demo plus transcript and asciinema-compatible cast.
assets/demo/self_improving_loop_demo.svg, .txt, .cast; README embeds the SVG.
Local path leakage
Raw command output included temp directories.
Demo artifacts use sanitized /tmp/sil_demo_* paths.
Sensitive-path grep is clean.