Backcheck
Combinations · chemistry

The Line Lab

Which trios actually work when they're on the ice together. We compare what a line shouldproduce given who's on it against what it actually produces — the difference is chemistry. Every unit below really played together at 5-on-5; nothing is hypothetical.

How to read this

Nothing here is invented: every line and pair below actually played together at 5-on-5, with real minutes attached. For each unit we show its observed — the slice of scoring chances it controlled while on the ice, where 50% means break-even.

The interesting part is the split. We first compute an : what those players should produce together if a line were just the sum of its parts. Whatever the real unit does beyond that is the — the closest a spreadsheet gets to measuring whether three players actually fit. Small-minute units are noisy, so read a big bonus on 30 minutes as a hypothesis, not a verdict — and remember this is a model estimate, not the coach's lineup card.

BASE
what the unit “should” produce — its members' individual profiles, added up
CHEM
observed share minus base: the whole vs. the sum of the parts
PROJ
base + chem — the model's expectation if the unit stays together
TICK
the hairline on each bar marks 50% — the break-even line
Fine print · verbatim from the model
  • Lines recovered from MoneyPuck lines.csv (5on5, observed deployment); chemistry = observed xG% minus additive archetype base. Model-estimated.
NSHNashville Predators62 units observed · 5v5As of Sun, Sep 20
Forward lines
3 units
  • 01
    Steven Stamkos – Ryan O'Reilly – Luke Evangelista
    Together this unit controls play outright — 60% of expected goals, 8 points better than these players' profiles predict. Real chemistry.
    267 min together 52 +8 → proj 60
    60.0%
    · obs.
  • 02
    Tyson Jost – Fedor Svechkov – Matthew Wood
    Together this unit tilts the ice — 54% of expected goals, 10 points better than these players' profiles predict. Real chemistry.
    86 min together 44 +10 → proj 54
    54.0%
    · obs.
  • 03
    Erik Haula – Jonathan Marchessault – Filip Forsberg
    Together this unit gets buried territorially — 37% of expected goals, 7 points WORSE than the talent predicts — worse together than apart.
    121 min together 44 -7 → proj 37
    37.0%
    · obs.
Defense pairs
2 units
  • 01
    Nick Perbix – Adam Wilsby
    Together this unit controls play outright — 56% of expected goals, 29 points better than these players' profiles predict. Real chemistry.
    100 min together 27 +29 → proj 56
    56.0%
    · obs.
  • 02
    Roman Josi – Nicolas Hague
    Together this unit controls play outright — 55% of expected goals, 26 points better than these players' profiles predict. Real chemistry.
    180 min together 29 +26 → proj 55
    55.0%
    · obs.
◆ The ModelObserved 5v5 deployment · MoneyPuck linesModel-estimated proxy, not a lineup card

Lines recovered from MoneyPuck lines.csv (5on5, observed deployment); chemistry = observed xG% minus additive archetype base. Model-estimated.