Celtics Ball Movement / Assisted Creation in Prosperity (2021-22 to 2025-26)

Generated 2026-05-11. Ball-movement layer on top of the existing prosperity / win-loss analysis. Separates true rim pressure from assisted creation so the prior "force_event" composite stops doing two jobs at once.

Coverage: 2,296 prosperity event rows (1,898 ender + 398 creator) across 81 Celtics playoff games, 5 postseasons (2021-22 through 2025-26).

Sources:

• data/celtics_all_years_tagged_events.csv
• data/celtics_prosperity_game_conversion.csv

Outputs:

• data/celtics_ball_movement_prosperity_events.csv (2,296 rows)
• data/celtics_ball_movement_win_loss_summary.csv (55 rows)
• data/celtics_ball_movement_lead_conversion_summary.csv (80 rows)
• data/celtics_ball_movement_tatum_brown_delta.csv (50 rows)

---

Definitions (this layer)

• Rim pressure event: ender event that is a rim attempt OR a free throw event. Excludes assisted-three creation (which was bundled into force_event in earlier reports). This is true rim/foul pressure from that player.
• Assisted made FG: ender row where the make has another Boston player credited as the assist (regex match on description).
• Assisted made three / assisted made two: subsets of the above.
• Assisted creation event: creator-role row. That is, the player was the assister on a Boston make.
• Assisted creation three / two: subsets of creator rows.
• Unassisted shot event: ender FGA (made or missed) that is not an assisted make. By construction, all missed FGAs and all unassisted makes qualify. This is a rough public-data proxy for self-created / isolation shots.
• Assisted make rate: assisted made FG / total made FG.
• Assisted three make rate: assisted made three / total made three.
• Rim pressure rate: rim pressure events / total ender events.
• Unassisted shot rate: unassisted shot events / total ender events.
• Unassisted three rate: unassisted three events / total ender events.
• Creator share (team): this group's assisted-creation events / team's total assisted-creation events.

All splits use the existing win/loss flag and lead-conversion category from celtics_prosperity_game_conversion.csv.

---

Sample sizes (prosperity ender events, all years)

By lead conversion category (team-wide, all years):

• lead_controlled: 1,624 events / 30 games (100% wins)
• lead_stressed: 317 events / 9 games (100% wins)
• lead_blown: 355 events / 17 games (47% wins)

Table A. Boston ball movement, ALL SEASONS, wins vs losses

Read:

1. Team assisted-make rate is essentially identical in wins and losses (63.53% vs 64.15%). When Boston makes a shot in prosperity, the same share of those makes are assisted in both contexts.
2. Team rim-pressure rate falls 10 pp in losses (39.56% to 29.56%). This is the largest single change.
3. Team unassisted-three rate rises 5.81 pp in losses (25.01% to 30.82%). More pull-up threes in games Boston loses.
4. Brown's rim-pressure rate falls 19.78 pp in losses (46.61% to 26.83%). His assisted-make rate is unchanged. He receives the same share of catches but stops attacking the rim with them.
5. Tatum's creator share of team rises in losses (24.18% to 38.24%). Other Celtics creator share falls (61.81% to 50.00%). Tatum picks up creation slack when the rest of the roster stops creating.

Table B. Lead-conversion split (ALL SEASONS)

Read: Ball movement is actually slightly BETTER in lead-blown games. Team assisted-make rate rises to 65.91% (vs 63.02% in lead-controlled). Other Celtics' assisted three make rate climbs to 94.44% (vs 87.05%). Brown's assisted-make rate climbs to 69.23% (vs 57.45%). The team moves the ball MORE in blown leads, not less. What it does not do is attack the rim more or convert at a higher rate per shot.

The lead-blown signal is not "Boston stops passing." It is "Boston keeps moving the ball into the same kind of looks, but the shots are slightly more outside and the rim attempts are no more frequent than in controlled leads."

Table C. 2024-25 NYK collapse (the dominant loss series)

Read on the worst conversion series in the five-year window:

• In NYK losses Tatum's unassisted-shot rate jumped to 75% (vs 56% in wins). He went into iso mode and his rim pressure dropped to 30%.
• Brown made only 2 prosperity field goals in NYK losses; both assisted (100% rate is on a 2-make sample). His rim pressure rate collapsed to 17.65%.
• The team unassisted-shot rate in NYK losses was 63%, the highest of any series-result split in the dataset. Boston went one-on-one in windows it should have been moving the ball.

