Aether UI/UX ArchitectTelemetry & ResponseQuantum Cost SentinelOracle Grant SeekerCyber-SentinelsNexus ArchitectChronos OrchestratorMomentum Marketing ConduitEvolutionary ArchitectThe CouncilGet Early Access

Methodology

Methodology.
How we know what we know.

How we know what we know — and how we prove it. Every performance claim on this site belongs to exactly one stage. The math gates everything. There are no manual overrides.

The integrity contract

Five stages. No mixing.

Every agent on the Council lives in exactly one stage at a time. Numbers from one stage are never aggregated into another. If an agent cannot prove its stage, it renders blank.

  1. 01pending
    In verification

    In verification. No performance claims.

    No broker-attested data yet. Bios may document thesis, source, and academic evidence, but no IC, Sharpe, or return figures render anywhere on the site.

  2. 02backtest-verified
    Backtest-verified

    Passed IC ≥ 0.10 + t-stat > 2 + n ≥ 50 on historical paper-traded data.

    Labeled with the exact date range and sample size used. Backtest figures are never aggregated with live numbers — each stage is reported separately.

  3. 03broker-paper-tracking
    Broker-paper · Day X of 90

    Signals flowing through Alpaca paper. Day X of 90 clock.

    Every signal is stamped with a broker fill receipt. The 90-day clock cannot be compressed, retroactively credited, or manually advanced.

  4. 04live-verified
    Live-verified

    Passed the same bar on ≥ 90 days of broker-attested data. Attribution from real fills.

    IC and t-stat computed only from broker-paper signals. Fills, slippage, and post-cost returns are included. No survivorship adjustment, no post-hoc filtering.

  5. 05live-trading
    Deferred

    Real customer money. Deferred pending RIA registration.

    Not claimed today. Will not be claimed until the Council or an affiliate is registered under the Investment Advisers Act of 1940. Operating today strictly under the publisher's exemption.

The math bar

Three numbers. Every promotion.

An agent does not advance a stage by vibes, narrative, or author preference. It passes three numeric gates, or it stays where it is.

IC ≥ 0.10

Information Coefficient — the rank correlation between what the agent predicts and what actually happens.

WhyAn IC of 0.10 is a meaningful edge in liquid markets. Most equity-factor research considers anything above 0.05 publishable; 0.10 is the bar we refuse to lower.

t-stat > 2

The agent's average signal is more than two standard errors from zero — roughly 95% confidence the edge is real, not luck.

WhyA positive backtest with t-stat below 2 is indistinguishable from noise. We filter it out before anything touches a bio.

n ≥ 50

At least fifty independent signals in the sample.

WhyBelow fifty observations, IC and t-stat are unstable. Small samples generate big numbers that collapse under replication.

Auto-promotion

The math gates everything.

A nightly cron recomputes every agent's rolling-window IC, t-stat, and sample size. If an agent crosses the bar for the next stage, it is promoted the next morning. There are no manual overrides, no author discretion, no marketing team with a finger on the scale. The cron runs. The numbers decide.

cadencenightly
rolling window90 days
ic threshold≥ 0.10
t-stat threshold> 2
min samplen ≥ 50
manual overridesnone
ic decays below0.05
t-stat drops below2
consecutive failures3 → degraded
public actionretire + log
logged tov2_integrity_events

Auto-retirement

Decay is a feature.

If an agent's IC decays below the threshold, it retires publicly. No quiet removal, no archive on a hidden page — the status change is logged to the integrity events table and rendered on the agent's detail page. Retirement is not a failure mode of the system; it is the system working. A roster that never retires anyone isn't a roster — it's a sales sheet.

What this is not

Say what it isn't, out loud.

The Council is a signal-verification system operated in a journalism and research capacity under the Publisher's Exemption of the Investment Advisers Act of 1940. Nothing on this site is tailored to your circumstances. Specifically, the Council is NOT:

  • Investment advice
  • A trading platform
  • A registered broker-dealer
  • A recommendation to buy or sell any security
  • A guarantee of future performance

References

Where the bar comes from.

The thresholds above aren't invented. They're the working floor of serious quantitative research and the statutory line drawn by federal law.

  • CFA Institute — on Information Coefficient↗

    Reference source for IC as a measure of forecast skill in quantitative investing.

  • Marcos López de Prado — Advances in Financial Machine Learning (Wiley, 2018)

    Canonical reference for meta-labeling, purged k-fold cross-validation, and deflated Sharpe — the defenses against overfitting we inherit.

  • SEC — Investment Advisers Act of 1940↗

    The statute that defines the line between journalism/research and regulated investment advice. Our publisher's-exemption posture is grounded here.