Methodology
How the signal engine actually works, in plain language. Last updated August 18, 2026.
This page explains what DEFINTEL measures, where the data comes from, how a score gets computed, and what our SHA-256 provenance claim does and does not prove. No performance figures live on this page, since methodology and results are different questions, and this one only answers the first.
What a signal is
Every trading day, the engine scores a universe of tickers across a set of independent data channels, including institutional filings, options activity, insider transactions, dark pool flow, macro conditions, and sector rotation. Each channel that fires on a ticker contributes to that ticker's composite score. A signal is the resulting ranked output: a ticker, a score, the channels that fired, and the market regime at the time it was computed.
A signal is not a trade recommendation. It is a statement about what the data currently shows, sourced and timestamped, for you to weigh yourself.
The 34-channel engine
The stock intelligence engine currently runs 34 channels, defined in a single source-of-truth registry that is validated at import time, so a channel cannot silently disappear from the count without the registry itself failing to load. Channel weights vary by market regime (for example, a credit-tightening regime weights different signals than a credit-easing one), and weights are re-evaluated on every scoring run rather than fixed once.
Not every channel fires on every ticker on every run. A channel firing depends on there being something in the underlying data worth flagging, such as unusual institutional accumulation, an options-structure anomaly, or a congressional trade disclosure. The channels that did fire for a given signal are shown alongside its score, not hidden behind a single number.
Where the data comes from
| Category | Source |
|---|---|
| Institutional holdings (13F) | SEC EDGAR quarterly filings, parsed directly from the primary filing data |
| Congressional trading | Public congressional trade disclosure filings |
| Insider transactions | SEC Form 4 filings |
| Options activity | Options chain and flow data from our market data provider |
| Macro conditions | Public macroeconomic and credit-market data (FRED and related sources) |
| Price and volume | Standard market data feeds |
Every category above is either a public regulatory filing or a licensed commercial data feed. None of it is synthetic or fabricated to fill a gap; where a source is temporarily unavailable or degraded, the affected channel is designed to contribute nothing rather than substitute a placeholder value into a live score.
SHA-256 provenance: what it proves and what it doesn't
Every signal is SHA-256 hashed at the moment it is generated, over its ticker, score, run time, and fired channels, before any market outcome exists. The hash is computed at generation time, not after the fact, so it cannot be back-dated to a signal that performed well.
What this proves: if you have a signal's published fields and its published hash, you can recompute the hash yourself and confirm the content hasn't been silently edited since it was published. That is a real, checkable property.
What this does not prove: the hash is stored alongside the data it hashes, in a system we control. That makes it tamper-evidence against accidental or casual edits, not a third-party-anchored, externally verifiable proof against deliberate retroactive editing by us. We are not going to overstate that distinction: a hash sitting next to its own data is a real integrity check, not a public, independently-anchored ledger. If we ever add external anchoring, this page will say so and explain what changed.
Regime weighting
Markets behave differently in different conditions, so a channel that matters most in a risk-on, credit-easy environment isn't necessarily the one that matters most in a risk-off, credit-tight one. The engine classifies the current regime from live macro and credit data and applies regime-specific channel weights accordingly, rather than a single fixed formula regardless of conditions. The regime label shown alongside a signal (for example, a bull or credit-tightening tag) reflects the conditions the engine measured at the time the signal was computed.
What "34 wired" means, precisely
When you see a channel count on this site, it describes channels that are defined, registered, and validated in the engine, wired in engineering terms. A wired channel contributes to scoring whenever its underlying data source has something to report; it does not fire on every single run for every single ticker, because most of the time most channels have nothing unusual to flag for most tickers. That is expected behavior, not a defect.
Congressional trading data
Congressional trade disclosures are public record, filed under the STOCK Act. We treat this as one input channel among many rather than a standalone signal: a single disclosed trade is a data point, and the engine weighs it alongside institutional, options, and technical channels rather than in isolation. You can browse the raw disclosures we track, with no account required, on our Congressional Trades page.
Limitations, stated plainly
- Filings-based channels (13F, Form 4, congressional disclosures) are inherently lagged, reporting positions and trades that already happened, sometimes by days or weeks, because that is how the underlying disclosure regime works. We do not claim real-time visibility into filed positions; we claim faster synthesis of what has been disclosed than most retail tools offer.
- A signal reflects what the data showed at computation time. Markets move; a signal is not continuously re-scored between runs.
- No amount of data synthesis removes market risk. Signals are informational, not investment advice, and past patterns in the underlying data do not guarantee future outcomes.
Questions about methodology that this page doesn't answer: chad@justiceapex.ai.