The system at a glance
Decifer is a deterministic intelligence engine. It does not have opinions. It has sensors, rules, and a synthesis layer.
The engine watches eleven real market instruments every cycle. From those prices, it applies a set of deterministic rules to determine which macro drivers are active. Active drivers activate themes. Active themes identify which companies are in the flow of capital. The engine then publishes analysis of what this means in plain English.
No analyst is in the loop between the sensors and the publication. The intelligence is computed, not written.
The sensor layer
The engine watches these eleven instruments:
| Symbol | What it measures | |--------|-----------------| | SPY | US equity market direction | | IEF | Intermediate Treasury yields | | HYG | High-yield credit conditions | | LQD | Investment-grade credit conditions | | USO | Oil prices | | ITA | Defence sector expectations | | SMH | Semiconductor sector | | NVDA | AI infrastructure bellwether | | UVXY | Volatility expectations | | GLD | Safe-haven demand and inflation | | IWM | Small-cap risk appetite |
Five-day returns are computed for each instrument on every cycle. These returns feed the driver rules.
The driver layer
The engine applies thirty-one deterministic rules to the sensor returns. Each rule evaluates whether a specific market driver is active. A driver is active when the price evidence clears a threshold. There are no judgment calls in this layer.
Examples:
- If IEF five-day return is negative (bond prices falling, yields rising), the yields_rising driver activates.
- If SMH and NVDA both post positive five-day returns above their thresholds, the ai_capex_growth and ai_compute_demand drivers activate.
- If ITA posts a meaningful positive return, the geopolitical_risk_rising driver activates.
When price evidence is absent or below threshold, the engine fails closed. No driver activates on missing data. This prevents the engine from fabricating activity when markets are quiet or data is unavailable.
The macro event layer
Price signals tell the engine what the market is doing. The macro event layer tells the engine what is happening in the economy that might explain it.
The engine reads news and earnings call transcripts. It classifies each article or call into one of fifty-three event types. The classification is done by a language model, but the result is a structured label, not an opinion. Events are stored with a time-to-live that reflects how long they are likely to be relevant.
Driver confirmation is a function of both price and events. A driver can be in three states:
- Confirmed — price signal and macro event evidence agree
- Price only — price is moving but no macro event explains it
- Event unconfirmed — a macro event implicates a driver but the price has not confirmed it
This three-state model prevents the engine from treating noise as signal. A single news article does not activate a driver. The price must confirm.
The earnings transcript layer
After market close, the engine processes recent earnings call transcripts. For each transcript, it extracts:
- Guidance direction (raised, lowered, maintained, withdrawn, or not given)
- Tone (confident, neutral, cautious, or defensive)
- Key topics discussed on the call
- Forward outlook summary
- Whether the transcript diverges from the headline result
Low-confidence extractions are suppressed. The guidance information is injected into the synthesis layer to give it context that a simple price read cannot provide.
The synthesis layer
One language model call per content piece. The model receives a curated fact payload: the active drivers, their confirmation status, the macro events, the market stress reading, the economic backdrop, and any relevant transcript intelligence.
It does not receive raw API data. It receives structured, curated facts. The model's job is to turn those facts into plain English that follows the voice rules.
The model does not make investment decisions. It explains what the engine's sensors and rules are detecting.
The content safety gate
Every piece of content passes through a safety gate before publication. The gate applies three checks:
Hard blocks — fail if any of these are present:
- Buy, sell, hold, short, or long used as instructions
- Price targets for individual securities
- Certainty language: "will happen," "definitely," "guaranteed"
Required hedges — fail if causal statements do not include:
- Could, may, historically, worth watching, tends to
Anchor check — warning if any implication does not reference a forward date or calendar event
Content that fails the hard block check is not published. No exceptions.
What the engine does not do
The engine does not predict stock prices. It does not tell you what to buy or sell. It does not have a portfolio. It publishes observations about what market instruments and macro events are signalling.
The intelligence is an input to your thinking, not a replacement for it.
The legal basis
Decifer operates as a bona fide financial publication of general and regular circulation under Section 202(a)(11)(D) of the Investment Advisers Act of 1940 and Lowe v. SEC, 472 U.S. 181 (1985). Content is published equally to all readers. It is not tailored to individual situations. No adviser-client relationship exists.
Not financial advice.