Celtics Prosperity, Split by Game Result (2021-22 through 2025-26)

Generated 2026-05-11. Win/loss conversion layer on top of the existing Celtics prosperity work. Splits the 1,898 prosperity ender events into games Boston won and games Boston lost. Adds game-level conversion stats (did Boston build a +8 second-half lead, did the opponent cut it back, what was the final margin) and joins the defensive-bleed file.

Sources:

• data/celtics_all_years_tagged_events.csv (11,394 tagged BOS events)
• data/celtics_playoff_games_discovered.csv (81 games, BOS W/L)
• data/league_playoff_tagged_events.csv (full event stream both teams)
• data/celtics_defensive_bleed_after_prosperity_events.csv (1,898 bleed rows)

Outputs:

• data/celtics_prosperity_game_conversion.csv (81 games)
• data/celtics_prosperity_win_loss_summary.csv (226 rows)
• data/celtics_prosperity_lead_conversion_summary.csv (65 rows)
• data/celtics_prosperity_win_loss_defensive_bleed.csv (247 rows)
• data/celtics_tatum_brown_win_loss_delta.csv (94 rows)

---

Definitions

• Prosperity event: BOS event, period >= 3, BOS margin >= 8 entering event, event type FGA / FT / Turnover. ender_only view unless noted.
• Reached +8 2H: at any point in period >= 3, Boston's lead crossed +8 (computed from full event-stream margin, not just BOS-event sample).
• Lead-stressed: reached +8 in 2H AND opponent later cut margin to <= 3 at some subsequent event.
• Lead-blown: reached +8 in 2H AND opponent later tied or led.
• Force event proxy: rim attempt OR free throw event OR assisted teammate three created.

---

Table 1. Game conversion table

How often did Boston build a 2H +8 lead, and what happened next?

Read: across 5 postseasons Boston reached a +8 second-half lead in 57 of 81 playoff games. Boston won 48 of those, lost 9. The 2024-25 second round (NYK loss) is the worst conversion year: only 60% of +8-2H-lead games turned into wins. 2023-24 (championship) is the cleanest: 14 for 14.

Table 2. Worst lead-conversion games (17 lead-blown games)

Sorted by final margin (most negative first):

Read: when Boston reaches +8 in 2H and the opponent later ties or leads, the win rate is 8 of 17 = 47%. Six of those 9 losses come from 2024-25 NYK or 2022-23 MIA (ECF). The 2025-26 PHI Game 5 (-16 finish from a +13 lead) is the single largest blown lead on the ledger.

Table 3. Boston prosperity offense in wins vs losses (ALL SEASONS, ender_only)

Headline: the simple "Tatum collapses in prosperity losses" reading the prior reports primed for is not what the data shows. Tatum's prosperity PPP is actually higher in losses (0.923) than in wins (0.828). Brown's is dramatically lower in losses (0.634 vs 0.915).

Table 4. Tatum vs Brown in prosperity losses

Read: when Boston loses a game after building prosperity:

• Tatum attacks the rim MORE (28.2% vs 19.5%), takes more threes, fewer midranges, turns it over slightly more (+3.75 pp), and is actually MORE efficient (0.923 PPP) than in wins. He behaves like he is trying to drag the team back.
• Brown attacks the rim much LESS (14.6% vs 24.6%), takes more threes, doubles his turnover rate (9.3% to 19.5%), and his force rate collapses by 20 pp (46.6% to 26.8%). His PPP falls to 0.634, which is replacement-level. He behaves like he is settling.

This is the most consistent player-specific signal in the entire five-year dataset.

Table 5. Defensive bleed in wins vs losses (ALL SEASONS, opponent's next possession PPP)

After Boston turnovers:

After Boston missed threes:

After Boston missed rim attempts:

Read: opponents score significantly more often on the possession immediately after a Boston failure (turnover or missed three) in games Boston loses than in games Boston wins. The difference is +0.30 to +0.50 opponent PPP across player groups. The largest single spike is opp PPP of 1.444 after a Brown missed three in a loss (vs 1.085 in wins) and 1.385 after Other_Celtics missed threes (vs 0.875 in wins). These are the "let them back in" possessions Stevens described, and they cluster in losses.

Table 6. Lead-controlled vs lead-stressed comparison (ALL SEASONS, ender_only)

Game classification:

• lead-controlled: reached +8 2H, opponent never cut to <= 3. 30 games, all wins.
• lead-stressed only: reached +8 2H, opponent cut to <= 3 but never tied/led. 9 games, all wins.
• lead-blown: reached +8 2H, opponent tied or led at some point. 17 games, 8 wins / 9 losses (47% win rate).

