Celtics Action-Based 3PA Generation Analysis

Generated 2026-05-19. Direct answer to the RealGM question: which Celtics offensive actions actually produce the most 3PA?

This is the companion analysis to reports/celtics_queta_roll_vs_3pa_tradeoff.md (which answered the PPP-equivalence question) and reports/celtics_pnr_dho_action_analysis_REPAIRED.md (which answered the action-mix question).

---

1. Executive verdict

Spotup produces by far the most 3PA in Boston's offense. It is worth roughly 7 inferred-3PM per game in the regular season (55-60% of all team 3PM) and held that role into the playoffs at 6.8/game. After Spotup, the 3PA-generation ladder is:

Spotup            (~7.0 inferred 3PM/g  RS, 6.8 PO)  - dominant
Transition        (~2.7 inferred 3PM/g  RS, 3.0 PO)  - second
PRBallHandler     (~2.4 inferred 3PM/g  RS, 2.3 PO)  - pull-up 3s, mostly
OffScreen         (~1.0 / 0.8)
Handoff           (~0.8 / 0.2)
PRRollman         (~0.7 / 1.2)                       - "pop" rollers
Isolation         (~0.7 / 1.1)
Misc              (~0.2 / 0.1)
Cut, OffRebound, Postup ≈ 0

So the poster's general claim - "coaching staffs prize actions that produce 3PA" - applies to Spotup, Transition, and PRBallHandler, in that order. Cut, OffRebound, and Postup produce essentially zero 3PA, so the implicit critique that "Queta on the roll doesn't generate 3PA" is technically correct: PRRollman generated only ~0.7- 1.2 3PM/g, and most of that was rollers popping rather than truly rolling.

But three counterweights make the objection only partially valid:

1. Synergy cannot credit roll gravity to the roller. A PNR that resolves with a kickout-three is attributed to the receiver's Spotup row, not to the roller's PRRollman row. So the roll-man inferred 3PM (0.7-1.2) is the roller's own pop 3PAs only - not the kickout-threes the roll forced.
2. Boston's actual playoff catch-and-shoot 3s went in at 31.2%. The "we shift to Spotup so 3PA goes up" logic only works if those Spotup 3s convert. In the 2025-26 playoff series, they didn't.
3. PPP comparison is unchanged. Queta playoff roll = 1.786 PPP = equivalent to a 59.5% three. No realistic NBA three matches that.

The poster's objection therefore survives as "coaches prioritize 3PA volume," but does not survive as "the actual 2025-26 Boston playoff usage was optimal." Even on the 3PA-volume logic, Boston's spot-up 3s weren't going in.

2. Exact question being tested

"Which Celtics offensive actions actually produce the most 3PA?"

And the underlying objection: "If Queta on the roll is returning good value but not a lot of 3-point attempts, that doesn't mean it's what the coaching staff wants."

3. Data availability: can public data show 3PA by action?

Direct 3PA-by-Synergy-play-type: NO. Synergy aggregate headers contain FGM, FGA, FGMX, FG_PCT, EFG_PCT but NOT FG3M, FG3A, FG3_PCT. Verified across all 22 Synergy player and team JSON files.

Inferred 3PM-by-Synergy-play-type: YES. Mathematically exact via the eFG identity: eFG = (FGM + 0.5 × 3PM) / FGA → 3PM = 2 × (eFG × FGA - FGM)

Estimated 3PA-by-Synergy-play-type: YES, but requires 3P% assumption. A column has been added for 3PA estimated at 33%, 36%, 39%, 42%.

Direct 3PA by shot-type (catch-and-shoot, pull-up): YES. Pulled via nba_api.stats.endpoints.TeamDashPtShots for Boston 2025-26 regular season and playoffs. Returns FG3M and FG3A directly. This is not a Synergy-action breakdown but is the closest direct 3PA-creation proxy available in public data.

Pass-out / kickout attribution: NO direct field. Synergy classifies pass-out outcomes under the receiver's play type. The PNR ball-handler kickout that becomes a corner three is recorded in the shooter's Spotup row, not in the handler's PRBallHandler row.

4. Direct 3PA results: tracking shot-type breakdown

From data/raw/tracking/teamdashptshots_BOS_2025-26_*.json (NBA Stats TeamDashPtShots endpoint, via nba_api):

Boston Regular Season 2025-26

Read: 64.5% of Boston's regular-season 3PA were catch-and-shoot (2,231 of 3,459). The remaining 35.5% were pull-ups or off-the-dribble. Catch-and-shoot 3P% was 38.1% (good).

Boston Playoffs 2025-26 (7-game PHI series)

Read: catch-and-shoot 3s collapsed to 31.2% in the playoffs. Pull-up 3s held up at 38.3%. The "feed Spotup to generate threes" logic broke down because the catch-and-shoot conversion broke down.

Dribble breakdown (RS)

65.1% of all 3PA came from 0-dribble (catch). Pure off-ball / pass-out attempts.

Touch-time breakdown (RS)

70.5% of 3PA came from <2-second touches - again, pure spot-up / kickout / pass-out attempts.

5. Inferred 3PM by Synergy action

From data/celtics_action_3pa_direct_or_inferred.csv. Inferred via the eFG identity. Estimated 3PA assumes a 36% conversion (close to the team's actual RS rate).

Boston team, regular season

Note: action shares sum >100% because attribution overlaps (a Spotup 3PA caused by a PNR-handler kickout still lives in Spotup). The "share of team 3PM" column treats each action's inferred 3PM as a share of the sum of all action inferences, which is the right denominator for relative ranking but not for absolute total.

Boston team, playoffs

PRRollman 3PM went UP from 0.74 (RS) to 1.23 (playoffs). That was Vučević popping instead of rolling, plus Garza pop-rolls earlier in the season. Queta himself attempted 3 threes total across the 7-game series (PBP).

6. Tracking proxy results

The TeamDashPtShots tracking endpoint adds an independent view:

• 64.5% of Boston RS 3PA were catch-and-shoot, 35.5% pull-up.
• 70.5% came from <2-second touches (quick release, pure spot-up).
• 65.1% from 0-dribble (pure catch).

These shares broadly match the Synergy inference: Spotup is the single biggest 3PA generator, but PRBallHandler (pull-up) plus the chain of pass-outs (still attributed to Spotup) account for the rest.

Independent confirmation: the 3,459 total RS 3PA divided 9 ways across Synergy actions returns roughly the same Spotup-dominance shape.

7. PBP context (descriptive proxy)

From data/celtics_pbp_three_context.csv. The 7 Boston-PHI playoff games yielded ~195 Boston 3PA events from PBP description matching (undercounts the official 320 because the PBP description string is not 100% complete - mostly missing the ones logged only via subType). Of the captured events:

• 39.5% made
• 30.8% had a description-detectable assist
• Shooter distribution: Tatum 42, Pritchard 36, White 23, Hauser 19, Brown 19, Vučević 18, Scheierman 14, Garza 9, Walsh 9, Queta 3, Harper 2, Shulga 1.

Queta took 3 threes in the entire playoff series. He is not a 3PA-generation source. Vučević's 18 attempts are the pop-roll contribution; Hauser/Scheierman are the spot-up volume.

The PBP cache description does NOT label PNR/DHO - that finding holds from the prior local-reconstruction lane. So PBP cannot tell us "this 3PA was off a Queta-rolling kickout" directly. The description field is too sparse.

8. What actions most likely generated 3PA

Direct evidence (inferred 3PM):

1. Spotup ~7.0/g (RS), 6.8/g (PO) - dominant.
2. Transition ~2.7/g (RS), 3.0/g (PO) - early-clock corner kicks and trailing 3s.
3. PRBallHandler ~2.4/g (RS), 2.3/g (PO) - pull-up 3s by the handler.
4. OffScreen ~1.0/g - Hauser specials.
5. Handoff ~0.8/g - DHO-into-three for the receiver.
6. PRRollman ~0.7-1.2/g - pop-rollers (Vučević, Garza), not Queta.
7. Isolation ~0.7-1.1/g - mostly Brown/Pritchard step-back 3s.
8. Cut, OffRebound, Postup ≈ 0.

Indirect evidence (Synergy attribution problem):

• A PNR ball-handler kickout-three is filed under Spotup.
• A PNR roll-man pop-three IS filed under PRRollman.
• A PNR roll-man kickout (roll, defender helps, ball gets swung, shooter fires) is filed under Spotup for the shooter, not PRRollman for the roller.

So the SAME action (a Boston PNR) shows up in multiple Synergy rows depending on how it resolved. The "PRBallHandler generates threes" read is strictly about handler pull-up 3s, NOT about the kickout-3s the handler created.

9. What actions generated the highest PPP

From the companion analyses, regular-season Boston team PPP by action:

The highest-PPP actions (Cut, OffRebound) are exactly the actions that produce zero 3PA. The 3PA-rich actions (Spotup, Transition) have above-average PPP but not the highest. The cleanest PPP-vs- 3PA tradeoff actions are Spotup (1.103 PPP / 7.0 3PM-per-g) and Transition (1.177 / 2.7). Those are the actions a coach focused on 3PA volume should run.

The under-exploited PPP-rich actions (Cut, OffRebound) require rim attackers and rebounders to create them, not three-point shooters. That is the exact handle on the Queta-roll argument: his roll PPP of 1.786 lives in this PPP-rich, 3PA-poor category, but it doesn't compete with Spotup - it creates Spotup chances downstream.

10. Queta roll vs 3PA-generation objection, revisited

The poster's claim: "Queta on the roll is returning good value but not a lot of 3-point attempts, [so it] doesn't mean it's what the coaching staff wants."

The data:

• PRRollman directly generates ~0.7-1.2 inferred-3PM/g. Not zero, but small. Most of that is the pop-rollers (Vučević, Garza), not Queta.
• Queta attempted 3 total threes across the 7-game PHI series. He doesn't shoot.
• The PNR roll-man branch returns 1.196 PPP team-level (1.786 for Queta). That is more PPP than every Synergy action except Cut.
• If the coach prioritizes 3PA, that 3PA still has to be GENERATED. Most of it comes from Spotup, which in turn comes from pass-outs created by drives, PNR-handler kickouts, and roll-man gravity. The Synergy attribution does not see the chain.

So the literal answer is: "Yes, PNR Roll Man does not generate much 3PA directly. The actions that DO generate 3PA are Spotup, Transition, and PRBallHandler." But that does not mean the roll branch should be under-deployed:

1. Boston's playoff Spotup 3s went in at 31.2%. So adding 3PA volume wouldn't have been worth much against this defense.
2. Boston already ran Spotup 27.0 times per game in the playoffs (1st of 16 playoff teams by share). They were not under-deploying the 3PA-generation action.
3. The under-deployed action was PRRollman with Queta, which is the PPP-richest finishing option Boston had access to.
4. Roll-man gravity creates Spotup chances. Synergy doesn't credit the roller, but it shows up downstream as someone else's Spotup 3PA.

11. What the data proves

• Spotup generates the most 3PA in Boston's offense (~55% of all team 3PM via inference; 64.5% of all 3PA were catch-and-shoot via tracking).
• Transition is second, PRBallHandler third (mostly handler pull-up 3s).
• Cut, OffRebound, and Postup directly generate essentially zero 3PA.
• PRRollman generates 0.7-1.2 3PM/g, mostly from pop-rollers, NOT Queta.
• Boston's playoff catch-and-shoot 3s converted at 31.2% - below league-average. So the 3PA-volume strategy didn't pay off in the series.
• Queta's playoff roll PPP of 1.786 = 59.5% equivalent three. No realistic NBA three matches that.

12. What remains unproven

• The kickout chain. Synergy cannot attribute the Spotup 3PA back to the action that created it. We cannot prove "X% of Boston's Spotup 3PA originated from a PNR roller forcing help."
• Queta-specific roll gravity. No public-data way to isolate the Boston threes that wouldn't have existed without his screen.
• Coaching intent. The data describes outcomes, not preferences. Joe Mazzulla may favor Spotup-heavy schemes for portfolio variance reasons that aren't captured in single-series efficiency.
• Volume scaling. Doubling Queta's roll-man possessions doesn't guarantee 1.786 PPP holds. Defenses adjust. The marginal next-rep PPP is likely lower than the average current-rep PPP.

13. RealGM-ready response

Fair point that coaching staffs design offense to maximize 3PA volume, and Synergy's PRRollman row directly generates only about 0.7-1.2 inferred 3PM per game for Boston - mostly from Vučević and Garza popping, not Queta rolling. Spotup is the dominant 3PA generator at ~7 inferred 3PM per game (about 55% of all team 3PM), and Boston already ran it 27 possessions per game in the playoffs, first of 16 playoff teams by share. So they weren't under-deploying the 3PA-generation action - they were running it aggressively. The problem is the actual conversion: Boston's playoff catch-and-shoot 3s went in at 31.2% (205 attempts, 64 makes). At that rate, each Spotup 3 returns about 0.94 PPP. Queta's roll possession returned 1.786 PPP. So even on the volume logic, shifting one possession from PNR-handler (which returned 0.856 PPP) to a Queta roll (1.786) only "loses" 3PA generation if you assume Spotup chances would have come out the other way - and the kickout from a roll IS what creates Spotup chances. Synergy just credits the shooter, not the screener. The honest read: Boston had a 3PA- heavy offense and the catch-and-shoot 3s went cold against PHI. The cure isn't to run even more catch-and-shoot 3s - it's to bend the defense first with rim pressure (drives, Queta rolls, cuts), and let the resulting threes come at higher quality.

---

Sources

• data/raw/action_types/nba_api/synergy_2025_26_*_offensive_PerGame.json (44 files, all 11 play types × 2 season types × team+player)
• data/raw/tracking/teamdashptshots_BOS_2025-26_Regular_Season.json
• data/raw/tracking/teamdashptshots_BOS_2025-26_Playoffs.json
• data/pbp_cache_v3/0042500111.csv through 0042500117.csv (7 PHI series games)

Deliverables produced

• scripts/analyze_action_3pa_generation.py
• data/celtics_action_3pa_direct_or_inferred.csv (148 rows: Phase A)
• data/celtics_tracking_3pa_creation_proxies.csv (52 rows: Phase B)
• data/celtics_pbp_three_context.csv (195 rows: Phase C)
• data/celtics_action_3pa_fetch_log.csv (3 fetch-log rows)

Limits, declared

• 3PA-by-Synergy-action is INFERRED, not observed. The eFG identity gives exact 3PM but 3PA requires a 3P% assumption (33/36/39/42% estimates provided).
• Pass-out / kickout attribution is invisible in Synergy.
• PBP descriptions do not label Synergy actions, so PBP context is proxy-only (assisted vs unassisted; prior 3-event window).
• Tracking shot-type categories (Catch-and-Shoot, Pull-up) are NOT Synergy-action categories; they are independent groupings.
• 7-game playoff sample. Per-game variance is real.
• Roll gravity is a basketball claim, not a measurable in this data.