Three things need to happen

The event tree

Each assumption case starts with mutually exclusive expansion scenarios: no expansion, one slot, or two slots awarded together or on a staggered timetable. Each expansion branch states the conditional chance that Seattle receives a slot. Seattle-selection branches then carry a dated award distribution, a conditional chance of successful delivery, and a readiness-lag distribution.

P(on-time Seattle launch)
= Σ scenario weight
  × P(Seattle selected | scenario)
  × Σ award-date probability
  × P(delivery succeeds | Seattle award)
  × Σ qualifying opening-aligned timing probability

With an annual opening policy, adding a lag to an award date produces a readiness date. The first-game planning date is the next configured annual season opening on or after that readiness date. A 21-month readiness lag therefore does not claim a game exactly 21 months after an award. The opening-aligned date counts only when it falls inside a configured planning-season window and strictly before the deadline. The displayed windows and opening date are planning assumptions, not NBA-published schedules.

Historical fixtures with seasonOpening is null preserve the earlier literal behavior: award date plus lag is used directly as the modeled first-game date. That compatibility mode is explicit and is not the current sourced-draft policy.

What the three stages mean

P(expansion)

Total posterior mass on scenarios in which the NBA expands.

P(Seattle awarded)

Total mass on branches where Seattle receives a slot. This already includes expansion-scenario uncertainty.

P(on time | Seattle award)

The chance of a qualifying launch before the deadline, conditional on Seattle having received a slot.

When the conditioning event has positive probability, the headline equals P(Seattle awarded) × P(on time | award). Expansion is not multiplied again because it is already included in the award probability.

Every branch has a home

Qualifying season probabilities, the probability of a later launch, and the probability of no Seattle launch form one complete partition:

Σ season probabilities + later + no launch + unallocated = 100%

Delivery failure and branches where Seattle is not awarded a slot belong in “no launch.” Successful launches outside the deadline belong in “later.” The headline is exactly the sum of qualifying seasons. “Unallocated” is normally zero; it holds probability only when direct terminal evidence resolves the headline but does not support retrospective claims about expansion or award stages.

Time and observations

The calculator receives an explicit “as of” date; it never reads the clock by itself. If no Seattle award has occurred, award branches on or before that date are removed. The remaining future-award and retained no-award branches are renormalized together. That is conditioning on elapsed evidence, not applying an arbitrary daily decay and not resetting future success to 100%.

A published no-award observation cannot be advanced beyond its verified effectiveThrough date. Choosing a later date in the sandbox extends that field only on a hypothetical in-browser draft clone. It does not claim that the live evidence has been checked through the selected date and never changes the published snapshot.

• An observed Seattle award makes expansion and selection certain; its real date replaces the prior award-date distribution.
• Evidence that rules expansion out through the deadline, or a formal deferral to a date on or after the deadline, resolves the target to 0%.
• An observed first game resolves it to 100% if strictly before the deadline, otherwise 0%.
• After the deadline without a first game, contradicted on-time branches are removed while later and no-launch branches remain.

Resolved outcomes use true zero and one. The old 1%/99% clamps do not apply.

Assumption cases and what-ifs

Cautious, central, and optimistic are three complete editorial assumption cases. The public assumption range is simply the lowest and highest resulting headline across those cases. It is not a confidence interval, credible interval, or claim that outcomes are statistically calibrated.

Deterministic what-ifs rerun the complete central event tree for one-slot expansion, six- or twelve-month award delays, a six-month preparation/readiness delay, and a 15-point post-award delivery setback. Under an opening policy, the preparation sensitivity moves readiness before the annual opening rule is applied. Their percentage point changes are scenario comparisons, not uncertainty bounds.

Evidence and the seven milestones

The seven existing milestones remain an observed supporting record: league action, ownership, fee, arena, rights and branding, draft rules, and schedule inclusion. They are not weighted peer probabilities and are not averaged into the headline.

Persistent evidence records identify publisher, source title, source publication date and precision, the separate review timestamp, the assumptions they support, and a counterargument. Primary source facts, attributed reporting, and editorial archive reviews remain visibly distinct. An editorial review records its reviewed URLs and cutoff; it is not an NBA confirmation. Public evidence can change conditional inputs only through a reviewed snapshot. Automated news sentiment, article volume, and LLM classifications do not move the forecast directly.

The six-period view

The forecast shows six consecutive Seattle civil months ending in the current period. Outside the approved April–August backcasts, each period keeps the latest replay-verified V3 run associated with that month. A run published late still belongs to the month its evidence covers, and the current run is never also drawn as an earlier period. A period with no verified run stays an explicit gap rather than a guessed value.

April through August 2026 are retrospective backcasts: the V3 method rerun on September 13, 2026 against the assumptions and evidence already available at each month's Seattle-time cutoff. They were not forecasts published at those month-ends, archives may be incomplete, and they are ineligible for scoring, calibration, or track-record claims. Movement reflects a change in the editorial assumption vintage, not proof that any individual story mechanically moved the number. Other months use replay-verified published V3 runs or explicit gaps.

Each mark spans the lowest and highest headline across the three complete editorial cases for that period, drawn on the same vertical scale as the current estimate and read the same way as the assumption range above. Per-period evidence cutoffs, reconstruction hashes, and the approved artifact boundary are recorded in the forecast specification and editorial approval record rather than on the dashboard.

Versioning, replay, and scoring

Every published V3 run must point to an immutable, validated assumption snapshot and its deterministic hash. Historical V1 and V2 runs remain replayable as explicitly labeled archives; they are not silently interpreted as V3 history or used for a V3 weekly delta.

A forward ledger pre-registers forecast questions before they resolve. Brier scores are descriptive summaries of those recorded forecasts. An empty ledger is shown as unscored with N = 0—never as evidence that the model is calibrated or outperforms a market.

Kalshi's KXNBASEATTLE contract asks a broader question about any qualifying NBA team in Seattle. It is not the same event as this expansion-only forecast, so it is unavailable as the default matched benchmark and is not used for Brier comparisons. If broader market odds are ever shown as context, the question mismatch must be disclosed beside them.