Read: Boston is highly efficient in prosperity even in games where the opponent later tied/led the score (team PPP 0.838 in lead-blown games vs 0.962 in lead-controlled). The drop is about -0.12 PPP. Brown's prosperity PPP in lead-blown games is 0.672, far below his 0.954 in lead-stressed-only and 0.921 in lead-controlled. Tatum's prosperity PPP actually rises in lead-blown games (0.897 vs 0.818 in controlled). Once again the within-game Brown drop is the signal.

---

A. How often does Boston win after having an 8+ second-half playoff lead?

48 of 57 = 84%. That is a high win rate by playoff standards.

B. How often does Boston lose after having an 8+ second-half playoff lead?

9 of 57 = 16%. Of those 9 losses, all were "lead-blown" games where the opponent later tied or led; none came from games Boston led the whole way.

C. Which seasons/series produced the worst lead-conversion failures?

By win rate among games where Boston reached +8 in 2H:

• 2024-25 (NYK series): 60% (6 of 10). Worst in the dataset. Boston lost 4 games after building a +8 2H lead, including Game 1 (-3 after +20) and Game 2 (-1 after +20).
• 2022-23: 79% (11 of 14). Three losses include the two MIA ECF home blowups (G1 and G2) and the ATL G5.
• 2021-22, 2025-26, 2023-24: all >= 75% win rate; 2023-24 was 100%.

By largest single-game blown lead:

• 2025-26 PHI Game 5: +13 → -16 final, swing -29.
• 2024-25 NYK Game 4: +14 → -8 final, swing -22.
• 2022-23 MIA ECF Game 1: +12 → -7 final, swing -19.

D. Does Boston's prosperity PPP collapse in losses compared with wins?

Modestly. Team PPP drops from 0.937 in wins to 0.855 in losses (delta -0.082). Force rate drops dramatically: 39.56% wins → 29.56% losses (-10 pp). Three rate rises in losses (35.77 → 41.51). So in losses Boston is still scoring at a respectable rate but doing so with far less rim/FT/AST3 pressure. The shift is consistent with Stevens's "lacks rim pressure" complaint.

E. Does Tatum's turnover rate spike specifically in prosperity losses?

Slightly, not dramatically. Tatum's TO rate rises from 11.63% in wins to 15.38% in losses (+3.75 pp). The prior multi-year "Tatum prosperity TO surge" finding (+1.13 pp vs his baseline) is real and slightly amplified in losses, but Tatum's overall efficiency in prosperity losses is actually higher (PPP 0.923) than in wins. His turnovers do not appear to be the proximate cause of the team's prosperity-loss collapse.

F. Does Brown's shot profile change in prosperity losses?

Yes, dramatically. Brown is the largest within-player shift in prosperity in the entire dataset when split by W/L:

• PPP: 0.915 → 0.634 (-0.281).
• Rim rate: 24.58% → 14.63% (-9.95 pp).
• Force rate: 46.61% → 26.83% (-19.78 pp).
• Turnover rate: 9.32% → 19.51% (+10.19 pp, more than double).
• Three-point rate: 25.14% → 34.15% (+9.01 pp).

In losses Brown stops attacking the rim, settles for threes more often, and turns the ball over more than twice as often as in wins. This is the single clearest player-level finding in the win/loss split.

G. Do Other Celtics still outscore the Jays in prosperity losses?

Yes, and the gap widens. In losses:

• Other_Celtics PPP: 0.937 (79 events).
• Tatum_Brown_combined PPP: 0.775 (80 events).
• Gap: 0.162 PPP.

In wins the gap was 0.128 PPP (Other 0.995, Jays 0.867). So the structural "role players outscore stars in prosperity" gap is present in both wins and losses, slightly larger in losses.

H. Does defensive bleed appear more strongly in losses?

Yes, very clearly. Opponent next-possession PPP is +0.30 to +0.50 higher in losses than in wins, after every type of Boston failure event with meaningful sample. After missed threes specifically:

• After Brown missed three: opp PPP 1.085 in wins, 1.444 in losses.
• After Other_Celtics missed three: 0.875 in wins, 1.385 in losses.
• After Tatum missed three: 0.897 in wins, 1.222 in losses.

After turnovers:

• After Tatum turnover: 1.160 in wins, 1.667 in losses (small N=3).
• After Brown turnover: 1.200 in wins, 1.250 in losses (small N=4).

The opponent scores more after every Boston failure in losses. This is the strongest support for Stevens's "let them back in" framing in the entire analysis.

I. Are missed threes or turnovers more connected to losses?

Missed threes by volume; turnovers by per-event severity.

In prosperity losses (159 ender events team-wide):

• 41% three-point attempts (~65 events, 47 misses likely).
• 14.5% turnovers (~23 events).

