Producer: archetype_chemistry.py (2026-09-04). Forward-only 5v5 units from unit_chemistry_2010_26.csv (era-adjusted WOWY, 2010-11 → 2025-26): 967 trios (≥100 together min) and 2,849 duos (≥150 min) where every member has a career archetype. Lift60 = together net xG/60 − same-season apart net xG/60. All means TogMin-weighted.
Macro roles: S scorer (First-line star, Top-six sniper) · P play-driver (Two-way top-six, Complementary skill winger) · G grinder (Net-front, Shutdown/PK C, Checker, Enforcer).
| Line | n | together min | Lift60 (±SE) | apart net60 |
|---|---|---|---|---|
| PSS | 176 | 63,497 | +0.310 ± 0.040 | +0.141 |
| SSS | 62 | 20,717 | +0.279 ± 0.060 | +0.025 |
| PPS | 148 | 45,232 | +0.263 ± 0.045 | +0.168 |
| GPP | 63 | 14,923 | +0.262 ± 0.064 | +0.096 |
| GGP | 86 | 19,700 | +0.260 ± 0.065 | +0.036 |
| GGG | 90 | 21,942 | +0.220 ± 0.054 | -0.065 |
| GGS | 45 | 9,798 | +0.218 ± 0.079 | +0.165 |
| GSS | 125 | 34,960 | +0.196 ± 0.045 | +0.093 |
| GPS | 150 | 31,473 | +0.104 ± 0.048 | +0.155 |
| PPP | 22 | 5,395 | +0.050 ± 0.120 | -0.154 |
| Line | n | together min | Lift60 (±SE) | apart net60 |
|---|---|---|---|---|
| 1 | 40 | 9,855 | +0.316 ± 0.097 | +0.106 |
| 2 | 435 | 139,226 | +0.256 ± 0.027 | +0.106 |
| 3 | 492 | 118,555 | +0.211 ± 0.024 | +0.092 |
League-wide trio lift: +0.238 (lines that stay together ≥100 min are, by selection, the ones that worked).
| Term | coef | SE | t |
|---|---|---|---|
| const | +0.255 | 0.095 | +2.7 |
| apart_net60 | -0.144 | 0.052 | -2.8 |
| 2 types (vs 1) | -0.062 | 0.085 | -0.7 |
| 3 types (vs 1) | -0.121 | 0.085 | -1.4 |
| spread finishing | +0.116 | 0.062 | +1.9 |
| spread playmaking | +0.094 | 0.085 | +1.1 |
| spread hits | +0.033 | 0.058 | +0.6 |
| mean finishing | -0.125 | 0.043 | -2.9 |
| mean playmaking | +0.142 | 0.056 | +2.5 |
| mean hits | -0.057 | 0.039 | -1.5 |
| Star | Sniper | Two-way | Skill wing | Net-front | Shutdown C | Checker | Enforcer | |
|---|---|---|---|---|---|---|---|---|
| Star | +0.26 53 | +0.18 122 | +0.27 143 | +0.10 122 | -0.01 131 | -0.08 38 | -0.01 35 | +0.01 20 |
| Sniper | +0.18 122 | +0.19 82 | +0.16 138 | +0.05 154 | +0.10 174 | -0.04 78 | +0.03 60 | -0.04 19 |
| Two-way | +0.27 143 | +0.16 138 | +0.15 68 | +0.08 139 | +0.14 162 | +0.06 47 | +0.02 50 | +0.22 24 |
| Skill wing | +0.10 122 | +0.05 154 | +0.08 139 | -0.02 77 | +0.10 157 | -0.13 77 | +0.01 75 | -0.00 33 |
| Net-front | -0.01 131 | +0.10 174 | +0.14 162 | +0.10 157 | +0.10 83 | +0.15 79 | +0.03 87 | +0.02 58 |
| Shutdown C | -0.08 38 | -0.04 78 | +0.06 47 | -0.13 77 | +0.15 79 | +0.01 25 | +0.06 89 | +0.06 45 |
| Checker | -0.01 35 | +0.03 60 | +0.02 50 | +0.01 75 | +0.03 87 | +0.06 89 | +0.01 43 | +0.10 51 |
| Enforcer | +0.01 20 | -0.04 19 | +0.22 24 | -0.00 33 | +0.02 58 | +0.06 45 | +0.10 51 | — |
Same-type duos: +0.147; mixed-type duos: +0.109 (TogMin-weighted).
apart_net60 is controlled in the regression because high-quality players have less room to show lift (regression to the mean).