Skip to content
HelixEvent-driven alphaLaunch dashboard
AI TRADING INTELLIGENCE · LIVE ON SODEX · VERIFIED TRACK RECORD

Event-driven alpha.Audited.

Every signal carries its full reasoning chain. Every outcome is tracked. Every framework is stress-tested. Trade what you can verify.

Launch dashboard How the audit works
3
AI AGENTS PER SIGNAL · CLASSIFY · VERIFY · DEBATE
100%
SIGNALS SCORED AGAINST OUTCOME
40+
ASSETS · CRYPTO · EQUITIES · TREASURIES
24/7
CATALYST MONITORING

Most AI signals are unverifiable.
Helix isn't.

Per-signal audit trails

Click any signal to see its sources, classifier prompt version, conviction formula breakdown, asset relevance score, corroboration status, and gate-rule outcomes. The reasoning is exposed, not hidden behind a confidence number.

Outcome-tracked calibration

When a signal's horizon expires, we record what actually happened — target hit, stop hit, or expired flat. The calibration dashboard shows hit rate by tier, by catalyst type, and conviction calibration curves. We measure the way real funds do.

Stress-tested frameworks

Two portfolio frameworks running paper-traded in parallel. v2.1 was stress-tested against a real -35% BTC bear market and contained the loss to -19%. Acceptance criteria are documented; failures are visible. Trade-offs are stated.

Eight stages from news to signal.

  1. STAGE 01

    Ingestion

    Sanitize and validate raw news. Strip HTML, normalize timestamps, attach source tier classification.

    ingestion-validation.ts · I-01, I-02
  2. STAGE 02

    Classification

    LLM categorizes the event. Detect promotional language, dedupe near-identical articles, classify catalyst subtype.

    classifier.ts · digest.ts · promotional.ts · I-03, I-04
  3. STAGE 03

    Asset routing

    Score each candidate asset's relevance to the event: subject (1.0), directly affected (0.8), basket member (0.5).

    asset-router.ts · I-05, I-06, I-07
  4. STAGE 04

    Conviction

    Multi-component scoring. Adjust by entity history: contradictory recent signals lower conviction.

    entity-history.ts · I-15, I-16
  5. STAGE 05

    Risk derivation

    Stop, target, and horizon drawn from base rates per (catalyst subtype, asset class). No flat templates.

    catalyst-subtype.ts · I-10 through I-20
  6. STAGE 06

    Conflict detection

    Flag signals that contradict each other on the same asset. Relevance-weighted to suppress incidental-mention noise.

    conflict.ts · I-09
  7. STAGE 07

    Tier assignment

    Auto / Review / Info gating. Single-source claims cannot reach Auto regardless of conviction.

    tier-resolution.ts · I-21, I-22, I-23
  8. STAGE 08

    Persistence

    Pre-save invariant gate enforces every rule above. Lifecycle sweeper expires stale signals.

    invariants.ts · lifecycle.ts · I-13, I-14, I-30

Built to be inspected.

The audit page for a single signal — reasoning, sources, classifier output, gate-rule outcomes, corroboration status.
Evidence 01
Open the box.

Click any signal to see its full reasoning, the triggering news event, classifier output, gate-rule outcomes, and corroboration status. No black boxes.

View live audit →
The event stream — recent news headlines with classifier verdicts (event type, sentiment, severity) and asset mappings.
Evidence 02
Live event ingestion.

Every news headline is fetched, classified by Claude for actionability, sentiment, severity, and asset mapping, then gated by a corpus-similarity filter before it can fire a signal. Thousands of events on file, auto-polling for new ones.

View live event stream →
The AlphaIndex live portfolio dashboard — NAV, holdings, rebalance history, and Claude-written rebalance reasoning.
Evidence 03
An AI-managed index, paper-traded live.

AlphaIndex allocates across BTC, ETH, L1s, RWA, sector indexes, and perps based on accumulated news signals, sector momentum, and ETF flows. Real SoDEX prices, simulated fills. Rebalance rationale written by Claude — visible for every rebalance.

View live AlphaIndex →

Two frameworks. Documented trade-offs.
You choose.

v1
Design philosophy
Unconstrained
BTC anchor
Adaptive
Concentration cap
None
Drawdown protection
None
Worst stress test
−53% (2× BTC)
Capture in trends
Full
Best for
Maximum upside
v2.1
Design philosophy
Risk-controlled
BTC anchor
40 – 70% band
Concentration cap
8% per position, 15% per cluster
Drawdown protection
Circuit breaker at −8% / −12%
Worst stress test
−19% (0.55× BTC)
Capture in trends
~80% of BTC
Best for
Bounded downside

Both frameworks run paper-traded in parallel. The calibration dashboard shows actual outcomes side by side.

Built to be measured.

Not just to look impressive.

Launch the dashboard