Selecting a different window or a comparison strategy silently made the tiles
stop matching the recommendation below, which is baked into the report and
cannot follow a dropdown. On load they now agree by construction; moving off
that basis says so.
Also: an absent production row produced no headline and no benchmark, but any
passing gate finding still rendered a green "no warnings" chip — a success badge
for missing data, directly beside "this report predates the portfolio monitor".
Missing baseline now reads "baseline unavailable".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Five UI problems, all reported from using the page.
Expanding "How this is measured" shoved every control down, because the
disclosure and the run controls shared one flex row. They no longer do: run
status and the controls that start a new run sit together on one line, and the
explainer is below them where growing it moves nothing.
A long strategy name wrapped the dropdown trigger onto three lines and dragged
the row out of alignment. The trigger now truncates with the full text on hover
— a wrapping dropdown is broken anywhere, so the fix is in the primitive — and
the twelve-character "Production: " prefix is a bullet.
"Sortino 2.72" answered nothing. Each risk-adjusted metric now carries a meter:
a track showing where the value sits, ticks at the band edges, and the band word.
Colour never travels alone. Bands are deliberately stricter than textbook ranges
because this universe is today's survivors replayed backward, which flatters
every ratio — that caveat is stated next to them rather than left implied.
The two tile rows were different sizes, which read as inconsistent rather than
as hierarchy. Every tile is the same size now and grouping carries the ranking:
top row is raw outcome and takes no meters, second row is risk-adjusted ratios
and all take meters. Sharpe moved down to join them — it is one of those ratios,
and leaving it above made it the only metered tile in a row of bare ones.
The recommendation led with a long bold sentence that describes the
configuration, not a verdict, while the actual findings were small grey text.
Findings now come first, each split into label and detail on the colon the
backend strings already carry, and the configuration is a footer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
One tab stacked three things that all called themselves a track record:
realized paper P&L, setup-outcome grading under the rejected take-profit model,
and the backtest portfolio simulation. Split into Setups | Paper Trades |
Backtest, one subject each. `track` stays the Paper Trades slug so the legacy
/performance redirect keeps working. The grading diagnostic and its Evaluate /
Reset controls go with Backtest, not Paper Trades — reset_track_record deletes
trade_setups, not paper trades.
BacktestPanel 439 -> 175 lines. Its run settings alone were 106 lines of
hand-rolled sr-only radio cards for two binary choices; they are now two
Dropdowns and a button on one wrapping row, with the per-option prose moved into
the existing explainer. The amber warnings survive as a conditional slot, so a
non-default choice still announces itself but the common path is silent.
The recommendation printed eight findings at equal weight, burying the verdict
in tuning detail. `topic` now splits them: production, benchmark and robustness
stay inline, gate/exit/cutoff collapse behind a disclosure, and any WARNING or
LAGS item is promoted out of the collapsed group regardless of topic. No topic
chips — every backend string already self-prefixes, so a chip would render
"GATE | Gate: ...".
Portfolio metrics are now two tiers: five headline tiles for what the book
returned, then a smaller labelled row for how good that return was (Sortino,
Calmar (MAR), Gain/Pain, Profit Factor $, EV/trade). Reports cached before those
metrics existed hide the second row rather than showing a half-populated line of
dashes.
Extracted EquityCurveChart, PortfolioMonitorPanel and BacktestRecommendationCard,
plus a StatTile primitive and shared formatters for the duplication in the files
this touched. DashboardPage and OpenTradesPanel deliberately keep their own
copies — migrating them is separate scope.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
dennisthiessenandClaude Opus 5