Prospects list — black headline numbers, 60-day default window
Two dictated one-line changes to the prospects list: the tile headlines stop being graded by colour, and the list opens on 60 days. The colour half can only be shown, never asserted from text — so the before/after pair is the evidence for it.
30 Aug 2026 · branch fix/pr-6616-evidence-prose · captured against propflow-stage · at cd8ac5c430+local
Data note. Every name and balance below comes from propflow-stage, which is anonymized by construction (ADR-0097/0110 — phones in the impossible +1000 NPA, emails @example.test). The identical capture against prod would carry real tenants' names and balances and is not safe to attach anywhere, which is why the generator refuses to run against one.
1 · BEFORE — 8a771250, the commit this PR landed on: two tiles graded green, a six-month window
This is 8a771250 — main immediately before this PR squashed in — captured from its own dev server, so the comparison below is two renders of the same page rather than a render and a description.
It names a COMMIT and not a branch on purpose. This PR is merged, so main now carries the change and a capture pointed at it would quietly show the after-state under a “before” heading — the exact failure a before/after artifact exists to avoid.
Two things to fix on it. Applications and Signed Leases print their headline in the success green while Inquiries and Tours print plain — a verdict rendered as colour, on two of four tiles that all measure the same funnel, with no threshold behind it (1 application was as green as 40). And the toolbar chip on the right reads Last 6 months, which is the window the list opens on.
Asserted in the captured DOM — Inquiries · Tours · Applications · Signed Leases · Last 6 months. All four tile labels are present, and the date chip names the OLD default — so the next section's chip is a real change and not a re-crop.
2 · AFTER — every headline is the plain text token, and the window opens at 60 days
The same page with the change in, captured from main at cd8ac5c430. Compare the four headline numbers against the section above — that is the whole of item 1.
Two other things move between the two shots, named here so neither is mistaken for a side effect of the colour change: the tile VALUES drop (a 60-day window holds fewer leads than a 180-day one), and the Signed Leases sub-line re-words itself from last 6 months to last 60 days. Both belong to item 2. What is genuinely unchanged is the labels, the layout, and each tile's brand-spectrum identity dot — the dot stays because it is constant per tile, so it can never say “good” or “bad” about a value, and it is what pairs a tile with its funnel bar (see the judgment call in the PR body).
“Black” here is var(--color-text), the standard body/heading token — near-black on light, near-white on dark. A literal #000 is what black means in a mockup, not in a themed app.
The chip now reads Last 60 days. Every windowed caption is derived from the same ladder row as the predicate, so a moved default relabels the chip, the tile sub-line and the funnel panel together rather than leaving a stale claim over fresh data.
Asserted in the captured DOM — Inquiries · Tours · Applications · Signed Leases · Last 60 days. The same four tiles, and the chip naming the NEW default. The colour claim itself is not assertable from text — it is the thing to look at.
3 · The window is a default, not a cage — the ladder still offers every step
Gera asked for 60 days as what the page opens on. The rows above and below it are unchanged and one click away, which is what makes narrowing the default safe rather than lossy: Last 30 days / Last 60 days / Last 3 months / Last 6 months / Last 12 months / All time. The 6-month step the page used to open on is still right there.
This is also the reason 60 could become the default at all. It was rejected on 2026-08-26 because a stored window matching no quick range gives the chip nothing truthful to print; the row landed on 2026-08-28, and the objection stopped describing this value.
Asserted in the captured DOM — Last 30 days · Last 60 days · Last 3 months · Last 6 months · Last 12 months · All time. Every ladder row is rendered in the open popover — the 60-day default did not remove a single option.
4 · A wider window still renders plain numbers — the green was the rule, not the values
The retired rule was count > 0 ? success : undefined, so a capture whose Applications and Signed Leases counts happened to be zero would look identical before and after. This section widens the window to 12 months to raise those counts and shows them still plain — the same check the unit fixture makes, taken against real stage data instead of six hand-written rows.
It doubles as the URL half of item 2: an explicit ?dateWindow= still wins over the new default, so a bookmarked wider view opens wide.
Asserted in the captured DOM — Applications · Signed Leases · Last 12 months. A non-default window, named by the chip, with the graded pair on screen.
What these captures are load-bearing for
Item 1 is a claim about colour and nothing else. There is no behaviour to assert, so a unit test can only pin the token the component was handed; the before/after pair above is the part that shows a person what actually changed on screen.
Item 2 is the other way round — its behaviour is testable and is tested — so these captures add something different: they show it holding on real portfolio data rather than on six hand-written fixture rows.
What these captures do NOT show: the loading shell. The WINDOW_CHIP_GHOST skeleton is sized for the loaded chip — 129px against a measured 128.5px at 1440×900 — but no section here photographs the skeleton itself, so the ghost-vs-loaded layout question is not closed by this artifact. The shell is a Suspense fallback and a capture cannot reliably hold a warm dev server on it. What IS guarded is the PAIRING: measuredFor is asserted against the default preset's own label, and the shell is asserted to consume the shared constant rather than a literal, so the two halves cannot ship apart. The pixel value itself is a measured claim with a recorded instrument, not a guarded one — the WINDOW_CHIP_GHOST docblock in _helpers/tableDefaults.ts states that same limit, and is the record to trust on it.
Captured from propflow-stage on both sides — anonymized by construction (ADR-0097/0110), never production.