Defensive Bleed After Prosperity Events: Win / Loss + Lead-Conversion Layer

Question: Does defensive leakage after Boston prosperity events show up specifically in games Boston loses after building an 8+ second-half lead?

Window: 81 Celtics playoff games, 2021-22 through 2025-26. 56 games contained at least one prosperity event and at least one second-half +8 moment.

Sample: 1,898 prosperity ender events, 1,878 matched to opponent next-possession outcome (98.9%). Of those, 156 occurred in the 9 losses Boston suffered after holding a +8 second-half lead, and 1,722 occurred in the 47 wins.

---

TL;DR (lead with the uncomfortable read)

The prior verdict was wrong, and it was wrong because the team-wide average hid a clustered failure pattern in losses.

Conditioned on the 1,898-event team-wide sample, opponent next-possession PPP after a Boston prosperity event is 0.966 (below the playoff offensive baseline). That was the prior finding.

Conditioned on win / loss:

• Boston wins: opp PPP = 0.944. Even better than the overall number.
• Boston losses: opp PPP = 1.205. A 0.26 PPP gap, on a 156-event sample.
• Lead controlled (32 games): 0.909 PPP. Lead stressed (8 games): 1.123. Lead blown (16 games, 7-9 record): 1.136.

And the prior "missed threes don't bleed defense" conclusion completely flips in the loss subset:

• Missed three in wins: 0.937 PPP. Missed three in losses: 1.463 PPP. +0.53 delta.
• The largest single bleed source in loss games is the missed three, not the turnover.

Updated verdict: defensive leakage IS materially worse in Boston losses (verdict #1). The prior "missed threes are neutral" claim was an averaging artifact. Star turnovers still bleed but the loss-side sample is tiny (7 events); the loss-side missed-three signal is more robust (41 events).

This supports the broader Stevens "let them back in" narrative when reframed as a leverage-timing pattern, not a steady-state team defect.

---

A. How many Celtics playoff games had a second-half +8 lead?

56 of 81 Celtics playoff games. The other 25 either never had Boston up 8+ in the second half, or were over by then.

B. Boston's record in those games

47-9. 84% win rate when Boston leads by 8+ in the second half. The 16% loss rate is the population this report investigates.

Table 1. Game conversion summary

Every loss in the 56-game sample came from a fully blown lead (opponent tied or led at some point after Boston first hit +8). Stressed leads always converted to wins in this dataset.

C-D. Defensive leakage in wins vs losses

Table 2. Opponent next-possession PPP by player group, wins vs losses (all seasons)

Every group bleeds materially more in losses. Brown's loss-side delta (+0.30) is actually larger than Tatum's (+0.18), which contradicts the prior report's claim that the bleed was Tatum-specific. The bleed is team-wide in losses, not concentrated in a single player.

The Other_Celtics loss delta (+0.28) is also striking. The Stevens "let them back in" pattern is not a star-driven phenomenon in this sample.

Table 3. Opponent PPP by Boston event type, wins vs losses (team-wide, all seasons)

The two largest deltas (turnover and missed three) sit on small loss-side samples. The missed-three loss sample (41 events) is the largest of the truly elevated buckets, which makes it the most defensible signal.

E-F. Tatum and Brown bleed in losses

Table 4. Tatum prosperity bleed: wins vs losses

The Tatum loss-side turnover sample is 3 events. The point estimate (1.667 PPP) cannot be defended.

Table 5. Brown prosperity bleed: wins vs losses

The most striking Brown finding is that made shots in losses (n=10) bleed at 1.50 PPP. Plausible mechanism: Brown's makes in lead-protection mode are often pull-up midrange jumpers off the dribble against a set defense, which neither force the opponent to a deep inbound nor get Boston to a comfortable defensive transition. Public PBP cannot prove this. The n=10 sample is too small to assert.

The Brown loss-side missed-three sample (n=8) bleeds 1.625 PPP — directionally consistent with Tatum (1.571) and the team total (1.463). It cannot be ruled out as a real signal but it cannot be confirmed either.

G. Star turnovers in losses

The seven star turnovers in losses (3 Tatum, 4 Brown) bleed 1.43 PPP. Compared to the 40 star turnovers in wins (1.175 PPP), that is a +0.25 delta. Sample is too small to be a load-bearing claim but it is directionally consistent with the prior report's star-turnover findings.

