Forecast trust and accuracy
Forecast accuracy is being tracked
The Star Window trust page explains what has been verified, which limitations remain, and which performance or outcome claims are not yet published. Use it to understand the boundary between supported forecast and astronomy features and claims that remain unavailable. Trust here means traceable evidence, explicit unavailable states, and honest publication status - not a promise that a forecast or observing session will be correct.
Use this page to see the boundary of what Star Window currently supports. A capability is described as verified only when its public contract and source evidence support it; missing atmospheric, access, imaging, alert, or accuracy claims stay unavailable rather than being implied by confidence or a strong score.
Reliability scores publish after enough verified nights are collected to avoid misleading small-sample claims.
Readiness: Tracking started
Sample window: More verified nights needed before publishing reliability scores.
What confidence means
Confidence describes how much support the current recommendation has from input coverage, agreement, and timing stability. It is not a guarantee that every part of a state or park will match the forecast.
- Higher confidence: usable numeric coverage, stable timing, and aligned providers.
- Lower confidence: sparse coverage, source disagreement, or rapid-formation risk.
- Curated destination guides use a verified forecast for one named map point when available. Large parks can vary, so the page identifies the point and keeps access and recheck caveats visible.
Forecast freshness
Forecast date: 2026-08-28
Status: fresh
Last generated: 2026-08-28T03:31:20.493162+00:00
7-day reliability
7-day accuracy: Pending
False go rate: Pending
False go nights: Pending
Average forecast error: Pending
Tracks nights that looked worth going but were later verified as poor.
Coverage + calibration
Verified outcome coverage: Pending
Evaluated rows: Pending
Verified coverage days: Pending
Conflicting source rate: Pending
Pending means the sample is still too small to publish responsibly.
Verification history and limits
Evaluated rows: Pending
Verified coverage days: Pending
Reliability metrics remain unavailable until the recorded sample is large and independent enough to support them. Pending values are intentionally not replaced with estimates.
Forecasts can still fail because of local cloud formation, terrain, smoke, access restrictions, and changes after the nightly dataset is generated.
Trust and verification questions
Does The Star Window publish a forecast accuracy percentage?
No approved public accuracy percentage is currently part of the verified product contract. Confidence describes evidence strength, not a measured probability that conditions will occur exactly as shown.
What happens when a required input is unavailable?
The affected claim or feature should fail closed: the value stays unavailable, the recommendation is limited or withheld, and unsupported data is not filled in with an estimate.
Does a Star Window forecast confirm access or safety?
No. Forecast evidence and access or safety information are separate. Official land managers, hosts, and other controlling sources own current access, closure, parking, and safety rules.