Tatum/Brown win/loss deltas (all years, key numbers)

In losses, Tatum:

• Has slightly more of his makes assisted (the rest of his prosperity scoring is self-created).
• Has fewer of his made threes come off an assist (50% in losses, 62% in wins).
• Drops 8 pp in rim pressure but rises 9.37 pp in unassisted threes.
• Picks up 14 pp of team creator share — he is the one still trying to create when Boston is losing.

In losses, Brown:

• Has the same assisted-make rate (60%) and assisted-three-make rate (80%) as in wins. The ball is finding him.
• Loses 19.78 pp of rim pressure while everyone else loses less than 10 pp. He stops attacking the rim more sharply than any other group.
• His creator share goes slightly down (he creates less for teammates).

---

A. Does Boston's assisted make rate drop in prosperity losses?

No. It is essentially flat. Team assisted-make rate is 63.53% in wins and 64.15% in losses (delta +0.62 pp, statistically meaningless). The same applies to every group: Tatum +4.26 pp, Brown +0.34 pp, Other Celtics +2.62 pp. Boston's ball-movement-to-makes pipeline is intact in losses.

B. Does Boston's assisted three rate change in losses?

Slightly down. Team assisted-three-make rate falls from 83.11% in wins to 77.27% in losses (-5.84 pp). For Tatum the drop is sharper: 61.76% to 50.00% (only 6 made threes in losses, so noisy). For Brown it is essentially flat (82.76% to 80.00%). For Other Celtics it actually rises (87.65% to 90.91%). The team-level drop is mostly Tatum-driven: in losses, his made threes are more often pullup isolation threes rather than catch-and-shoot looks off a teammate.

C. Does the offense become more unassisted / self-created in losses?

Yes, moderately. Team unassisted-shot rate rises 4.45 pp in losses (51.52% to 55.97%), and team unassisted-three rate rises 5.81 pp (25.01% to 30.82%). The shift is consistent across player groups: Tatum +8.05 pp unassisted shot rate; Brown +5.07 pp; Other Celtics +2.70 pp. The biggest single shift is in the 2024-25 NYK losses where team unassisted-shot rate hit 63%.

D. Does Tatum create fewer assisted shots in losses?

No, the opposite. Tatum's creator-role events are 88 in wins (518 total events) and 13 in losses (52 total events). On a per-event basis: 17.0% of Tatum's prosperity rows in wins are creator rows; 25.0% in losses. His share of team creation rises from 24.18% in wins to 38.24% in losses. When Boston loses prosperity windows, Tatum is doing MORE creation, not less. Sample is small (13 creator events) but the pattern is consistent across seasons (in 2024-25 specifically, Tatum's creator share in losses is 50%).

E. Does Brown create fewer assisted shots in losses?

Yes, dramatically. Brown's creator-role events are 51 in wins (in 405 total events) and only 4 in losses (in 45 total events). That is 12.6% of his prosperity rows being creator events in wins, dropping to 8.9% in losses. His team creator share drops from 14.01% to 11.76%. Brown's role as a connecting passer in prosperity windows nearly vanishes when Boston is losing.

F. Do Other Celtics become more dependent on assisted threes?

Yes, slightly. Other Celtics' assisted-three-make rate is already high in wins (87.65%); in losses it rises to 90.91%. Their unassisted-three rate also rises from 29.11% to 34.18%. So Other Celtics shoot more threes in losses, and a higher share of the made ones come off an assist (consistent with their catch-and-shoot role), but more of their attempts (made or missed) are pullups they have to create themselves. The interpretation is that the team is asking role players to shoot more threes, and the made ones still come off the catch but the attempted pool drifts more toward pullups.

G. Does ball movement degrade in lead-blown games?

No. Ball movement is slightly better in lead-blown games than in lead-controlled games. Team assisted-make rate in lead-blown windows is 65.91% vs 63.02% in lead-controlled (+2.89 pp). Other Celtics' assisted-three-make rate rises from 87.05% in controlled to 94.44% in blown (+7.39 pp). Brown's assisted-make rate jumps from 57.45% to 69.23% (+11.78 pp). The "Boston stopped moving the ball when the lead got tight" framing is not supported by the assisted-make data.

What does change in lead-blown games is the same pattern as in losses: rim pressure is similar, unassisted-shot rate is similar, but the attempts skew slightly more toward outside threes and the finishing-at-the-rim conversion declines. Ball movement is fine; the shot diet and the finish rate are the problem.

