Public research snapshot
The evidence layer is learning what not to trust.
This snapshot shows how Stock Analysis Desk treats early hit rates, failed signal families, sample-size gates, and paper-only research status. It is a public summary of aggregate research evidence, not a private portfolio view.
- Snapshot date
- 20 Jul 2026
- Mode
- Research only
- Execution
- Locked
Current stock evidence
81.8%
11 / 60 decisive samples
Encouraging, but still too thin to trust as an edge.
Option evidence
100%
4 / 30 decisive samples
A good example of a result that looks impressive but is too fragile to overread.
Price-volume momentum
0%
12 decisive samples
This signal family is being treated as failed evidence, not promoted scoring.
Combined-regime method
75.0%
16 decisive samples
The strongest current family, but still collecting evidence before trust.
01 / What changed
The product now collects evidence before it asks for trust.
The most useful finding so far is that a high hit rate can still be weak evidence. Current stock evidence is showing 81.8%, but it only has 11 decisive outcomes against a 60-outcome target. Option evidence is even more extreme: 100%, but only 4 decisive outcomes against a 30-outcome target.
That changed the direction of the workspace. The goal is less show me a signal and more show me whether this idea has earned trust yet. Early evidence stays visible, but execution remains locked while the sample is small.
02 / Collection rules
What the system watches before promoting evidence.
Decisive outcomes only
Flat or tiny moves are not allowed to inflate accuracy. The model waits for meaningful outcomes before counting evidence.
Version-aware hit rate
Accuracy is tracked by method and model version, so a newer rule cannot borrow trust from older behavior.
Failure is preserved
A weak signal family can be quarantined while its observations remain visible for review.
Execution stays locked
The workspace can show research progress without turning early evidence into a live trading instruction.
03 / Promotion gates
Why the research state is still locked.
| Gate | Progress | Read |
|---|---|---|
| Stock evidence | 11 / 60 | More current-version decisive outcomes needed. |
| Option evidence | 4 / 30 | More execution-adjusted option outcomes needed. |
| Paper trades | 0 / 30 | Closed paper outcomes are still collecting. |
The current blocker is evidence maturity, not UI readiness. Stock, option, and paper-trade samples have to grow before the system can treat a pattern as trustworthy.
04 / How to read it
The point is skepticism with a memory.
The snapshot is not a performance claim. It is a record of what the collection layer has seen so far: which evidence is promising, which evidence is fragile, and which signal family is already failing.
That is the product direction: keep the original thesis, collect outcomes, preserve contradictions, and make it harder to mistake a small-sample hit rate for a reliable edge.