AI automation demo
You are viewing: Model
SignalOS turns news into a clear decision, with proof.
Live mode checks the latest saved data.
SignalOS turns news into a clear decision, with proof.
The project has two prediction systems. Both are shown in plain English.
Starts with a base rate, then treats each news item as diagnostic evidence that can move the YES prediction up or down.
Looks at the wider evidence pattern and compares today's story with earlier decision paths.
The app does not invent a number. It combines clear pieces of evidence.
1. Is the real situation moving toward the outcome?
2. Did an official source say it clearly enough?
3. Is the wording strong, vague, or contradictory?
4. The final YES prediction is the combined answer.
The app checks whether a news item supports the event, comes from a reliable source, and uses wording that matches the market rule.
For By Jun 30, those pieces produce a YES prediction of 34.6%.
This is how I would convince myself first, then someone else.
Hide the final market result. Feed news in time order. Check whether the prediction moves before or with the outside view.
Group past predictions into buckets like 20%, 50%, 80%. Check whether outcomes happen at roughly those rates.
Remove one evidence type at a time. If performance does not change, that input is not useful.
Compare against simple baselines: no-change, Polymarket-only, and headline-count-only.
A starter validation table for comparing SignalOS against Polymarket and simple baselines.
| Market | Model | Probability | Outcome | Brier score | Use |
|---|---|---|---|---|---|
| June 30 Iran operations | SignalOS archive read | 93.0% | YES | 0.0049 | Showcase replay check |
| June 30 Iran operations | Polymarket reference | 94.0% | YES | 0.0036 | Outside-view baseline |
| June 30 Iran operations | Naive no-change | 50.0% | YES | 0.2500 | Simple baseline |
| June 30 Iran operations | News-volume only | 68.0% | YES | 0.1024 | Weak baseline |
Lower Brier score is better. This table is a starter validation view, not enough by itself. The next step is running it across many resolved markets.
The chart should explain the market over time, not just decorate the page.
Use the market question as the chart title. Example: Trump announces end of U.S. military operations against Iran by June 30.
Show SignalOS estimate and Polymarket probability as separate time-series lines.
Keep news above the chart as annotated vertical markers, with CHANGED or READ state visible.
The app checks what happened and whether it was clearly confirmed.
Did the real situation improve?
Did an official source say it clearly?
Both checks shape the final number.
The prediction only rises strongly when the facts and the wording both support it.
For By Jun 30, the app combines real-world progress, official wording, and the wording gap.
Checks whether the real situation is improving.
Checks whether the public wording is clear enough.
Reduces the prediction when the wording is vague.
Combines the three checks.
The final number is lifted slightly because later dates should not be lower than earlier dates.
Later dates usually have higher predictions because there is more time for events to happen.
The final number shown to the user.
Whether the real situation is improving.
Whether someone official says it clearly enough.
A discount when the news is strong but the wording is vague.
The final number in simple parts.
Chance the real situation has improved by the deadline.
Chance that clear public wording appears.
Penalty when reality and wording do not match.
The date being checked.
How much the real situation appears to have improved.
How likely it is that clear public wording appears.
How much vague wording reduces the prediction.
The final prediction after the checks are combined.
It is simple enough to explain and specific enough to be useful.
Gives one answer. Fast, but misses why wording matters.
Could read many items, but it is harder to explain in an interview or live decision.
Good at asking when something may happen. Not enough for rule-based outcomes.
First asks what happened. Then asks if it was officially confirmed. That matches the real decision.
The app should not trust a headline by itself. It should check what happened and whether the wording is clear.
Raw news becomes a score, then a user-facing update.
Small daily chances add up over time.
Imagine checking every day and asking, "Did the event happen today?" The slider changes that daily chance.
Where the approach helps most.
A situation can improve before anyone says it clearly. The app checks both.
It helps most when the news is improving but the wording is still unclear.
It is weakest during sudden shocks that no news system can see early.
It does not just summarize news. It checks if the news should change the decision.
The biggest triggers to watch.
A clear official statement would raise the prediction fast.
A new strike or casualty report would lower the prediction fast.
The real question is whether official wording becomes clear.
Current values used by the prediction.
| Setting | Value | Range | Updated |
|---|---|---|---|
| Daily event chance | 1.8% | 0-15% | Live |
| Daily statement chance | 1.2% | 0-15% | Live |
| Wording gap | 0.22 | 0-1 | Live |
| Trust score | 22.1% | 0-100% | Live |
| Wording risk | 18% | 0-100% | Live |
| Statement after event | 22.1% | 0-100% | Live |