The formula.
points = 0.002064 × age_mult × level_mult × condition_spent × efficiency,
where condition_spent = slots × intensity and
efficiency = 112.5 below an average of 80, 90 from 80, 67.5 from 100, 45 from 120 and
22.5 from 140 to 180, with avg the drill average. Slot count, drill intensity and drill size all act through condition spent — they
need no separate terms. Note that condition cost per session is flat at any level; what a higher drill average
reduces is the points that session buys.
Efficiency comes in steps. Inside each band of the drill average the rate is
flat, and at 80, 100, 120 and 140 it drops suddenly, each time by a quarter of the 80–100 rate (rates
1.25 / 1.00 / 0.75 / 0.50 / 0.25). There is no step at 60. Measured session by session on one player who
trained one drill from an average of 49.5 to 145, and found again, independently, in 42 runs from 16 other
players. Older versions used a smooth slope, which made a player who started just above a step look slower
than he is. Above an average of about 157 the rate is untested.
Age. ×1.0 for 18–21, then
×0.785 per year (95% CI 0.705–0.863). Measured on identical runs at
ages 18, 19, 21, 23 and 26; beats a no-age model by ΔAIC = 28.6. Not a level effect in
disguise — the 26-year-old had the lowest skills of any subject, where the model predicts it should
have been fastest. Age 22 is measured directly: the same player trained at 21 and
again at 22, and his rate fell to ×0.64 (95% CI 0.59–0.70), so 22 uses
that value. Ages 23 and up follow the fitted curve. The 18–21 plateau is checked on the same players:
six players repeated their exact earlier run a season older, and 18→19 came out
×1.00 ± 0.03, 19→20 ×0.99 ± 0.04.
Rotation. One pass over every skill the drill trains, then a second pass
over the highlighted ones. Verified exactly: on Carioca with Ladders the rotation predicted 26.00 and 13.00
points; the real log gave 26 and 13.
Player speed. Each player has his own training speed and keeps it. Players
trained in the same sessions differ clearly (eleven measured so far span 0.83–1.11×), while the
same player repeats his speed within about 1–3% across runs weeks apart, other drills, a promotion and a
birthday. The average player sits at 1.00×; most players fall within about
±11%. Calibrating with one logged run replaces that with the player's own number: learning from a
player's first run and predicting his latest missed by 3.7% on average, against 11.8% for the average
player.
Overall (OVR). A player's OVR is simply the average of all 15 skills,
rounded. Roles don't change it. Checked against 17 in-game screenshots of 12 players in 9 positions: every
one matched. So the OVR after training is exact given the skill values; the uncertainty is only in how fast
the skills get there. Goalkeepers are not covered.
Stars. Stars = 1 + skill value ÷ 20: one star per 20%, so 100% is
6 stars and the 180% training cap is 10 stars. A player's star bar follows his unrounded OVR; this matched
the in-game bar on 3 screenshots. You can enter targets as stars; the plan is identical either way, since
stars are just another way of writing the skill value.
Condition. slots × intensity (Very Easy 0.75% · Easy 1.5% ·
Medium 2.25% · Hard 3% · Very Hard 3.75%), flat in level and age. Rest Packs restore 15%.
A Campus condition drain boost cuts the condition drained but not the points: at −10% a
25-session run drained 0.89× the listed cost while gaining exactly what the listed cost predicts.
Bigger boosts (−20% to −50%) are assumed to work the same way.
Drill level. The game says upgrading a drill adds 10% at Semi-Pro, 20% at
Pro and 30% at World Class, and that holds up. One player trained 25 sessions at each level, so
player-to-player differences cancel out. Measured: Semi-Pro ×1.11, Pro
×1.23, World Class ×1.31, each within half
a standard error of the advertised value. The calculator uses the advertised
1.10 / 1.20 / 1.30. Level speeds up gains only: every run drained exactly the
listed condition, so upgraded drills are not cheaper per session. (An earlier test that found only +4% at
Semi-Pro is overturned. Its "Amateur" comparison run was most likely on an upgraded drill.)
Where it's weakest. Age 22 rests on one player measured against himself;
23 and up come from players compared with each other. Ages past 26 are extrapolated. After a promotion every
skill drops about 20; training follows the new numbers, so just enter what the game shows. Most runs logged since October used 6 slots and fit
the model like the rest (the early hint that 6 slots run ~15% low did not hold up); a direct 5-vs-6-slot test
on one player hasn't been done. Player-to-player variation of about ±11% is real and unexplained. Drill levels were
measured on one player; the +10% steps are the game's advertised values, which that test matched. The rate was fitted on
six drills; the rest of the catalogue is an educated extension. Teamplay drills and Goalkeeper Training
aren't included at all. Ranges shown are 10th–90th percentile, not guarantees. Session-to-session luck
turned out to be small (each skill banks a hidden fraction until it becomes a whole point), so for a player you
have calibrated the range is narrow; without calibration it mostly reflects the ±11% between players.