Core Protocol

Verification Protocol

No single model is allowed to surface a signal alone. The 2-source rule and the SkepticValidator gate everything that reaches the Diary or the broker.

Why two sources, not one model

Single-model AI hallucinates. Single-source signals get faded by the market. The Telumin pipeline solves both at once: every signal must be backed by at least two independent sources of meaningfully different type, and the SkepticValidator must fail to disprove it. Anything that can’t clear both gates is logged for review but never auto-traded.

The pipeline that emits signals

1

DataForager

Pulls candidate facts from filings, exchanges, primary news, calls, and (carefully tiered) social. Tags every fact with source, timestamp, and tier.

2

Normalizer

Resolves the raw facts to canonical entities (issuer, ticker, instrument, event type) and de-duplicates across sources.

3

QuantAnalyzer

Pulls market context: price, vol, options skew, holdings overlap, sector beta. Attaches the quant frame the Skeptic and Diary will read against.

4

SkepticValidator

Adversarial. Tries to disprove the candidate signal using independent sources, prior thesis state, and statistical priors. Outputs PASS / DRIFT / BREAK with rationale.

5

SignalPublisher

Emits the cleared signal to the terminal, the Diary, the Signal API, and any subscribed workflow (rebalance, trailing stop, thesis guard).

Skeptic pass requirements

A signal is only emitted when all of the following are true:

Independent sourcesAt least 2, of different tier where possible
Tier mixTier 1 (filings, exchanges) preferred — social alone never qualifies
No disqualifierSkeptic must fail to find a contradicting Tier-1 fact
ConfidenceWeighted score ≥ 65 (0–100 scale)
Quant contextQuantAnalyzer frame attached

Confidence scoring

Every emitted signal carries a 0–100 confidence score made of four weighted factors:

Source tier & diversity (35%)

How many independent sources, and what tier they sit at

Skeptic agreement (30%)

How firmly the Skeptic could not disprove the candidate

Historical accuracy (20%)

Track record of similar signals from these sources

Recency (15%)

How current the corroborating evidence is

Pipeline metrics

2+

Sources required

< 5 min

Avg pipeline latency

100%

Audit-logged trades

Diary

Append on every signal