Weather evidence in. A personal opportunity shortlist out.

Whole-product architecture

Separate engines combine; none silently rewrites another

The customer selects a starting point, travel mode, and reach. Bluebird then joins independent evidence in a fixed order. Green engines work now, amber engines have an enforced contract but no live provider, and grey engines are planned product services.

Opportunity layer · version 2

Different trips use different published weather lenses

Each lens grades every day as Suitable, Marginal, or Poor from temperature, rain, and wind, then applies the same 50-member evidence discipline. The free read consistently uses Default Explorer; a signed-in personal search applies the member's chosen lens. Cloud cover is context, never a hidden rejection. Personal absolute heat limits can remove a place from the positive pool; they cannot rewrite its stored forecast evidence.

The complete path

Eight canonical gates from eligible place to answer

A row cannot skip a gate. The separate reach engine changes which destinations are evaluated; canonical weather evidence supplies the compatibility verdict; the opportunity engine may add a complete lens-based presentation without changing those source rows.

1. Accept an eligible set

   Receive destinations from the selected origin, mode, and reach. Verified routes and explicitly estimated Fly reach remain different evidence states.

2. Load one complete run

   Require every destination in one hashed target set and all four forecast windows from one completed ECMWF ensemble run. Never mix runs.

3. Validate 50 members

   Every destination needs exactly 50 complete numeric member series, valid dates, expected grid coordinates, and sane units.

4. Build four windows

   Seven-day windows begin at lead days 1, 3, 5, and 7. Their midpoint lead days are 4, 6, 8, and 10.

5. Judge every day

   A day matches only when all three published high-temperature, rain, and wind rules pass.

6. Classify every member

   Count matching days per seven-day window, then place the member in Strong, Borderline, or Poor.

7. Adjust decisive evidence

   After midpoint day 8, reduce Strong and Poor equally as lead-time uncertainty grows.

   effective = round(raw × lead weight)

8. Verdict, then selection

   Apply the symmetric 40-of-50 rule. Prefer Go over Hold, then higher effective Strong, shorter route band, and stable destination ID.

Weather contract

What makes one day match

All three conditions must be true. The boundaries are deliberate: temperature is inclusive; rain and wind are strict.

Challenge the threshold

Move the same controls the engine uses

Strong plus Borderline plus Poor always equals 50. Lead adjustment affects both decisive sides symmetrically and can only move an extreme verdict toward Hold.

Why range clicks matter

A retained SFO example shows how the pool grows

In this historical reference, the selected winner may remain Reno if none of the newly admitted candidates beats its weather evidence. That is different from the range failing: coverage, new rows, map shortlist, and counts must all change.

Examples added: 3h admits Albuquerque, Calgary, Denver, Seattle and five more; 4h admits Austin, Dallas–Fort Worth, Edmonton, Minneapolis and five more; 8h admits Boston, Honolulu, Miami, New York, Washington and fourteen more.

Input boundary

What changes the answer—and what does not

Real primary-window evidence

Three rows from the completed production replay

Trust boundary

When the board must show nothing