Who to borrow

What a point is made of

Every skill a spark can hand over is valued on this course, for this style, as the lengths it is expected to gain at the line: what it gains wherever it can fire, times how often it fires there, with the races where it does not fire counted as zero. The fold skills ranked for this Cup and style lists every one, with how it was valued.

A skill that is worth nothing alone is valued in company: Groundwork needs three skills to have fired in the opening leg, so it is run on a reference kit. A skill that depends on who is around you (Nimble Navigator, Uma Stan) is read off recorded races of a reference field. A recovery skill adds no speed; it is valued as the stamina it stands in for, and only as many count as the distance has use for.

Ten points are one expected length of white spark. The other kinds are weighted against that — pink first, whites close behind, the unique, then blues — and that weighting is a stated priority, not a measurement: in plain expected lengths the whites lead.

Whether a spark fires: where it sits, its stars, your affinity

A legacy is the borrowable parent plus its two grandparents, and a spark fires at an inspiration event by three things, in this order: a grandparent's spark at half the parent's rate; more stars more often (a white at 3 / 6 / 9 % per event, a blue at 70 / 80 / 90 %, a pink at 1 / 3 / 5 %); and affinity scaling all of it by (1 + points / 100). Those are the community's numbers, and they were checked against the game's own inspiration-event captures at each capture's real affinity: they predict what fired with half the error of the flat rate this page used before, and they land within a few points on every kind and slot (whites 17 % / 4 % observed on parent / grandparent against 11 % / 5 % predicted; blues 100 % / 89 % against 100 % / 82 %).

A white is a hint, not the skill. Measured on 13 fired whites: every one that was not already owned raised the hint level of the skill of its name — by 1 or 3 levels, at random — and none granted the skill. So a white counts for what the skill is worth after the skill points it still costs, and a white not worth its points here counts for nothing. A parent's own unique is guaranteed at career start; a grandparent's rolls like a white.

A pink is a chance at a grade. The career-start grant is taken to stop at A (the community's rule); the step to S comes from a proc at one of the two inspiration events, so nine stars of the right pink reach S in a third to a half of careers, and one star on a grandparent in about two in a hundred.

COMMUNITY rates, CROSS-CHECKED on 9 unambiguous observations that share legacies — enough to prefer the model, not to recalibrate it. python -m umahelp constants --tag proc prints every number with its source; python umahelp/experiments/studies/inheritance_events.py reproduces the check.

How much to trust the order

A length is about a tenth of a second here. MEASURED in a simulated 9-runner meta field (Kyoto 2200m, 1,000 races per step, median-entrant build): a +0.10s solo advantage moved P(top-3) by +20 points, from 54% to 75% — about seven times the measurement's own error bar, and monotone all the way down. So a legacy whose sparks are worth two expected lengths is worth having.

But a spark is a chance, not a delivery: a top row's points are an average over careers in which most of its sparks do not fire. Rows inside one tier (five points) are the same borrow. Choose between them on the things the row shows beside the number — whether the sparks are the ones you are short of, and whether the trainer will accept you.

What is assumed, said once: the opponents a positional skill is tried against (eight runners built like yours, no skills — no real Champion Meeting field has been captured); the weights between pink, white, green and blue; and that the races the simulator was checked on predate the July 2026 balance patch, while the skill data it values is today's.