← Vigi-Sky   Methodology & Sources FR

Methodology & Sources

Vigi-Sky combines scientific astronomy, Claude AI analysis (Anthropic) and the Hatch UDB database (18,116 worldwide UFO cases). This page documents every data source, the AI analysis methodology, and the assumed limits of the approach. Full transparency.

1. Databases and external sources

Hatch UDB
18,116 historical worldwide UFO/UAP cases, compiled by researcher Larry Hatch over 30+ years. Covers 1900-2002, including military, civilian, multi-witness and physical-trace cases. Used for pattern cross-referencing by our AI.
larryhatch.net ↗
NASA APOD
Astronomy Picture of the Day — astronomy image curated daily by professional astronomers since 1995. Our site shows the daily image with Claude AI contextual analysis.
apod.nasa.gov ↗
NASA DONKI
Database Of Notifications, Knowledge, Information — coronal mass ejections (CMEs), geomagnetic storms, solar flares. Real-time updates.
DONKI ↗
NASA NEO API
Near-Earth Object API — near-Earth asteroids, closest-approach distance, velocity, potentially hazardous status (PHA). JPL/Caltech.
api.nasa.gov ↗
NOAA SWPC
Space Weather Prediction Center — planetary Kp index (geomagnetic activity), 3-day aurora forecasts, storm alerts.
swpc.noaa.gov ↗
Where The ISS At?
Live ISS position — latitude, longitude, altitude, velocity. Updated every 5 seconds.
wheretheiss.at ↗

2. Claude AI

Vigi-Sky uses Claude (Anthropic) via the official API for four distinct tasks:

  1. Submitted UFO observation analysis — cross-references the description (shape, color, motion, location, time) with Hatch UDB and proposes plausible natural hypotheses (aircraft, Venus, Starlink, weather balloon) before flagging the unexplained elements.
  2. PDF report generation — contextualized synthesis with pattern analysis, historical comparisons, and graphical rendering.
  3. Wiki Intelligence (PromptForge) — evolving contextual base that enriches each prompt with the site's accumulated knowledge.
  4. VigiChat — conversational assistant to explain an astronomical phenomenon or a famous UFO case.
Important — AI accuracy: the 98% shown on the homepage refers to internal coherence between Claude's analysis and the natural hypotheses documented in Hatch UDB on a manually validated subset. It is not a measure of "absolute truth" about the nature of UFOs — a topic that is, by definition, open.

3. Editorial approach

Vigi-Sky takes a scientific, open and honest stance:

4. Assumed limits

Limit 1 — AI does not decide

Claude proposes analyses, never verdicts. The "unidentified" box remains legitimate as long as it is argued.

Limit 2 — the database is historically biased

Hatch UDB over-represents Anglo-Saxon countries (USA, UK, Australia) and under-represents Asia, Africa, Latin America. A Japanese sighting has fewer comparable "neighbors" than a Texan one.

Limit 3 — the figures evolve

The 12,840+ UFO observations sum user submissions and the Hatch historical archive. The number is updated at each deployment, not in real time.

Limit 4 — models evolve

Claude is an evolving AI model. An analysis in May 2026 is not strictly reproducible in May 2027: we archive signed analyses (model + date) for traceability.

5. Privacy and data

Vigi-Sky is free, no mandatory signup, no advertising. No data resale. Observation submissions are stored with explicit consent. GPS coordinates are rounded to one degree for public hot-spots.

6. Open source and contributions

The Vigi-Sky code is partially open. The Wiki Intelligence (PromptForge) system is publicly documented. To propose a case correction, report a factual error, or contribute to the database: contact page.

7. Support

The Claude AI API has a cost (~$0.50 per generated PDF report). Vigi-Sky is supported by Ko-fi ↗ to cover these costs. 100% of donations fund the API and hosting, never any commercial operation.