Breakout
Price clears defined resistance on confirming volume.
How it works
From close to publish, in five steps.
Every weeknight after the US market closes, an automated pipeline runs end-to-end and ships a ranked, summarized list of chart setups before pre-market opens. The full process is below; the detector internals are not.
The pipeline
What runs each night.
- 01Ingest
Pull EOD candles for the full US universe
Around 8,400 US-listed equities and ETFs. Adjusted OHLC, volume, and corporate-action history flow in from regulated data vendors. The pipeline waits for late prints and re-runs after each settlement.
Universe~8,400 tickers - 02Detect
Six pattern detectors, one pass each
Each setup type — Breakout, Breakdown, Bull Flag, Uptrend Pullback, Mean Reversion, Downtrend Rip — has its own detector. They run independently across the universe. We do not publish what the detector does; we publish what it finds.
Detectors6 independent passes - 03Rank
Score each detection on five points
A detected pattern is not the same as a clean one. Every hit is scored on a 5-point quality scale that considers structure, trend context, and volume confirmation. Scores rank, they do not predict.
Output1–5 quality dots - 04Summarize
Generate the AI market digest
A guard-railed LLM reads the ranked list, the macro context block, and a fixed set of structural prompts. It writes a short summary of the day's themes — never names a "pick," never makes a price call, never invents a ticker.
Modelclaude-haiku-4-5 - 05Publish
Ship before US pre-market opens
The list, charts, milestone badges, and digest are stamped with a generation timestamp and published to subscribers. Once published, that day's list is immutable — corrections appear as new entries with explicit deltas.
CutoffBefore 6:00 AM ET
The six setups
The complete published taxonomy.
These are the only setups we publish. We do not detect (or publish) anything else. The illustrations are schematic — they show the shape of the pattern, not historical data.
Breakdown
Mirror of breakout to the downside — defined support gives way.
Bull Flag
Sharp uptrend (pole), tight consolidation (flag), then resolution.
Uptrend Pullback
A shallow dip inside an established uptrend, resolving with trend.
Mean Reversion
Inside an uptrend, price reaches a statistical extreme and turns back.
Downtrend Rip
Sharp downward momentum inside an established downtrend.
Quality scoring
A 5-point screening aid.
Every detection is scored from 1 to 5. Higher scores indicate a cleaner pattern relative to the detector's structural ideal — not a prediction of how it will resolve.
The score is the same number the app uses to filter the list. Five dots is rare; three is common; one means the detector saw the shape but the structural read was weak.
Read the methodology Five — Textbook
Clean structure, strong trend context, confirming volume.
Four — Strong
Tight setup with one mild structural concession.
Three — Workable
Recognizable pattern with notable structural weakness.
Two — Borderline
Pattern present but trend or volume context is poor.
One — Detected
Shape only; structural read failed most checks.
Macro context
Setups don't exist in a vacuum.
Each day's digest sits in front of a short macro context block — index breadth, sector leadership, volatility regime. The detectors don't see this; it's there for you, so a high-quality Breakout in a deteriorating tape reads differently from one in a clean uptrend.
Macro · specimen
redacted · 2026-05-12
Index breadth
Strong
Sector leadership
Cyclical
Volatility regime
Low
Risk-on / off
Risk-on
The macro block describes the tape; it does not advise. It is not used as a detector input or as a scoring weight.
Want the full transparency document?
The methodology page covers what the quality score is, what it isn't, and why the digest never names a ticker. The disclosures page covers everything else.