web-og ResultCard ignores result.partial, so a truncated traversal unfurls as a final answer

[lukaso-bot]

Found while reviewing #158 (thread on packages/web-og/src/index.tsx:242). Pre-existing on main, not introduced by that PR, so it is split out rather than bundled into it.

What

ResultCard in packages/web-og/src/index.tsx never reads result.partial. The web page does — packages/web/src/ui/result-card.tsx renders PartialResult (:23) and BestEffortBanner (:31) — so the two surfaces disagree about the same result, and the one that disagrees is the one that gets screenshotted into Slack, X and GitHub unfurls.

Two shapes render wrong today:

1. partial with firstRelease: null. The card shows the green --ship dot, "First released in", and not yet released. A truncated traversal is not evidence the commit is unreleased. If the commit is released, the unfurl asserts the opposite of the truth.
2. partial WITH a firstRelease. The card shows the green SHIPPED badge with no caveat. The gallop hit is the release we happened to land on, and the bisect that would confirm no EARLIER release contains the commit is exactly what the soft deadline cut short — so v2.0.0 can unfurl as final when v1.9.0 was the true earliest.

That is the CLAUDE.md guardrail — "Partial state ≠ 'not yet released'" — broken on the share surface.

Why it is not urgent-but-still-real

#158 gives both shapes a 300s PENDING_CACHE instead of the 24h pin, so a mislabeled card now lives 5 minutes at a time rather than a day. Right direction, but the copy is still wrong for those 5 minutes, and unlike the web page there is no caveat anywhere on the PNG.

Interaction with the PRs in flight

• fix(web): align /internal/* result cache with the public routes (#143) #144 makes /internal 503 every partial, so once it merges no partial reaches web-og at all: the card becomes the neutral placeholder. That removes the wrong copy by removing the card.
• Deadline-heavy repos get a permanently blank OG card and one full traversal per minute after #144 #156 is the counterpart that reopens the route deliberately: render the gallop hit with a caveat and keep it revalidatable. This issue is what Deadline-heavy repos get a permanently blank OG card and one full traversal per minute after #144 #156's card has to say — it is the copy half of that change.

So: fix this with #156, not before it. Fixing it before #144 lands would be writing copy for a path #144 closes; fixing it after #156 lands without this would reopen the path with the same wrong copy.

Shape of the fix

Mirror what result-card.tsx already does, within the OG card's space budget:

• partial + firstRelease → keep the tag, replace the SHIPPED badge with a "best effort" / "not confirmed earliest" treatment.
• partial + null → do not say "not yet released"; say the lookup was cut short.

Guard it with a routing test per shape asserting the rendered copy, alongside the lifetime test added in #158 (a soft-deadline partial with NO release renders the not-yet copy, PENDING-cached), which will need updating in the same commit.

Related: #159 (the same partial pinned as terminal on the web + badge surfaces), #151, #156.

[lukaso-bot]

Pairs with #156 — both are the follow-on that closes #144's regression window.

#144's round-19 review confirmed the shape: after #144, /internal 503s on every partial and the OG card falls back to PlaceholderCard permanently for repos that reliably blow the soft deadline. #156 is the fix (render the gallop hit, short-cache by terminality); this issue is why #156 cannot simply return 200 today — ResultCard ignores result.partial, so a truncated traversal would unfurl as a final answer with no caveat, which is the exact failure #144's 503 exists to prevent.

So the dependency runs: #144 (safe-but-degraded) → #160 (card can express a partial) → #156 (stop dropping the gallop hit). #160 and #156 can reasonably land as one web-og PR.