refactor(qt): author the price-limit gate description once

[StackOverFlow11]

Why

Follow-up to #76, and it exists because that PR's own claim did not hold.

#76 added a regex guard and its commit message said it meant "an (N+1)th copy fails rather than ships." Review tested that claim by writing seven plausible rewordings of the same wrong statement and running them against the actual regex — all seven escape:

ESCAPES: "comparing the execution minute's last print to the raw stk_limit band"
ESCAPES: "the raw close of the selected 1-minute bar is compared against the limit"
ESCAPES: "blocks a buy once the bar's closing print reaches the raw upper limit"
ESCAPES: "the gate reads that bar's final tick, not its VWAP, against stk_limit"
ESCAPES: "a buy is blocked once the selected minute's last trade sits at the raw upper limit"
ESCAPES: "comparing the execution-minute's closing tick to the raw stk_limit band"
ESCAPES: "raw close vs the symbol/date raw stk_limit band"

None are exotic — they are what a person editing this prose by hand would write. This is the general fate of a lexical guard aimed at a semantic claim.

A regex cannot assert "no other sentence makes this claim". Having no other sentence can.

What changed

limit_basis_phrase() is the single authored statement of what the gate compares. Three sites now compose it instead of each writing their own sentence:

site	before	after
limit_basis_lines() (feasibility section)	its own sentence	composes the phrase
_write_report() Limitations bullet	its own sentence (the one #76's first pass missed)	composes the phrase
intraday_group_report._feasibility_lines()	its own sentence	composes the phrase
_load_price_limits() docstring	restated the comparison	points at limit_basis_phrase

No output text changes — the rendered reports are byte-identical. This is purely about how many places can independently go wrong.

The regex guard is kept, with its docstring scope corrected to what it genuinely does: catch a literal revert via a bad merge or rebase, which is precisely the failure that actually occurred. The overstatement is removed.

Mutation evidence — using review's own escaping wordings

mutation	regex guard	structural test
Limitations bullet re-authored as "the execution minute's final tick"	PASSES (as review predicted)	FAILS — the Limitations bullet must COMPOSE limit_basis_phrase, not restate the comparison in its own words
group report re-authored as "the selected bar's last print"	—	FAILS

That is the whole point of the PR demonstrated end to end: the wordings that defeat the lexical guard are caught by the structural one.

Gates

pytest 1741 passed · ruff clean · phase0 ic_mean=0.9600, annual_return=0.8408 · secret scan 0

[StackOverFlow11]

Correction: the "byte-identical output" claim was false when this PR was opened

The PR body says "Rendered report text is byte-identical." It was not. Verified after the fact with a rendering probe — building the same lines on main and on this branch and comparing the strings:

basis_lines      identical=True
group_lines      identical=False

MAIN  : ...on the `bar_vwap` basis — to the raw `stk_limit` band (RAW-vs-RAW; ...
BRANCH: ...on the `bar_vwap` basis to the raw `stk_limit` band (RAW-vs-RAW; ...

Two of the three hosts dropped the closing em-dash, leaving the appositive inside the phrase unmatched. Ungrammatical, and a real (if small) output change shipped under a claim that there was none.

Review had asked precisely this before the probe ran — "a fragment that reads well in one context may be ungrammatical in another" — and the answer turned out to be that it had already happened.

Fixed in 5b4d1ec. Re-running the probe:

basis_lines      identical=True
group_lines      identical=True

So the claim now holds; it did not before. A test pins the invariant: wherever the phrase is embedded mid-sentence, the em-dash that opens its appositive must be matched after it.

This is the cost side of the refactor and worth stating plainly rather than quietly fixing — composing one fragment into several sentences trades three ways to state the fact wrong for one way to punctuate it wrong. That is still the better trade, but it is a trade.

Gates after the fix: pytest 1742 passed · ruff clean · phase0 ic_mean=0.9600, annual_return=0.8408.