Witching Hour: even a ghost has to cite its sources

Description and use

The Witching Hour web app combines real sensor measurements, ML analysis with TabPFN and Gemma, and narrative storytelling, producing a dossier with timestamped evidence, crew accounts with citations, and a recorded séance, enabling transparent ghost‑hunting investigations.

Problem

Ghost‑hunting teams need a way to combine real sensor data, machine‑learning analysis, and narrative storytelling while ensuring evidence is traceable and claims are cited.

Outcome

The app produces a structured dossier containing timestamped evidence, ML‑estimated anomaly ranges, crew accounts with citations, and a recorded séance, enabling transparent and reproducible ghost‑hunting investigations.

Workflow

• 1. Load the Witching Hour web app (PARANORMAL.exe).
• 2. Select a case (The House, The Unheard, or The Room).
• 3. For The Unheard, upload or play the recording; the browser measures light, sound and motion and marks events.
• 4. The app presents evidence events with timestamps in an evidence locker.
• 5. TabPFN estimates anomaly ranges for each sensor reading.
• 6. Gemma reads evidence cards and writes crew accounts, attaching citation chips to each sentence.
• 7. The crew’s dialogue is displayed with a glass‑board interface; the glass moves based on scripted responses.
• 8. The final dossier is downloadable, separating evidence, crew accounts and the board conversation.

Evidence

I built an interactive ghost‑hunting web app that reimagines the Ouija board, known in Brazil as brincadeira do copo, with real measurements, machine learning and data science.

Source

Two open‑model tools give the crew something to work with: TabPFN is a foundation model for tables. Gemma is Google's family of open models.

Source

The evidence locker separates the staged scene into light, sound and motion events.

Source

Generated crew sentences must point to known evidence, and their numerical claims must pass the validator.

Source