Repository navigation
chore: fix the tier glossary, and drop the py2mcp workaround - #14
Merged
Merged
Conversation
Two follow-ups from the surface work. The catalogue header still defined tiers as confidence that a model wrote it -- E as "near-certain artifact". That stopped being true when the per-rule `weight:` was split out: tier is now how firmly a rule is worth enforcing as writing advice (which `acquaint` reads), and weight is how much a hit is worth as evidence about authorship (which `ductus` reads). Measurement against 350 human-written documents falsified the old reading for several rules while leaving them worth fixing as advice, which is why the axes were split rather than the rules demoted. Comments only; both suites unchanged. That also settles the open question about `summary-closer`. It keeps tier E: the case for demoting it rested on the stale definition, and once tier means "worth fixing whoever is reading" there is nothing to demote. Its weight of 0.15 already carries the measured finding that it is near-worthless evidence. `_resolve_annotations` is gone. py2mcp 0.1.12 requires fastmcp>=3.2, where keyword-only defaults survive `from __future__ import annotations`, so the workaround has nothing left to work around. The unit test that pinned it goes with it; `test_the_optional_arguments_really_are_optional` stays and is the real guard -- it passes with the workaround removed, which is the evidence that removing it was safe.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two follow-ups. No behaviour change in either.
The tier glossary was stale
tells.yamlstill defined tiers as confidence that a model wrote it —E near-certain artifact of a model having written it. That stopped being true when the per-ruleweight:was split out in #5. The two axes now mean different things:tier— how firmly a rule is worth enforcing as writing advice.acquaintreads it and never readsweight.weight— how much a hit is worth as evidence about who wrote the text.ductusreads it;acquaintnever does.Measuring 350 human-written documents falsified the old reading for several rules while leaving them worth fixing as advice, which is exactly why the axes were split rather than the rules demoted. The header now says that. Comments only.
This settles the open
summary-closerquestionIt keeps tier E. I went in expecting to demote it — it trips 25 of 350 human documents and matched machine ground truth zero times, which makes "near-certain artifact" plainly false. But that argument rests on the stale definition. Once tier means "worth fixing whoever is reading", there is nothing to demote: as style advice, "In conclusion," is worth flagging even for a reader who tolerates AI-sounding prose. Its weight of
0.15already carries the measured finding.The contradiction was in the glossary, not in the rule.
The py2mcp workaround is gone
_resolve_annotationsexisted because fastmcp < 3.2 dropped every keyword-only default underfrom __future__ import annotations, which made every verb on the MCP surface uncallable. i2mint/py2mcp#14 pinsfastmcp>=3.2,<4, so there is nothing left to work around.The pin here moves to
py2mcp>=0.1.12. The unit test that pinned the workaround goes with it;test_the_optional_arguments_really_are_optionalstays, and it passing with the workaround removed is the evidence that removing it was safe rather than an assumption.309 tests pass, ruff and mypy clean.
acquaint's 555 also pass against the changed catalogue.