Coaching Deep-Dive

📈 Win-Probability model (pure-Go logistic regression trained on our own rounds — a v1 baseline, not gradient-boosting)

Model accuracy

75.7%

vs 50.2% base-rate · log-loss 0.48

Our win-probability through the round, averaged at 10-second marks — green = rounds we won, red = rounds we lost. Where the lines separate is where rounds get decided.

Win-Probability Added (WPA) — impact weighted by how much each action swung the round

Player	Role	WPA (pts)
KingG	Support / trade	2393.4
Knoxjq	Entry rifler (AWP later)	3727.0
Sh3nl0ng	IGL / flex / secondary-AWP	628.8

Where win-probability is won & lost — the round-phase walk (Δ our WP per phase; in LOST rounds, the most-negative phase is the breakdown to fix)

Side	Rounds	Opening (→ 1st blood)	Mid-round (→ plant)	Post-plant (→ end)	n
T	won	1.8%	26.0%	4.3%	491
T	lost	-2.3%	-19.6%	-2.6%	511
CT	won	2.2%	23.7%	1.9%	555
CT	lost	-2.4%	-25.3%	-3.2%	481

⏱ Tempo (we have every timestamp — this is where the clock lives)

Avg time-to-plant

52.3s

Avg first contact

17.2s

First-contact spread (σ)

±8.5s

A low first-contact spread = predictable timing (anti-strattable). Late plant times = slow/late commits.

💰 Economy decisions

Win the round AFTER we force

51.0% n=578

Win the round AFTER we eco

51.5% n=565

Dying-rich ($ unspent, lost rounds)

$21793

🛡 Discipline

Impossible defuses started

defuse you couldn't finish in time

🔮 Predictability — plant-site preference (entropy: 0 = always same site, 1.0 = perfect 50/50; low = exploitable)

Map	A	B	A %	Predictability (entropy)
de_anubis	1	3	25%	0.81
de_overpass	8	22	27%	0.84
de_mirage	44	21	68%	0.91
de_dust2	91	52	64%	0.95
de_cache	17	10	63%	0.95
de_nuke	49	62	44%	0.99
de_ancient	11	13	46%	0.99
de_inferno	43	46	48%	1.00

📍 Positional predictability (where we die in the same few spots — lowest entropy = most readable; ≥15 deaths)

Player	Map	Side	Deaths	Distinct spots	Predictability (entropy)
KingG	de_nuke	CT	154	28	0.56
Knoxjq	de_nuke	T	181	35	0.59
KingG	de_nuke	T	157	38	0.61
Sh3nl0ng	de_nuke	CT	139	35	0.63
Knoxjq	de_dust2	T	244	53	0.64
Knoxjq	de_mirage	CT	101	29	0.65
Knoxjq	de_inferno	T	168	42	0.66
Sh3nl0ng	de_nuke	T	145	43	0.67
KingG	de_dust2	T	228	56	0.67
Knoxjq	de_dust2	CT	222	53	0.67
Knoxjq	de_nuke	CT	155	43	0.68
KingG	de_inferno	CT	126	40	0.69

🎯 Clutch by situation (which spots we actually fold)

Player	Situation	Won	Attempts	Win %
KingG	1v2	13	71	18%
KingG	1v3	2	55	4%
KingG	1v4	0	59	0%
KingG	1v5	0	25	0%
Knoxjq	1v2	7	37	19%
Knoxjq	1v3	4	42	10%
Knoxjq	1v4	0	33	0%
Knoxjq	1v5	0	15	0%
Sh3nl0ng	1v2	23	85	27%
Sh3nl0ng	1v3	10	91	11%
Sh3nl0ng	1v4	1	76	1%
Sh3nl0ng	1v5	0	41	0%

🗺 Map record (veto picture)

Map	W	Played	Win %
de_dust2	302	586	51.5%
de_nuke	255	454	56.2%
de_inferno	218	412	52.9%
de_mirage	145	263	55.1%
de_cache	51	126	40.5%
de_overpass	45	113	39.8%
de_ancient	23	64	35.9%
de_anubis	7	20	35.0%

🧱 Role accountability — time alive, utility & info-trade

Player	Role	Avg time alive (anchor = buys time)	Unused util $ at death	Win % after their 1st-death (info-trade; team avg 34.5%)
KingG	Support / trade	48.1s	$151	22.3%
Knoxjq	Entry rifler (AWP later)	37.3s	$109	29.5%
Sh3nl0ng	IGL / flex / secondary-AWP	48.8s	$311	38.0%

"Win % after their 1st-death" above 34.5% = that player's early deaths still tend to win the round (they buy info/space — a productive sacrifice); below = costly.