Celtics Prosperity Possessions: Tatum vs Brown

Generated 2026-05-11. Public data research package. Compiled from NBA stats PlayByPlayV3 for the 2025-26 first-round playoff series vs Philadelphia (BOS lost 4-3). Game ids 0042500111 through 0042500117.

---

Purpose

Test whether Jayson Tatum or Jaylen Brown is more associated with the Celtics losing force in the "prosperity" possessions Brad Stevens described at his end-of-season press conference: Boston gets ahead, slows down, tries to protect the lead, lets the opponent back in, struggles to generate quality first-shot offense, lacks rim pressure.

This is a comparison, not a verdict on either player's career value.

Data sources used

• nba_api.stats.endpoints.playbyplayv3 for full play-by-play of all 7 games in the 2025-26 BOS-PHI first round.
• nba_api.stats.endpoints.leaguegamefinder to enumerate Celtics playoff games and identify the most recent BOS-vs-PHI series.
• No private tracking data. No Second Spectrum. No defender-distance. No shot-clock data. No tracking-derived drive counts.

Discovery file: data/celtics_playoff_games_discovered.csv (81 rows, covering 2021-22 through 2025-26).

Exact definitions

Prosperity possession in this study:

• Boston playoff event.
• Period >= 3 (second half, includes OT if applicable).
• Boston's score margin BEFORE the event was >= 8.
• The event is a Boston shot attempt, free throw, or turnover.
• The player on the event was Jayson Tatum or Jaylen Brown.

Margin is computed by forward-filling scoreHome/scoreAway and then shifting one row (so the value is the margin entering the event, not the margin after the event).

Tagged role:

• ender the player whose action ended the possession (FGA, FT, or TO).
• creator the player whose assist set up a teammate's made shot.

When Tatum assists Brown's made three, both rows are written: one for Brown (ender) and one for Tatum (creator). The summary metrics group by player, so this counts as one Tatum possession and one Brown possession. The PPP statistic computed with all rows is therefore inflated by this double-credit; the ender-only PPP is the cleaner comparison and is reported below.

Force event proxy: is_rim_attempt OR is_free_throw_event OR is_assisted_three_created. A public-data approximation for whether the possession produced pressure instead of settling. It is not a perfect metric.

What public PBP can prove

• Who took the shot.
• Whether the shot was at the rim, midrange, or three (from actionType, subType, and shotValue).
• Whether it was made or missed.
• Who turned the ball over.
• Who got the assist (parsed from description).
• Game-clock timing.
• Score margin context.

What public PBP cannot prove

• The true possession initiator. PBP only records who finished or assisted. A possession that Tatum brought up and gave to Brown for a contested midrange will appear in the data only as a Brown FGA.
• Shot quality (defender distance, openness, contestedness).
• Shot-clock state at the moment of the shot. The is_late_clock_proxy flag in this dataset is a weak signal that only triggers in the last five seconds of the period clock, not the shot clock.
• Drive vs catch-and-shoot vs pull-up classification beyond what subType encodes ("Pullup Jump Shot", "Driving Layup", etc).
• Whether a turnover was forced by the defense vs unforced.

These limitations are large. Treat the comparison as a directional read, not a settled finding.

Sample size warning

The 2025-26 BOS-PHI series produced only 67 tagged prosperity events across 4 games (the other 3 games yielded zero or one event because Boston did not build an 8+ second-half lead in those games). This is a small sample. Below the table I split the data multiple ways to test whether the headline numbers survive.

Critically: in three of the four Boston losses (Games 2, 6, and 7), there were no prosperity possessions to study because Boston never built the lead in the first place. The Stevens "prosperity" pattern only directly applies to Game 5, where Boston held a second-half lead and surrendered it at home in a -16 loss. That game produced only 5 events. Read the single-game breakdown for Game 5 carefully and weight it accordingly.

Summary table (all tagged rows)

The all-rows PPP is inflated by the creator double-credit. The cleaner ender-only view, where each row represents a possession the player actually finished, is below.

Ender-only view (no creator double-count)

Ender-only PPP is essentially identical. The shape of the possessions is very different.

Tatum vs Brown comparison

Where Tatum is worse than Brown in prosperity (ender-only):

• Turnover rate: 18.5% vs 7.1% (Tatum nearly 3x). This is the single most damaging number for the Tatum thesis. Five of his 27 prosperity possessions ended in a Boston giveaway.
• Force events that are FT-driven are lower (FT events count: 6 vs 7).

Where Brown is worse than Tatum in prosperity (ender-only):

• Midrange rate: 35.7% vs 7.4% (Brown nearly 5x). Brown takes the classic "settle" shot far more often. Six of his 10 midrange attempts missed (60%). The Stevens "lacks rim pressure" complaint shows up in Brown's shot diet far more cleanly than Tatum's.
• Three-point rate: 14.3% vs 33.3%. Tatum is taking more threes; in a protect-the-lead context, that can be either good (efficient spacing) or bad (settling from deeper). Three-point efficiency was 56% (5 of 9) for Tatum and 75% (3 of 4) for Brown, so the volume difference is the variable here.

Where they are essentially tied:

• Ender PPP: 1.074 vs 1.071.
• Rim attempt rate: 18.5% vs 17.9%.
• Force events per 100: 45.95 vs 46.67 (all rows); 40.7% vs 42.9% (ender-only).

Where Tatum is clearly better than Brown:

• Creation. Tatum had 6 assisted-three-creations and 4 assisted-two creations in prosperity windows. Brown had 2 and 0. Tatum is the higher- leverage decision node when Boston has the lead.

Game 5 (2026-04-28, BOS lost 97-113 at home)

This is the one true Stevens-pattern game in the series: Boston led by 8+ in the second half and lost. Only 5 tagged prosperity events:

Both players were ineffective in this window. Tatum: 0-1 with a turnover. Brown: 1-3 with a pullup midrange miss and a missed driving layup. Five events is not enough to assign blame; what it shows is that neither star provided force in the only true prosperity-collapse game of the series.

Evidence supporting the Tatum-as-common-thread thesis

• Tatum's prosperity-window turnover rate (18.5% ender-only) is far higher than Brown's. Five giveaways while protecting a lead is real cost.
• In Game 5, the one collapse game, the most visible Boston giveaway is a Tatum backcourt turnover with Boston up 11 in the third quarter. That is exactly the "let the opponent back in" mechanism Stevens described.
• Tatum's three-attempt rate is higher in prosperity. If the read is that Boston "settles" with leads, Tatum's diet contributes via lower-leverage three-point pulls rather than rim attacks.

Evidence supporting the Brown counter-thesis

• Brown's midrange rate (35.7%) is the single most settle-prone number in the dataset. Stevens' "lacks rim pressure" complaint maps onto Brown's shot chart, not Tatum's: when Brown shoots in prosperity windows, it is most often a pullup or fadeaway two-pointer.
• Brown's midrange efficiency in prosperity windows is poor: 4 of 10 made (40%). That is the kind of shot that ends possessions without producing fouls or offensive rebounds.
• Brown's creation rate is one third of Tatum's (2 assisted threes vs 6). When Brown ends a prosperity possession, it is almost always with a shot, not a kickout. That removes a kind of force from the offense.

Net verdict

Mixed and inconclusive on the simple thesis, leaning slightly against the Brown counter-thesis on shot diet.

The data does not support a clean "Tatum is the problem" read. Ender-only PPP is identical. Force-event rate is identical. Tatum's higher turnover rate is real and damaging, but it is offset by his much higher creation rate (six assisted threes is a lot in a 67-event sample).

The data also does not support a clean "Brown is the problem" read. His midrange-heavy diet is the most clearly Stevens-coded behavior in the dataset, but he turns it over less and finishes at the rim slightly more often than Tatum does.

If forced to pick a single sentence: in prosperity windows, Tatum is the higher-creation higher-turnover player and Brown is the higher-settle lower-creation player. Neither pattern is the obvious culprit. The bigger finding is structural: in three of the four Boston losses, there was no prosperity sequence to analyze because Boston never built the lead. The Stevens narrative may overweight one game.

Caveats

• 67 events. 4 games. One true prosperity-collapse game (Game 5) with only 5 events in it.
• Possession initiator is not in public PBP. Tatum-led and Brown-led are measured by who ended or who assisted, not who brought it up.
• Late-clock proxy in this dataset only fires in the last 5 seconds of the period clock. It is not a shot-clock signal.
• Lead threshold of 8 is the user-specified definition. A lower threshold (say 5) would include more events from Games 5, 6, 7 where leads were smaller before being surrendered. Recommend re-running with --lead-threshold 5 for a wider denominator.
• Free-throw "events" are individual FT attempts, not trips. Two FTs on one trip count as two events.
• This is a single playoff series. Drawing a broader pattern requires pulling 2023-24, 2024-25, and prior playoff runs through the same pipeline (the script is parameterized to do so).

Next steps

1. Re-run the same pipeline on the 2024-25 second-round series vs New York (game ids 0042400211-0042400216) and the 2023-24 finals run, both already in data/celtics_playoff_games_discovered.csv, to see whether the Tatum-TO and Brown-midrange patterns repeat or were one-series artifacts.
2. Re-run with --lead-threshold 5 to capture the moments where Boston had a smaller cushion but Stevens' described "slowdown" still applied.
3. Add a possession-start filter (look at the event before the tagged event: was the prior event a Boston defensive rebound or made FG by the opponent?) to better isolate true initiating possessions.
4. Pull synergyplaytypes data (still public via stats.nba.com) for isolation efficiency in playoff context as a complementary signal.
5. Layer in second-chance points: if Boston turned a Brown midrange miss into an offensive rebound and a Tatum putback, current pipeline scores that as Brown miss + Tatum rim event. Track the chain explicitly.

Message-board paragraph

In Boston's 2025-26 first-round prosperity windows, Tatum and Brown finished possessions at the same scoring rate (1.07 points per possession). The difference was the shape. Tatum's prosperity possessions were higher-creation and higher-turnover six assisted threes for teammates, but five live-ball giveaways including a backcourt turnover in Game 5 with Boston up 11. Brown's were lower-creation and lower-turnover, but settle-heavy: 36% of Brown's prosperity possessions ended in a midrange jumper, and only 40% of those midrange jumpers went in. Neither is exonerated and neither is the obvious culprit. Tatum is closer to Stevens' "let them back in" pattern through turnovers; Brown is closer to Stevens' "lacks rim pressure" pattern through midrange settling. The honest read is that the Celtics' prosperity problem is structural, not one star's identity in three of Boston's four losses in the series, there was no prosperity sequence to study because Boston never built the lead long enough to surrender it.