The volume of misses far exceeds the volume of turnovers, and missed threes in losses produce opponent PPP of 1.1-1.4. Turnovers in losses produce opponent PPP of 1.25-1.67 per event but the count is small. Multiplied out: missed-three opponent value in losses ~= 47 events x 1.25 avg opp PPP = ~59 opp points; turnover opponent value in losses ~= 23 events x 1.4 = ~32 opp points. Missed threes contribute more total bleed in absolute terms; turnovers are more dangerous per event.

J. Does this change the conclusion from the league baseline?

Yes, in two specific ways.

1. The original multi-year and league-baseline conclusion was that Boston is a top-quartile prosperity team and that within Boston, Tatum's TO rate is the only player-specific signal. The W/L split adds: Brown is the cleanest player-specific signal in the loss subset (PPP -0.281 in losses, force rate -19.78 pp, turnovers 2x higher). Tatum's TO surge survives but is smaller than Brown's collapse on every dimension once you split by result.

2. The league baseline showed Boston is above league mean in prosperity PPP. The W/L split shows Boston is above league mean because of the wins: team PPP 0.937 in wins is ~150 points above the league average of 0.79. In losses team PPP drops to 0.855, which is closer to league average. So Boston's "league-best prosperity team" status is largely a story of being excellent when winning, normal when losing. That is not unusual (every team is more efficient in games they win), but it does soften the original headline.

K. Is the real issue possession efficiency, lead conversion, or clustered leverage failure?

Lead conversion and clustered leverage, more than possession efficiency. Three reasons:

1. Boston's prosperity PPP does not collapse in losses (team total 0.855 vs 0.937 in wins; -0.082 is meaningful but not catastrophic). Possession-by-possession Boston still scores at a respectable rate even in games it loses.

2. The "lead-blown" win rate is 47%, far worse than the 84% win rate when Boston only needs to hold +8. The conversion failure mode is not "we play worse possession by possession" — it is "we give it back enough times that we live in coin-flip territory."

3. Defensive bleed is the metric that diverges most cleanly between wins and losses (+0.30 to +0.50 opp PPP after BOS failures). In losses, the same Boston miss or turnover produces a much higher probability of an opponent score. That is clustered leverage failure in the Stevens-described "let them back in" sense, not pure possession-level inefficiency.

The cleanest characterization: Boston's prosperity problem is not a prosperity-possession-efficiency problem. It is a defensive-after- failure problem plus a Brown-specific shot-creation collapse in losses.

---

Verdict mapped to the user's options

Best single-sentence verdict: Boston's prosperity problem in losses is two specific things Brown's individual collapse (shot selection and turnovers) and a team-wide defensive-bleed surge on the next possession after Boston failures not a Tatum-specific signal and not a team-wide possession-efficiency problem.

---

Message-board paragraph

Split by game result, the Tatum-vs-Brown story flips. Across five postseasons of Celtics playoff data, Tatum's prosperity PPP is actually higher in losses (0.923) than in wins (0.828), and his rim attempt rate goes up, not down. Brown is the opposite: in prosperity losses his PPP drops 28% (0.915 to 0.634), his force-event rate collapses by 20 percentage points (he stops attacking the rim and stops drawing free throws), and his turnover rate more than doubles (9.3% to 19.5%). The other big finding is defensive: the opponent scores 0.30-0.50 more points per possession after every type of Boston failure event (missed three, turnover, missed rim) in games Boston loses than in games Boston wins. The "Stevens collapse" is not a Tatum offense problem. It is a Brown-specific offensive collapse plus a team-wide defensive bleed in the same windows. Boston's overall prosperity PPP only drops slightly in losses; what changes is which failures convert into opponent runs.

---

Caveats

• 9 losses is a small sample for player-level inferences. The Brown shift (41 events) and Tatum behavior (39 events) are real numbers but the per-season splits become very thin.
• 2024-25 NYK series is the dominant loss contribution (4 of 9 prosperity-game losses). The Brown collapse leans heavily on that series specifically.
• Possession initiator is not in public PBP; "Tatum events" and "Brown events" are who finished or assisted, not who initiated.
• Free throw events are attempts, not trips.
• No shot-clock or defender distance.
• "Defensive bleed" is a public-data proxy: opp next-possession outcome within the same period. It does not isolate true transition vs half-court, and it does not measure shot quality of the opponent's response.
• Event-level PPP can hide leverage. A team can have strong prosperity PPP and still lose if the decisive failure happens outside the prosperity window (e.g. after the opponent first cuts the lead below +8).
• The 2025-26 series has only 8 prosperity events in losses (the Game 5 collapse). Single-series win/loss claims for that year are on tiny samples.
• "Lead-blown" definition uses a strict tied-or-led threshold. Games where the opponent cut to within 1 or 2 without ever tying are classified as lead-stressed, not blown. Boundary cases exist.