H. Does this explain the defensive bleed after missed threes?

Partly. From the prior win/loss report: opponent next-possession PPP after a Boston missed three is 1.085 in wins, 1.444 in losses. This ball-movement layer adds context:

• In losses, Boston attempts more unassisted threes (Tatum +9.37 pp, Brown +6.03 pp, team +5.81 pp).
• Unassisted three attempts are mechanically more likely to be contested pull-ups than catch-and-shoot looks (public PBP cannot prove this, but it is the standard correlate).
• A long-rebound miss from a pull-up three is the canonical "let them back in" transition opportunity.

So the ball-movement layer is consistent with the defensive-bleed finding: in losses, Boston takes more self-created threes, and missed self-created threes produce worse defensive consequences. The mechanism is plausible from public data but not provable. It "complicates" the story rather than "proves" it: assist data alone does not separate a contested pull-up from an open one.

I. What can public PBP prove?

• Whether a made shot was assisted or unassisted.
• Who the assister was, by name.
• The basic shot type (rim / midrange / three) and whether the shot went in.
• That the team-level assisted-make rate does not drop in losses.
• That Brown's rim-attempt rate drops disproportionately in losses.
• That Tatum's creator share of team rises in losses while Brown's falls.

J. What can public PBP not prove?

• Number of passes per possession.
• Drive-and-kick chains, paint touches, dribble counts.
• Touch time before each shot.
• Defender distance and shot openness ("wide open" vs "tightly contested").
• Whether an unassisted three was a pull-up off the dribble or a catch-pump-and-shoot from a swing pass.
• Whether ball-screen action was used.
• Whether the next-opponent-possession was true transition or half- court.

Public PBP gives us the assist credit. It does not give us the mechanic that produced the assist or its absence. Calling an unassisted shot "self-created" is shorthand; the same row could be a swing-pass-pump-fake-step-back or a true isolation pull-up, and PBP cannot distinguish them.

---

Updated bottom-line characterization

The simple "Boston stopped moving the ball" reading of the Stevens collapse narrative is not supported by the assist data in this five-year sample. Team assisted-make rate is unchanged in losses. The ball-movement pipeline keeps producing assisted makes at the same rate.

What changes in losses is two specific things:

1. Rim pressure drops 10 pp team-wide, and 20 pp for Brown specifically.
2. Unassisted-three rate rises ~6 pp team-wide, driven by Tatum (+9 pp) and Brown (+6 pp).

In lead-blown games specifically, the ball moves slightly more than in lead-controlled games but is converted into similar-quality shots (no extra rim attempts). That is consistent with the prior finding that the loss problem is not possession-by-possession efficiency but rather conversion + defensive bleed.

---

Message-board paragraph

Ball-movement data complicates the "Boston stopped moving the ball with the lead" version of the Stevens narrative. Across five postseasons and 1,898 prosperity ender events, Boston's team assisted- make rate in losses is 64.15%, essentially identical to its 63.53% in wins. In lead-blown games the assisted-make rate is actually slightly higher (65.91%). Other Celtics' assisted-three-make rate climbs to 90.91% in losses and 94.44% in lead-blown games. The pass-to-make pipeline is not what breaks. What does break is rim pressure (-10 pp team-wide in losses, -20 pp for Brown) and shot diet (unassisted-three rate +6 pp). Brown's signal is the cleanest: in losses he stops attacking the rim, his creator events drop from 51 in wins to 4 in losses, and his assisted-make rate is unchanged, meaning the ball is finding him but he is finishing fewer rim attempts. Tatum's signal is the opposite: in losses he picks up team creator share (24% to 38%) but his own attempts become more unassisted. The ball is still moving; the shots are just worse, and Brown is the player whose role changes most when the team is losing.

---

Caveats

• Public PBP only. Assist credit is the only ball-movement signal available. It does not say how many passes preceded the shot, whether it was a drive-and-kick, whether the shot was openly defended or pressured.
• Brown's loss sample is 41 ender events (and only 4 creator events). Per-season splits are very thin.
• The 2024-25 NYK series accounts for 87 of 159 team-wide prosperity events in losses; signal is heavily weighted to that one series.
• "Unassisted shot event" includes both missed FGAs and unassisted makes. Missed FGAs are unassigned by definition, so the unassisted- shot category is partially driven by which shots failed rather than how the play started.
• Free throw events are attempts, not trips.
• Creator share is computed within Boston only. It does not reflect absolute creation rate.
• This layer does not separate end-of-period heaves, garbage-time late-clock attempts, or fouled-game free-throw runs.