Skip to content

Ask an environment finding where its value would differ - #29

Merged
tauanbinato merged 5 commits into
weak-spotsfrom
weak-spots-2
Sep 26, 2026
Merged

tauanbinato merged 5 commits into
weak-spotsfrom
weak-spots-2

Conversation

@tauanbinato

@tauanbinato tauanbinato commented Sep 26, 2026 •

Copy link
Copy Markdown
Contributor

Stacked on #28. After #28, the weakest reviews were hardcoded values' "fixes a value that differs between deployments": about a third of the labeled ones were right.

What changes

A hardcoded-value review or consider that rests only on the environment question is asked where its value would differ. It's asked once the locate names the value or constant, with the function around the value or the lines that use the constant.

  • Kept: a value each installation must set, or the author's own account or domain.
  • Note, when those options together reach 0.80:
    • the same value in every copy on purpose (the program's own service or registered client id, a provider's fixed address, a path the platform fixes)
    • a fallback used only when configuration gives none
    • code no deployment runs (fixtures, generator templates, scripts)

The constants locate now lists each constant's use lines, as #28 added elsewhere lines for values.

Results

Labeled findings on the tuned and fresh projects, #28 → this branch:

Right Wrong
Environment findings 17 → 15 36 → 19
All reviews 519 → 518 (74% → 76%) 179 → 167
All considers 830 → 829 337 → 332
  • Now notes: a desktop app's own license server, a provider's registered redirect, fallbacks behind environment variables, release scripts' fixed /usr/bin paths, a mock's base URL, and a generator's install template.
  • Right ones lost: a frontend's API host that was edited in code three times, and a template author's domain used as a fallback.
  • Held-out projects: unchanged. No new review or consider findings, so nothing to label.
  • Cost: $0.004 for the new questions on the corpus.

Review notes

  • 0.80, not 0.50: at 0.50 the step would have lowered 25 wrong and 6 right findings, against 17 and 2 at 0.80.
  • Tried and dropped: the same kind of question for shared logic ("what would one shared implementation save?"). The model answered "a few calls, assignments or checks" for almost every pair, right or wrong (0.84 on average), so it doesn't separate anything. The patch and numbers are recorded outside the repository.
  • Self-check: no unbaselined review or consider. It flagged the new tests' repeated setup, now a shared helper.

A hardcoded-value review or consider that rests only on a value changing
between environments is asked, once its value or constant is named and
with the code that uses it, where the value would differ: each
installation or the author's own account keeps it; the same value in
every copy on purpose, a fallback used only when configuration gives
none, or code no deployment runs, at 0.80, make it a note. The Choice
that names a file's constant lists the lines that use each one. Labeled
such findings went from 17 right and 36 wrong to 15 right and 19 wrong.
A path check left undecided after its trace and recheck is asked, with
the functions that call it, where the paths come from: the program's own
directories or configuration, the command line or settings of the person
running a local program, a whole path its caller gives, or another
party's input. The program's own or the local user's clear it: 74 units
on the corpus, for about $0.02.
Parameters in a path, URL or redirect were a note once a check found
them, but stayed uncertain while the check was undecided. Once the
Choice that settles the check was asked and does not lean toward another
party's input, such a unit is a note naming the path, URL or redirect:
450 injection units had stayed uncertain that way.
A test whose hollow checks stay undecided once its recheck is asked is
clear when they lean below 0.50. Labeled from the code, 4 of 43 such
tests below 0.50 were hollow, against 10 of 35 above; undecided tests on
the corpus went from 792 to 156, with nothing asked again.
Undecided units on the corpus went from 1.39% to 0.83% of judged units
with these changes and the environment Choice before them; no review or
consider changed.
@tauanbinato
tauanbinato merged commit 538e361 into weak-spots Sep 26, 2026
9 checks passed
@tauanbinato
tauanbinato deleted the weak-spots-2 branch September 26, 2026 22:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant