Methodology

Bracket score

Each team gets a composite bracket score on a 0–1 scale. The score determines at-large selection order, national-seed order, and margin-from-cutline.

The 15-game Q1+Q2 prior is hand-picked at roughly half a season of plausible high-quad opportunities so that an 8-game sample carries ~65% prior weight while a 30-game sample carries ~33%. This keeps a 5–3 vs Q1/Q2 from outscoring an 18–12 over a full season.

The marquee-series bonus was added May 2026 to address a structural gap: the committee weighs series wins over national-seed contenders much more heavily than a single Q1 win over an RPI #25 team, and the composite couldn’t see that distinction. Tuned on 2022–2025 backtest (top-8 agreement +3 of 32, no fold regresses; host agreement flat at 56 of 64). Honest disclaimer: not yet validated on held-out years; we’ll know much more after 2026 Selection Monday.

What we optimize for

Not every part of the bracket carries equal weight in our calibration work. In rough priority order:

1. Top 8 national seeds — the headline of selection night. Marquee-series bonus targets this metric directly.
2. Top 16 hosts — who gets to host a regional. This is what most users open the site to check.
3. 2-seed assignments — controls who travels to which regional and is the most-discussed early-round matchup.
4. The 64-team field — bubble inclusion is the hardest problem but also the part of the bracket the public scrutinizes least.

RPI formula

• WP — Location-weighted win pct: road wins count 1.3×, road losses 0.7×, home wins 0.7×, home losses 1.3×, neutral 1.0×
• OWP — Average of each opponent’s raw (unweighted) D-I winning percentage, head-to-head games excluded
• OOWP — Average of each opponent’s OWP

Quadrant thresholds

Games are classified by opponent RPI rank and game location:

Ties

Ties show up rarely in college baseball (curfew or conference decision) but they do exist. Win-pct components in the bracket score treat each tie as half a win and half a loss — i.e. (W + 0.5×T) / (W + L + T). The official RPI formula does not currently include a tie convention; we leave that untouched and only fold ties into the composite components above.

Auto-qualifiers

Each conference contributes one automatic bid. The selection rule runs in two passes:

1. Recorded champion. If an admin has entered the conference tournament champion, that team takes the AQ — even for a conference flagged as ineligible by team count.
2. Fallback. Best tie-aware conference win pct, with ties broken by bracket score, then RPI rank, then team id. Conferences with fewer than 6 D1 baseball teams are skipped.

The fallback is a stand-in until the actual conference tournament finishes — it makes the bracket model reasonable in March and April but it is not the truth once tournaments begin.

Field of 64

Up to 30 automatic bids fill first. The remaining at-large slots go to the highest bracket scores. The four at-larges with the smallest margin above the cutline are the Last Four In; the four highest scores below it are the First Four Out.

The cutline itself is the midpoint between the weakest in-field at-large and the strongest team out, so margin-from-cutline is symmetric: +0.05 inside and −0.05 outside mean the same distance on either side.

Bid-status badges

• AQ — Conference auto-qualifier (recorded champion or projected best conference record).
• Lock — At-large with margin-from-cutline of at least 0.04 — roughly five-plus ranks of separation from the bubble.
• Safe — At-large in the field but inside the lock margin.
• Bubble — One of the Last Four In.
• 1st Out — One of the First Four Out.

Host eligibility

The 16 regional hosts are the highest-scoring teams that pass at least one of these gates and have a hostable home venue:

• Gate A. RPI rank ≤ 25 and bracket-score rank ≤ 16 — the standard host profile.
• Gate B. Bracket-score rank ≤ 12 regardless of RPI — a hot team that out-resumes its RPI.
• Gate C. Conference auto-qualifier with RPI ≤ 20 — a champion whose composite might be borderline but whose RPI is solid.

If fewer than 16 teams clear the gate, it is relaxed and a Host gate relaxed warning is surfaced. If still fewer than 16, non-hostable venues are filled in last — the bracket always produces 16 regionals.

S-curve placement

Each non-host seed line is placed against the host pool in standard committee S-curve order:

• 2-seed line. Best 2 (rank 17) goes to region 16 (weakest 1). Formula: 16 − indexInLine.
• 3-seed line. Snake reverses — best 3 (rank 33) goes to region 1. Formula: indexInLine + 1.
• 4-seed line. Snake reverses again — best 4 (rank 49) goes to region 16. Formula: 16 − indexInLine.

When the natural top-16 by score also pass the host gate, each regional’s seed indices sum to a constant — standard committee S-curve balance. When canHost or gate fallback shifts the host pool, the seed lines stay monotone but per-region balance can drift.

Same-conference avoidance

Two teams from the same conference may not meet in the first round of a regional. Only the 1-vs-4 and 2-vs-3 pairings are checked (3 plays 2 first, 4 plays 1 first). When a conflict appears we look for a near-rank swap partner within 4 ranks whose own region is conflict-free for our team. If no such partner exists, the rule is relaxed and a Same-conference relaxation warning surfaces; that’s preferred over breaking the seed line.

Calibration notes

• The composite weights were retuned in May 2026 against a 2023–2025 backtest of NCAA committee selections. RPI (35%), SOS (15%), and Q1+Q2 (20%) anchor the score; conference record, bad-loss penalty, and road record each take 10%. The 15-game Q1+Q2 Bayesian prior and the 0.04 lock margin are hand-picked. Expect further retuning as more seasons of post-season data accumulate.
• Auto-qualifier fallback (best conference record) is a stand-in during March/April; once a conference tournament champion is entered via the admin endpoint, that team takes the AQ.
• Bracket score is on a 0–1 scale but small differences are meaningful: real bubbles cluster within ~0.05–0.10 of the cutline.

FAQ