Team outcomes desk

Up: 2026-27 Fantasy Basketball Encyclopedia Related: 2026-27 NBA team outcomes · Projection methodology desk · Schedule and streaming desk · Official news and roster desk

Scope and cutoff

• Research cutoff: 2026-07-18, America/New_York.
• “Fact” below means a dated source observation or a documented method. A forecast, market line, or modeled probability is explicitly labeled as such.
• Team outcomes covered: regular-season wins, conference seed, play-in/playoff probability, title probability, pace, offensive and defensive efficiency, and the fantasy consequences of those states.
• The full 2026-27 game schedule was not available from an official NBA source at the cutoff. Schedule-sensitive outputs therefore belong in a later snapshot; see Schedule and streaming desk.
• This is a research desk, not betting advice. Prices are used only as noisy external forecasts.

Desk verdict

There is no single authoritative 2026-27 team projection yet. The cleanest public baseline found is NumberBench’s July 6 model, but its own methodology says it only regresses final 2025-26 records one-third toward the mean and makes no roster or offseason adjustment. It is useful precisely because it is simple and falsifiable, not because it is complete.

The market layer is richer but less stable. FanDuel’s post-Finals article, VegasInsider’s futures tracker, and FOX’s dated odds snapshot disagree because they capture different books, dates, liability, and roster news. They should be stored as timestamped observations, never overwritten as if they were one consensus.

The encyclopedia’s eventual team forecast should combine a roster-and-minutes model, player impact, pace and efficiency, health scenarios, and schedule simulation. A market ensemble can be a comparison prior, but not the hidden answer key.

The maintained output table belongs in 2026-27 NBA team outcomes. This research note audits external inputs and the translation method; it does not supersede that shared projection.

Source register

Public historical-regression baseline

NumberBench describes a deliberately minimal model: take the final 2025-26 standings and regress each team one-third toward league average. Its output is schedule-neutral. These are model projections, not reported expectations from the teams or league.

The largest disagreements with the July 17 secondary FanDuel transcription are useful audit flags, not automatic “edges”:

† The 30 totals were not independently captured across FanDuel’s team pages, so the transcription is unverified. Its league sum is 1,229 wins by desk calculation, close to the NBA’s 1,230-game conservation total; that supports internal plausibility but does not authenticate it. The shared table in 2026-27 NBA team outcomes should retain its own provenance rather than silently inherit these lines.

These gaps expose where a prior-record-only model is processing radically less information than a market. Every large difference requires a roster-version, health, and competitive-context audit. A bookmaker total is also closer to a priced median threshold than a clean expected-win estimate; the over/under juice, alternative-win ladder, limits, and timestamp matter.

Championship-price snapshots

These are market observations, not title forecasts endorsed by the desk.

FanDuel research-article snapshot

The FanDuel article listed Oklahoma City and San Antonio at +260; New York +850; Boston +1300; Toronto +1900; Philadelphia, Denver, Detroit, Miami, Minnesota, and Cleveland +2200; Indiana +3500; Golden State and Houston +4000; Orlando and the Lakers +5000; Portland +7000; Atlanta +8000; Charlotte and Phoenix +15000; Dallas +17500; Washington +22500; Utah +25000; New Orleans +50000; Memphis +70000; and Brooklyn, Milwaukee, the Clippers, Chicago, and Sacramento +75000.

Store those values with the article timestamp. The page itself says lines are subject to change.

Movement and market-position context

VegasInsider reported San Antonio and Oklahoma City as BetMGM co-favorites at +260 and New York at +900 around July 16. It also reported that Oklahoma City held 32% of money on 3.1% of title tickets, and that Golden State moved from a +6600 opener to +3000 amid LeBron-related speculation. Those ticket/handle splits describe a bookmaker’s positions and customer behavior; they should never be converted directly into team-strength rankings.

The same page preserved a DraftKings opening snapshot led by San Antonio and Oklahoma City at +250, Boston +550, New York +700, Indiana and Denver +2800, Minnesota, the Lakers, Detroit, and Cleveland +3000, and longer prices thereafter. Openers are particularly vulnerable to becoming stale after transactions.

FOX labeled its numbers “as of June 23” despite a July 1 update date: San Antonio +230, Oklahoma City +250, New York +850, Boston +1300, Philadelphia +2000, and Denver, Detroit, and Toronto +2200. That mismatch is a useful provenance warning: the odds-as-of field, not just the article’s latest-modified date, determines comparability.

How to use a betting market without laundering it into fact

For positive American odds +A, raw implied probability is:

p = 100 / (A + 100)

For negative American odds -A, using the positive magnitude A:

p = A / (A + 100)

For a mutually exclusive title field, remove the book’s overround with:

q_i = p_i / Σp_j

That normalization is only a first pass. Futures can have different limits, stale teams, asymmetric shading, and substantial hold. A better market ensemble should:

1. capture book, jurisdiction, market, price, and exact timestamp;
2. remove each book’s overround within the complete field;
3. weight liquid, recently updated books more heavily;
4. retain the dispersion across books rather than publishing only an average;
5. flag prices tied to rumors or unresolved transactions;
6. freeze snapshots so later movement cannot rewrite what was knowable.

