Add a guaranteed line, list every place head to head, drop the leverage grid

[dgoodenough]

Three follow-ups to #75, all on the What's left tab. Front-end only — no pipeline change, no bundle-schema change.

1 · A guaranteed line on the cutline table

A fourth column, and the columns reordered hardest-first: Guaranteed · High · Median · Low, still after Short. That is the order a chaser reads them in, because the question is what it takes to stop worrying.

#	Player	Points	Short	Guaranteed 868	High 826	Median 798	Low 777
25	Marwede	794.42	4	24th	46th	any	clear
28	Tamm	730.1	68	7th	17th	25th	38th
36	Smith	661	137	2nd	4th	7th	12th
41	Welck	596.2	202	—	1st	2nd	3rd

Guaranteed is the cutline at its 99.9th percentile — clear it and you are in under all but one simulated season in a thousand, which is as close to a guarantee as 25,000 seasons can express.

It is a column and not a mark on the chart, deliberately: 868 lands past the last histogram bar (the bars stop at the 99.5th percentile), so a guide line for it would hang off the right edge with nothing under it. The panel note now says exactly that, so the number does not look like it arrived from nowhere.

The bands also gain keys and are looked up by key rather than by index. The chart reads its median from band(bands, "mid") now, so the solid line cannot silently follow the next time the display order changes — which is what would have happened here, since the reorder moved the median from index 1 to index 2.

The row filter moves with it: the table now lists everyone not already past the guaranteed line rather than the high one, since a player clear of high still has something to read in the new column.

2 · Head to head lists every place

The sampled ladder (1, 2, 3, 5, 8, 12, 20, …) is gone. It was defensible about points and wrong about people: a top-heavy curve does make the deep rows nearly identical in value, but the row someone actually wants is the one their rival finishes in, and nothing here can guess which that is.

81 rows in MPO, 42 in FPO — contiguous 1..N, verified. That is the roster (76 / 37) plus five places of slack, because entries move right up to the first tee and a ladder that stops at today's headcount would be missing real finishing positions by Sunday.

That one depth (fieldDepth) now feeds every search and every ladder in the file. It matters: if the ladder ran to 81 while needAt still searched to 76, rows 77–81 would price their answer against a shallower field and the margins would drift. Same place asked for, same place priced.

The table moves into pv-scroll pv-tall — the same 460px box with a sticky header that the other long tables on this tab already use.

3 · The leverage grid is hidden while one event is left

"Where the season actually gets decided" answers which of the events left matters most to you, which stops being a question when the answer is the only column there is. With one event on the calendar it had become a restatement of the auto-bid odds printed beside it.

Conditional on d.events.length > 1 rather than deleted — it is the most interesting panel on this tab in March, and gating it also skips a 24-row conditional re-simulation (~1.2s of chunked work) computing something nobody can act on. The panel count in the intro follows automatically: "These five panels".

Happy to make it unconditional if you would rather it were simply gone.

Verification

• Both divisions at 390px and 1180px, every tab walked and returned to: five panels, no leverage grid in the DOM, no page errors, no horizontal overflow.
• Needs table renders Guaranteed 868 / High 826 / Median 798 / Low 777 in MPO and 1100 / 1063 / 1034 / 1011 in FPO.
• Head-to-head ladder asserted contiguous 1..81 (MPO) and 1..42 (FPO); sticky header measured pinned to the scroll box at scrollTop 900 / 678, at both widths.
• python -m pytest tests -q → 258 passed (unchanged — this touches no Python).

🤖 Generated with Claude Code

https://claude.ai/code/session_01JsBC4pyUqbxW9y5o9tAeRr

---

Generated by Claude Code