Market Perception
Two machines: a register for data and bets, and sensors for topology, herd behaviour and regime change.
Market perception has two parts: a register for data and bets, and a set of sensors that detect topology, herd behaviour and regime changes in markets. The sensors read the shape of the market rather than its level: persistence landscapes over a sliding window, an order parameter for herding, the topology of the correlation network. A separate dissonance sensor puts four things side by side, what is said in the news, what is bet in the markets, what is feared in the macro numbers, what is done on the ground, and flags where they disagree.
This is a perception, not a trading system. What it produces are readings and a regime verdict, dated, so that they can be checked against what happened next.
As of 8 October 2026: 161 instruments in eight classes, daily data since 31 May 2024 (crypto since 20 January 2025), four sensors, and a regime archive of 2,580 readings kept since 28 August. The correlation sensor does not yet see every instrument: it works on 57 of them, and two classes, indices and futures, are still outside its landscapes.
What goes in
Snapshot 8 October 2026Daily candles for 161 instruments across 8 classes, back to 2024-05-31 at the oldest: 860 days of depth. Each class starts on its own date, shown next to it. Public price data, nothing proprietary. The register also keeps bets: predictions with a date, so they can be scored later.
Instruments by class
- ETFssince 2024-0681
- defence stockssince 2024-0526
- indicessince 2024-0626
- cryptosince 2025-019
- equitiessince 2024-067
- FXsince 2024-095
- commoditiessince 2024-084
- futuresince 2024-123
Four sensors, and their state right now
| Meta-regime Combines topology, herding and correlation structure across all classes into one state. | calm |
| Topology Persistence landscapes over a sliding price window: how the shape of the market changes, not just its level. | normal |
| Herding An order parameter for how synchronously the instruments of a class move. | disordered |
| Correlation structure The topology of the correlation network between instruments: whether the market moves as one or falls apart. | decorrelated |
What comes out
Not trades. Readings: which way the market is moving against its own recent norm, whether the move has a story behind it, and what regime the whole thing is in.
Dissonance reading, 2026-10-08 12:48 UTC
Move over the window and its z-score against the instrument's own baseline. Left leans down, right leans up. Amber marks a hot reading.
- Crude oil (WTI)+3.5 % · calm
- Gold-0.6 % · calm
- Silver-3.3 % · calm
- Natural gas+7.9 % · calm
- Defence basket-3.2 % · calm
- S&P 500+1.5 % · calm
- Bitcoin-3.0 % · calm
- ^N225+5.4 % · elevated
Register hot. The sensor reads three things side by side: what is said, what is bet, what is done.
Regime: unpriced geopolitics
Since 2026-10-08. Before that: unpriced geopolitics, broad stress. The regime is the sensor's one-line verdict on the whole picture; it changes when the sensors do, not on schedule.
One past signal
- the sensor said
- Macro tipped without a story or a kinetic trigger: VIX at panic level, MOVE above 120, credit stress. Verdict: financial stress without a trigger.
- what followed
- A recovery within days, after the tariff announcement was rolled back; the volatility sensors normalised.
The verdict, and one that can be checked
Where the market stands right now
unpriced geopolitics
Held since 2026-10-08. One state for the whole board, formed out of the four sensors rather than out of any single price.
- unpriced geopolitics2026-10-06 to 2026-10-08
- broad stress2026-10-08 to 2026-10-08
- unpriced geopolitics2026-10-08 to now
A past reading, and what happened next
A verdict is only worth something if it can be wrong. This one is closed and dated, so it can be held against the record.
2025-04-07 · VIX/MOVE
The macro sensor flipped with no narrative and no kinetic trigger: VIX at panic level (45+), MOVE above 120, credit stress. Verdict: financial stress with no trigger.
A few days later the market recovered, after the tariff announcement was rolled back; the volatility sensors returned to normal.
The shape of each asset class, right now
The topology sensor does not look at prices, it looks at the shape the prices make. Over a sliding window of 50 points it builds a persistence landscape for each class and measures how much structure is in it. A low number means the class is moving smoothly; a high one means it is folded.
- commodities0.050
- crypto0.175
- defence stocks0.467
- equities0.202
- ETFs0.029
- FX0.022
L1 norm of the persistence landscape, 50-point window. The spread is the point: FX at 0.022 against ETFs at 0.029 is a factor of thirty between the smoothest and the most folded class.
Two machines stand behind this. One holds the data and the bets: daily candles, instruments, predictions with a date. The other runs the sensors and forms the verdict. They are separate on purpose, so that a change to how the market is judged cannot quietly rewrite what was recorded.