Core Concepts
The primitives behind Telumin: watchlists with theses, the threat engine, the Investment Diary, and the Evidence Vault.
Signals
A Signal is the published output of the threat engine: a structured tuple of (ticker, direction, confidence, evidence_chain) with attached quant context. Signals only ship after the Adjudicator rules and the materiality bar is cleared.
BULLISH
Increase exposure
BEARISH
Reduce or exit
DRIFT ALERT
Day-1 condition broken
NEUTRAL
No action recommended
The Threat Engine
Every candidate threat travels through three stages. Each stage runs as a Temporal activity and writes its intermediate output to the Evidence Vault.
2-Source Verification
The SkepticValidator enforces a hard rule: no signal ships without at least two independent source confirmations. There is no override flag. There is no “just this once.” This is the discipline guarantee behind every Telumin-published signal.
| Stage | Responsibility |
|---|---|
| DataForager | Surfaces candidate documents (10-Q snippets, transcripts, sell-side, social-graph) |
| Normalizer | Strips duplicates, normalizes structure, attaches source-credibility weights |
| QuantAnalyzer | Attaches momentum, EPS revision, IV percentile, insider activity |
| SkepticValidator | Counts independent sources; vetoes if < 2; logs every veto |
| SignalPublisher | Writes the signal + full evidence chain to the Vault and emits |
Investment Diary
Every position on a watchlist should be backed by an Investment Diary entry: a thesis, Day-1 conditions, and a stop policy. Thesis-guard checks have nothing to verify against without a written thesis.
Drift, Stops, and Thesis Guard
Three engines run continuously against your live portfolio:
Rebalancing
Drift > threshold → queue orders
Trailing Stop (Ukemi)
Lift floor on every new high
Thesis Guard
Day-1 condition broken → alert
Platform Concepts
Evidence Vault
Immutable record of every signal, source citation, and alert. Exportable to your SIEM or compliance file.
Investment Diary
Required before the protect loop arms. Write your thesis, exit trigger, and position rationale — the engine watches them continuously.
Quant Context
Momentum, EPS revision delta, implied vol percentile, insider activity attached to every published signal.
Confidence Score
0–100 score combining source diversity, signal recency, and historical accuracy of the underlying thesis.
Backtest Lab
Run any signal strategy against historical data to validate a thesis before relying on it.
Signal API
REST + WebSocket access to verified signals for B2B desks. Reproducible across analysts.