Computed from 649 analyses, August 24, 2026 to September 19, 2026
Five stages. One written thesis.
Market Scanner does not score a stock. It runs an argument — and hands you the argument.
Below is the same sequence you watch while your own scan runs. Every figure on this page is computed from the analyses we have published.
What we see across every analysis
Five things that are true about how it behaves.
The analysts almost never agree.
Four analysts read four different things and vote independently, before anyone rules on them. They rarely land in the same place — and that disagreement is what the debate stage is handed, rather than a number we averaged away first.
Did not all reach the same verdict
79.32%
Reached the same verdict
20.68%
Counted only across analysts that reached a verdict. An analyst is allowed to decline, and a decline is not a vote.
The debate changes the answer.
Stage two is not a tally of stage one. A bull and a bear argue it out, and a third agent rules on the argument — which regularly lands somewhere the analysts did not.
The ruling differs from the analysts' leaning
41.9%
The ruling follows the analysts' leaning
58.1%
Counted across analyses where the analysts reached a leaning at all. A split vote points nowhere to differ from.
Risk review makes it less cautious, not more.
A stage named risk is assumed to shave exposure. It does the opposite. When it overrides the proposal, almost every override softens the call.
Softened the call
93.24%
Made it more cautious
6.76%
Share of the overrides themselves, not of every analysis.
Where the stages agree, they agree strongly.
Disagreement is real, but it is not the norm — which is what makes the cases where they diverge the ones worth reading.
87.52%agree
On 87.52% of analyses the debate ruling, the proposed position and the risk revision all land on the same call.
Its own confidence predicts nothing.
Every analysis states how sure it is. We checked whether that number tracks the outcome. It does not — so we do not lean on it, and neither should you.
How sure the model says it is
64%
What that number tells you about the outcome
no signal
This is why our calibration is keyed to the stock itself rather than to the model's confidence.
Deliberately absent
What this page does not claim.
No track record
We do not publish a hit rate. Split per stock it is almost always too small a sample to be honest about, and a precise-looking number the data cannot support is worse than none.
Nothing per stock
Every figure here is across the whole platform. Nothing on this page says anything about whether a particular stock is worth buying.
No model-tier comparison
Almost all analyses so far ran one configuration, so a comparison table would be one populated column. Whether behaviour varies with depth is unanswered, not answered no.
FAQ
Common questions.
Are these numbers kept up to date?
Yes. They are recomputed from the analyses we have published, and the window they were computed over is shown at the top of the page. Nothing here is written in by hand.
Do you publish how accurate the analyses are?
No. We publish how the agents behave, not how often they turn out to be right. We would rather show nothing than a precise-looking number the sample cannot support.
What happens when there is no data on a stock?
An analyst is allowed to say so, and that is treated as a distinct answer from a balanced view. A balanced view is a claim; declining to form one is not.
Is any of this financial advice?
No. Market Scanner produces AI-generated research and educational analysis, not personalised financial advice or a recommendation to buy or sell any security.
Read one for yourself
Every analysis on the platform is built this way.
Pick a stock and read the whole thesis — the call, the bull case, the bear case, and the reasoning behind each.