The hero card on / draws a seven-day axis over returns that took about a month to happen.
SAMPLE_CURVES (BoardPreview.tsx:56) runs "7d ago" → "5d ago" → … → "Now", and its own doc comment describes Live-board shape, "one session at a time".SAMPLE_STANDINGS beside it (:57-61) carries +21.0% / +5.5% / +2.8% / +1.4% / −1.5% — Competition-window figures over roughly a month. Race.tsx:9 imports the same array and heads it "Competition Standings", under copy reading "Competition: one fixed window of market history".- The curves were built to land exactly on those returns, so nothing on the card contradicts itself. It just says DeepSeek made +21.0% in seven days.
Pre-existing — #367 did not introduce it. Filing it now because #367 roughly triples the chart and makes that axis the most prominent claim on the acquisition page, which raises the cost of leaving it wrong.
#367 corrected the caption from "over the competition window" (which contradicted the axis) to "over the last seven days" (which matches it). That is as far as wording can go: the caption now truthfully describes a chart whose underlying data is borrowed.
Why this is polish, not a correctness bug
The card carries an Illustrative example chip, and the numbers are openly fabricated — no user or model is being misrepresented. The cost is credibility on the page whose job is credibility.
Options
- Re-derive
SAMPLE_CURVES at plausible seven-day magnitudes and give BoardPreview its own returns. Race.tsx keeps the Competition figures, which is where they belong. Costs a second array and a re-derivation of the chart-height reserves if the chip text changes width.
- Relabel the axis to the competition window (
"Apr 15" … "May 15"). Cheaper, but re-introduces the contradiction with SAMPLE_CURVES' documented Live-board shape.
Preference is (1) — the sample data is fabricated anyway, so fabricating it at the right magnitude costs nothing but the edit.
Note if touching this: the chip strip's width feeds two measured constants (--board-chart-reserve, 590 stacked / 390 at lg+) and the comment above them says to re-derive both in-browser if the strip's content changes. Don't reason them out.
The hero card on
/draws a seven-day axis over returns that took about a month to happen.SAMPLE_CURVES(BoardPreview.tsx:56) runs"7d ago" → "5d ago" → … → "Now", and its own doc comment describes Live-board shape, "one session at a time".SAMPLE_STANDINGSbeside it (:57-61) carries +21.0% / +5.5% / +2.8% / +1.4% / −1.5% — Competition-window figures over roughly a month.Race.tsx:9imports the same array and heads it "Competition Standings", under copy reading "Competition: one fixed window of market history".Pre-existing — #367 did not introduce it. Filing it now because #367 roughly triples the chart and makes that axis the most prominent claim on the acquisition page, which raises the cost of leaving it wrong.
#367 corrected the caption from "over the competition window" (which contradicted the axis) to "over the last seven days" (which matches it). That is as far as wording can go: the caption now truthfully describes a chart whose underlying data is borrowed.
Why this is polish, not a correctness bug
The card carries an
Illustrative examplechip, and the numbers are openly fabricated — no user or model is being misrepresented. The cost is credibility on the page whose job is credibility.Options
SAMPLE_CURVESat plausible seven-day magnitudes and giveBoardPreviewits own returns.Race.tsxkeeps the Competition figures, which is where they belong. Costs a second array and a re-derivation of the chart-height reserves if the chip text changes width."Apr 15" … "May 15"). Cheaper, but re-introduces the contradiction withSAMPLE_CURVES' documented Live-board shape.Preference is (1) — the sample data is fabricated anyway, so fabricating it at the right magnitude costs nothing but the edit.
Note if touching this: the chip strip's width feeds two measured constants (
--board-chart-reserve, 590 stacked / 390 atlg+) and the comment above them says to re-derive both in-browser if the strip's content changes. Don't reason them out.