The strongest defensible turnover statement remains the one from the prior report: in the full multi-year sample, Tatum and Brown turnovers bleed (1.21 PPP) and role-player turnovers do not (0.81). The win/loss split does not move that needle either way.

H. Missed threes in losses

This is the headline shift from the prior report.

Table 6. Missed three leakage: wins vs losses, by player group

The Other_Celtics missed-three loss sample is 26 events — the most robust subset in the table. Even there, opponents score 1.385 PPP on the next trip, ~50 percent higher than the matching win-side number.

The prior verdict ("missed threes do not create defensive bleed") was correct at the team-wide level but masked a strong loss-conditional effect. Stated correctly: missed threes are defensively neutral in wins, materially damaging in losses.

Possible mechanisms (none provable from public PBP):

• Lead-protection three attempts come from worse offensive sets in tighter games and therefore produce more open opponent runouts.
• In losses, Boston's defensive transition cohesion is worse, so the same shot result produces a worse defensive consequence.
• Reverse causality: games in which Boston's defense was already slipping are more likely to be losses, and those defensive slippages are what produced the bleed. The prosperity-event shot type is correlated, not causal.

The data cannot adjudicate between these.

Lead-conversion view: defense degrades monotonically

Table 7. Team prosperity bleed by lead-conversion category (all seasons)

Defense degrades materially between LEAD_CONTROLLED and the two stressed categories. The gap between LEAD_STRESSED and LEAD_BLOWN is small in opp PPP terms (1.123 vs 1.136) but enormous in outcome terms (100% win rate vs 43.8%). The defensive bleed is not the whole story of the loss; it is one observable correlate of the broader failure.

Table 8. Lead conversion by player group (within blown-lead games)

Within blown-lead games, Tatum's prosperity events bleed 1.39 opp PPP across 72 events — the most robust extreme bleed signal in the entire dataset. This is the cleanest version of the "Tatum prosperity damages defense" story. It is specific to games where the lead was actually blown, not games where it was merely stressed.

I. Worst loss games

Table 9. Boston losses after holding a 2H +8 lead

The two large-sample loss games (NYK G1, n=27; NYK G2, n=44) tell consistent stories: Boston held 20-point second-half leads, gave up 1.0 to 1.15 opp PPP on the next trip after prosperity events, and lost by 1 and 3. The 2024-25 NYK series is the single clearest in-season example of defensive bleed correlating with loss outcomes.

The 2022-23 BOS-ATL G2 (n=35) is the counter-example: Boston was up 13, lost by 2, and opp prosperity PPP was 0.886 — actually below team baseline. That loss was driven by something other than next-possession defensive bleed.

J. Does this change the prior verdict?

Yes.

Prior verdict: "Turnovers create defensive bleed, missed threes do not, and the bleed is concentrated in Tatum prosperity possessions, not team-wide."

Updated verdict: "Defensive leakage is materially worse in Boston losses (verdict #1 in the analytical menu), and missed threes are a major contributor to that loss-side bleed when conditioned on game result. The bleed is team-wide in losses, not concentrated in one player. The prior 'missed threes are neutral' conclusion was an averaging artifact that washed across 1,722 win-side events; the loss-side picture shows the opposite."

Specifically:

• The team-wide 0.966 opp PPP figure remains true and is no longer the most useful number for this question.
• The win-conditioned 0.944 / loss-conditioned 1.205 split is the new headline.
• The Tatum-versus-Brown gap shrinks in losses; the bleed is broader than a single-player story.
• Star turnovers still bleed but the loss-side sample (n=7) is too small to be the main signal.

K. Stevens "let them back in" narrative

Supported, but with the loss-side qualification. Boston's prosperity defense in lead-controlled games is excellent (0.909 PPP, well below playoff offensive baselines). The "let them back in" pattern shows up specifically in the games where the lead is eventually blown, which is exactly the population Stevens was describing. The defensive cost of lead-protection offense is not paid every game; it is paid in the small set of games where the lead actually slips, and in those games it is paid in a clustered, statistically visible way.

The narrative is best stated as: Boston's prosperity defense is fine on average. Boston's prosperity defense in the games it loses is materially worse, and that worse defense correlates with elevated next-possession opponent scoring on missed threes specifically.

