Evidence

Prediction accuracy is not trading edge.

A correct prediction may be non-executable. A good execution proves nothing about predictive quality. One repeated, correlated signal is not many independent opportunities. Paper trading is evidence collection, not proof of profitability. Alpha Yazan is built so these can never be confused.

Forward Paper validation activeLive money disabledEdge not confirmedas of 2026-09-08

Five classes

What is recorded, and kept apart.

MODEL EVIDENCE

Frozen model and feature contracts with hashes; the historical decision-rate record of the frozen model; replay parity on identical vectors.

INPUT EVIDENCE

Decision-time captured inputs, hash-chained per day; independent rebuild of every feature vector; provider bar-revision rate reported separately.

EXECUTION EVIDENCE

Broker order, fill and position stream; deterministic client ids; reconciliation state for every trade; permanent append-only trade history.

RISK EVIDENCE

Governor state at order time (positions, exposure, daily loss, per-symbol, hard stop); deadman health; authority epoch and filter reasons for every blocked decision.

FORWARD EVIDENCE

Prediction outcomes judged after frozen horizons; independent-opportunity accounting; natural Paper trades kept apart from engineering trades and from shadow populations.

Evidence today

What exists, and what does not.

Forward Paper validation is active for both Brains. The outcome ledger holds tens of thousands of judged decisions. One natural Paper execution has completed end to end and was reconciled on both legs. Certification harnesses pass before any authority is granted.

independent quantitative Brains (Swing, multi-day, LONG/WAIT; Intraday, ~60-minute, SHORT/WAIT) behind one shared risk and authority engine

37,000+

Brain decisions recorded in the prediction outcome ledger and judged after a frozen horizon, traded or not

natural Paper execution completed end to end: decision, authority, order, fill, broker reconciliation, outcome

certification harnesses passing before authority is granted: input parity, execution route, end-to-end, provisional execution, input capture, finalized-data path, dashboard truth

What is not proven yet

Trading edge (sample size insufficient)
Fully calibrated Intraday execution timing (provisional 45-second offset)
Live-money authority (disabled by design)
Customers, revenue, partnerships (none; none claimed)

No return, win-rate, Sharpe or P&L figure is published on this site. Paper results are hypothetical by nature and are reviewed only inside Investor Access, with their limitations stated.

Lineage

One decision, end to end.

The first natural Intraday Paper execution, sanitized. It is shown because the governed path worked and because the exit was closed by a legacy guard, which the reconciliation layer detected, recorded truthfully and which was then corrected. Faults are evidence too.

decision      NUR-29814678      brain INTRADAY  action SHORT  p_short 0.610  threshold 0.60
inputs        captured-input ledger · parity PASS 31/31 · boot 20260908T123056Z
authority     AUTHORITY_PASS · epoch 2026-09-03T22:12:10Z · timing PROVISIONAL_PAPER_45S
governor      positions 0/10 · exposure $0/$2,000 · per-symbol 1 · hard stop 0.70%
order         sell 1 · market · client_id INTRADAYV3-…-E · dispatched T+45s
fill          15:19:45Z @ 5.35 · broker position −1 · reconciliation OK
exit          15:36:59Z @ 3.88 · reason EXTERNAL_CLOSE_BEFORE_H60 · both legs reconciled
outcome       judged at horizon end · evidence class NATURAL_PAPER · counted as 1 independent opportunity
edge          NOT CONFIRMED · n = 1 completed natural trade · required n ≥ 20

Engineer view

For technical readers.

How is an "independent opportunity" defined?

One independent opportunity per contiguous directional episode of (brain, symbol, direction, population); a new episode starts only after the previous episode's latest horizon end; the representative decision is the earliest non-censored component. Repeated correlated decisions never make evidence strong or confirmed.

What does "fail closed" mean in practice?

Every authority input is verified at decision and at order time. A missing captured-input ledger, a failed parity confirmation, a missing authority epoch record, a stale decision, an unreadable timestamp or an unfrozen execution offset each block submission with a recorded reason. Nothing defaults to open, and there is no automatic fallback to weaker data.

How are provider data revisions handled?

Captured inputs are authoritative for the decision. Later provider revisions are measured and reported as evidence, never used to rewrite a decision. When a systematic timing skew was found (provisional daily bars at 17:00 ET versus finalized consolidated bars), the Swing decision was moved to finalized session-T bars behind a two-snapshot stability gate, with a strict session-T information boundary.

How are natural and engineering trades separated?

Engineering legs (plumbing tests, certification orders) carry distinct client-id lanes and are excluded from natural evidence and P&L. Shadow populations are never combined with the live-forward population; pre-authority decisions are labeled and cannot become orders.

What is deliberately not exposed publicly?

Feature definitions, model parameters, thresholds beyond the published contract shape, hostnames and infrastructure details, credentials and operational secrets. The read-only investor dashboard shows system state and evidence, not proprietary internals.

Investors

See the evidence, then decide.

Investor Access opens the read-only system dashboard. Request Investor Materials to start a conversation with the founder. Nothing here is an offer of securities.

Investor Access Request Investor Materials