fix(platform): five high-severity bugs on /repos and repo-detail - #156
Merged
Conversation
- getRepoTimeSeries / getRepoAITimeSeries fetched the OLDEST N runs
instead of the newest N: `.order("created_at", { ascending: true
}).limit(52)` sorts ascending then takes the first 52, i.e. the
oldest 52 once a repo/window has more than 52 analysis runs. The
repo-detail page's top MetricCards, ChangeAlert, and every chart fed
by these two queries would freeze on a stale run and never advance,
while the rest of the page (sourced from getRepoLatestPayload, which
already did DESC + limit(1) correctly) kept showing the true latest
state — two contradictory pictures of the same repo on one screen.
Fixed by fetching DESC + limit then reversing in JS, mirroring the
pattern getOrgReposSummary's sparkline already uses correctly.
- Investment Hotspots' fix-magnet cards rendered "0% of changes, 1% of
fixes" instead of "35%, 70%". The engine's code_share_pct/
fix_share_pct (iris/analysis/fix_targeting.py) are 0-1 fractions
despite the "_pct" name; the platform copied them through unscaled
while its two sibling hotspot types (stabilizationRatio,
couplingRate) correctly multiply by 100. Scaled at the query layer
and documented the misleading engine field name on both the raw
FixTargetMetrics type and the platform's own FixMagnetHotspot type.
- /repos and /ai-exposure called getOrgReposSummary (and, for
ai-exposure, getOrgLatestPayloads) without resolving the org's
actual available analysis window first — unlike /dashboard and
/compare, which both call getAvailableWindowDays + resolveWindowDays.
Both pages silently defaulted to the hardcoded 90-day window; any org
ingesting under a different window saw every repo as unanalyzed on
these two pages while the other two showed real data. Added the same
window resolution plus a WindowSelector, for parity with the rest of
the app.
- The repo-scoped DORA card never got the MTTR P90 hint that was added
to the org dashboard's DORAOverview and to RepoDORA's own type — a
wiring gap, not a missing capability. Added it, reusing the same
translation key.
- The Weekly Activity table's "AI%" column divided by HUMAN+AI_ASSISTED
+BOT, while the canonical ai_detection_coverage_pct metric (used by
the "AI Adoption" chart on the same page) explicitly excludes BOT
from its denominator. A week with a dependency-bump burst would show
two different, both-labeled-"AI%" numbers on the same screen that
don't reconcile. Excluded BOT from the table's denominator to match.
Added a regression test for the fix-magnet scaling bug (confirmed it
fails against the pre-fix code: 0.35 instead of 35). The other four
fixes aren't covered by an automated test — three are UI/page-level
code this codebase doesn't unit-test anywhere, and the time-series
ordering fix is a direct Supabase query with no existing DB-mocking
convention to build on (platform/CLAUDE.md: "integration tests use
real DBs, unit tests for analysis modules"). Validated via tsc + careful
review against the already-correct sibling patterns in the same files.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This was referenced Aug 13, 2026
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.
Follow-up from a fresh audit of
/repos(list + detail pages) — five confirmed 🔴 bugs, each traced end-to-end before fixing.Fixes
getRepoTimeSeries/getRepoAITimeSeriesfetched the oldest N runs, not the newest N..order("created_at", { ascending: true }).limit(52)sorts ascending then takes the first 52 — the oldest 52 once a repo/window has more than 52 runs. The repo-detail page's top MetricCards,ChangeAlert, and every chart fed by these two queries would freeze on a stale run and never advance, while the rest of the page (viagetRepoLatestPayload, alreadyDESC + limit(1)) kept showing the true latest state — two contradictory pictures of the same repo on one screen. Fixed by fetchingDESC+ limit then reversing in JS, mirroring the patterngetOrgReposSummary's sparkline already uses correctly.Investment Hotspots' fix-magnet cards rendered "0% of changes, 1% of fixes" instead of "35%, 70%". The engine's
code_share_pct/fix_share_pct(iris/analysis/fix_targeting.py) are 0-1 fractions despite the "_pct" name; the platform copied them through unscaled while its two sibling hotspot types (stabilizationRatio,couplingRate) correctly multiply by 100. Scaled at the query layer and documented the misleading engine field name on both the raw type and the platform's own type./reposand/ai-exposureignored the org's actual analysis window, unlike/dashboardand/compare(which both callgetAvailableWindowDays+resolveWindowDays). Both silently defaulted to the hardcoded 90-day window — any org ingesting under a different window saw every repo as unanalyzed on these two pages while the other two showed real data. Added the same window resolution plus aWindowSelector, for parity with the rest of the app.Repo-scoped DORA card never got the MTTR P90 hint added to the org dashboard and to
RepoDORA's own type — a wiring gap, not a missing capability. Wired it in, reusing the existing translation key.Weekly Activity table's "AI%" column divided by HUMAN+AI_ASSISTED+BOT, while the canonical
ai_detection_coverage_pct(used by the "AI Adoption" chart on the same page) excludes BOT. A week with a dependency-bump burst would show two different, both-labeled-"AI%" numbers on the same screen that don't reconcile. Excluded BOT from the table's denominator to match.Verification
0.35instead of35).platform/CLAUDE.md: "integration tests use real DBs, unit tests for analysis modules"). Validated viatsc+ review against the already-correct sibling patterns in the same files.npx tsc --noEmit: clean.npm run test: 254/254 passing (253 + 1 new).npm run lint: 0 errors, same 72 pre-existing warnings.Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com