Required: Small-sample warnings

Table 10. Sample-size warnings for this report

Message-board-ready paragraph

Defensive bleed after Boston prosperity events looked fine across 1,898 playoff prosperity possessions from 2021-22 through 2025-26: opponents averaged 0.97 points per possession on the next trip, below typical playoff offense. But split by game outcome, the picture flips. In Boston's 47 wins after holding a +8 second-half lead, opponents averaged 0.94 PPP on the next trip. In the 9 losses, opponents averaged 1.20. The single biggest event-type contributor inside those losses was the missed three, where opponents scored 1.46 PPP versus 0.94 in wins. The bleed is team-wide in losses (Tatum, Brown, and Other Celtics all show +0.18 to +0.30 deltas), and it tracks Boston's lead-conversion category cleanly: 0.91 PPP allowed when leads hold, 1.12 when leads are stressed, 1.14 when leads are blown. Public play-by-play cannot prove which way the causal arrow runs between the lead-protection offense and the worse defense, but the leverage-timing pattern is consistent with the Stevens "let them back in" framing. The previous finding that "missed threes don't bleed defense" was correct only as a team-wide average and is reversed once games are split by result.

What public data can / cannot prove (carry-forward from prior report, updated)

Can:

• Boston went 47-9 in playoff games where it held a 2H +8 lead. The 9 losses concentrated in 2021-22 (1), 2022-23 (3), 2024-25 (4), 2025-26 (1).
• Opponent next-possession PPP after Boston prosperity events is 0.94 in those wins, 1.21 in those losses. Delta +0.26.
• Missed threes account for the largest event-type delta with a defensible sample size (41 loss events).
• Tatum's prosperity bleed is at its worst in blown-lead games (1.39 PPP across 72 events).
• 2023-24 (championship year) had exactly one blown-lead game in this dataset, which Boston won. That season's defense did not exhibit the loss-side bleed pattern.

Cannot:

• Whether the elevated next-possession opp PPP after a Boston missed three in a loss is caused by long rebounds, by transition defense lapses, by shot-clock state, or by selection bias (the games where Boston loses are the games where the opponent was already attacking better).
• Whether star turnovers in losses are live-ball or dead-ball.
• Whether Brown's elevated "made shot in losses" bleed (1.50 PPP, n=10) is a real pattern or a small-sample artifact.
• Whether the 2024-25 New York series losses were driven by defensive transition specifically or by a broader scheme problem.
• Whether the prosperity-event-to-defensive-bleed mapping is causal or merely co-occurrent. The most rigorous causal test would require possession-level lineup, defender-distance, and shot-clock data not in public PBP.

---

Methodology notes

• Lead-conversion flags computed from the cached PlayByPlayV3 stream per game. lead_stressed_flag = True if the opponent cut the BOS margin to <= 3 at any point after BOS first hit +8 in the second half. lead_blown_to_tie_or_deficit_flag = True if the opponent tied or led at any point after that first +8 moment. Categories are ordered: LEAD_BLOWN_TO_TIE_OR_DEFICIT (strongest), LEAD_STRESSED (intermediate), LEAD_CONTROLLED (default).
• Series outcomes derived from the win counts in data/celtics_playoff_games_discovered.csv. Boston won the series in 11 of the 14 (season, opponent) pairs in scope; lost in 3 (2021-22 GSW, 2022-23 MIA, 2024-25 NYK, 2025-26 PHI).
• Win/loss split inherits directly from the wl field already in the discovered-games CSV.
• Bucket collapse: made_shot = made_rim + made_midrange + made_three. All other Boston event types retained as in the prior report.
• "Force event" inherits the prior definition: rim attempt OR free throw event.

Files

• data/celtics_defensive_bleed_game_conversion.csv — one row per game with prosperity events.
• data/celtics_defensive_bleed_win_loss_summary.csv — events x (season, result, group, bucket).
• data/celtics_defensive_bleed_lead_conversion_summary.csv — events x (season, stressed, blown, group, bucket).
• data/celtics_defensive_bleed_tatum_brown_loss_delta.csv — Tatum / Brown / combined wins vs losses delta.
• data/celtics_defensive_bleed_worst_loss_games.csv — 9 worst loss games with per-event-type bleed.
• data/celtics_defensive_bleed_win_loss_run.log — append-only run log.
• scripts/build_celtics_defensive_bleed_win_loss.py — reproducible build.
• Prior outputs preserved: data/celtics_defensive_bleed_after_prosperity_events.csv and the original summaries.

Generated 2026-05-11.