Do not infer expected wins from a bare 44.5 total. At minimum, retain the over and under prices. Prefer the alternative-win ladder and fit a discrete win distribution; if a parametric variance assumption is required, label it.

Canonical team-outcome model

1. Shared roster and minutes snapshot

The team model and player fantasy projections must use the same roster version and the same 240 regulation minutes per team-game. Each player needs base minutes, role, games-available distribution, and offense/defense impact. Otherwise the encyclopedia can simultaneously project a traded player on two teams or invent more opportunity than exists.

2. Possession-level team state

Build projected pace, offensive rating, and defensive rating from:

• minutes-weighted player contributions;
• lineup fit and role interaction rather than simple box-score addition;
• coaching/system and continuity priors;
• age and development curves;
• health and return-to-play states;
• roster slots still subject to transactions.

Use an impact metric only with its name, season window, source, and known limitations. Do not add an all-in-one impact estimate to the same box-score components from which it was built; that double counts information.

3. Wins from games, not isolated team ranks

A simple prior can map projected offense and defense to win percentage with a calibrated Pythagorean form:

Win% = ORtg^γ / (ORtg^γ + DRtg^γ)

But the final model should simulate head-to-head games. For every game, sample availability and role states, estimate possessions and margin from both teams, and account for home court, rest, travel, and schedule density once the official schedule is published.

Head-to-head simulation automatically enforces the most important league-wide identity: barring canceled games or ties, the 30 teams’ expected wins must sum to 1,230. Independent team forecasts should be reconciled to that total before publication.

4. Correlated outcomes

Seed, play-in, playoff, and title probabilities are joint outcomes. One team’s win is another team’s loss, conference seeds cannot be occupied twice, and a major injury changes several related markets. Preserve each simulation draw long enough to calculate:

• wins mean, median, and 10th/25th/75th/90th percentiles;
• seed distribution;
• top-six, play-in, playoff, conference-title, and championship probability;
• pace, offense, defense, and net-rating distributions;
• correlations with player games, minutes, and fantasy value.

5. Scenario architecture

Publish at least three causal scenarios, not arbitrary plus/minus bands:

Separate transaction branches from injury branches. “If player X is traded” and “if player X misses 30 games” can have similar team-win effects but radically different fantasy beneficiaries.

Fantasy translation layer

Team success is context, not a fantasy category. Translate it through mechanisms:

• Pace: more possessions can raise counting-stat volume, conditional on minutes and usage.
• Offensive efficiency: can lift assists, threes, scoring efficiency, and lineup stability, but does not distribute equally.
• Defense: affects steals, blocks, transition chances, and matchup quality without guaranteeing an individual category.
• Rotation stability: competitive teams may have predictable roles; deep teams may also cap minutes.
• Blowout distribution: dominant or noncompetitive teams can lose fourth-quarter starter minutes.
• Late-season incentives: seeding races, development priorities, and shutdown risk can change April availability, but should be scenario-weighted rather than asserted months early.
• Trades and injuries: the same event must update team outcomes, depth charts, minutes, usage, and player ranks together.

Never multiply a player’s rank by team win total or title odds. The causal chain is roster state → role/minutes → possessions and rates → fantasy production; wins are a downstream summary.

Data contract for every team page

Each team-outcome block should carry:

Calibration and maintenance

Freeze at least four forecast vintages: initial offseason, post-schedule, end of preseason, and opening night. Re-score them after the season with:

• mean absolute error and root mean squared error for wins;
• calibration curves and Brier score for binary outcomes;
• log score for mutually exclusive seed or title fields;
• continuous ranked probability score for full win distributions;
• coverage of the published percentile intervals.

Compare the model with historical-regression baselines and dated market ensembles. Closing market prices are a benchmark, not “truth”; actual outcomes score every forecast.

After the official schedule lands, rerun rather than hand-adjust. After every material transaction or injury, update the shared roster/minutes snapshot before changing team wins.

Evidence gaps at cutoff

• No primary, timestamped 30-team win-total capture was recovered. The full table above remains a secondary Reddit transcription.
• The official 2026-27 NBA schedule was not available, so opponent, rest, travel, Cup, and fantasy-playoff schedules cannot yet be simulated.
• NumberBench is transparent but intentionally omits offseason roster information; it cannot serve as the encyclopedia’s final projection.
• Public title odds were captured at different dates and books. No cleaned, de-vigged multi-book consensus has yet been stored.
• No public source found supplied a complete 30-team 2026-27 pace/ORtg/DRtg forecast with documented roster and minutes assumptions.
• Major rumor-driven prices require transaction resolution before they can inform a base case.

Open questions

• Can a primary-source scraper capture all 30 win totals, both sides’ juice, and alternative-win ladders at one timestamp and jurisdiction?
• Which player-impact source is licensed and reproducible enough for the canonical roster-to-team bridge?
• What simulation architecture will let player availability, minutes, team wins, and fantasy totals share the same random states?
• Which dates should be frozen as official encyclopedia forecast vintages after the schedule release and preseason?
• How should unresolved transactions be weighted between the base roster and explicit scenario branches?