Deskscale
Methodology

How the read is constructed

Every state Deskscale publishes has an input, a weighting, and a timestamp. This page shows the architecture and, more usefully, the boundaries. A method that never states its limits is a claim rather than a method.

Analytics and research. Not investment advice.

The architecture

Seven inputs, one read

Market data enters the top and a priority read leaves the bottom. Each layer answers one question well and is honest about the questions it cannot answer at all.

Market regime

What kind of day

Measures

Whether the session is trending, balanced or compressed, and how consistently that state has held.

Cannot tell you

Direction. A trend regime says moves tend to run; it does not say which way.

Price structure

Where price sits

Measures

Position against value, prior ranges, and the levels the tape has actually respected.

Cannot tell you

Timing. A level matters until it doesn't, and structure alone carries no clock.

Volatility

How wide, and changing

Measures

Realised range against implied, and whether conditions are expanding or contracting.

Cannot tell you

Which way a widening range resolves. Expansion is a statement about size, not sign.

Momentum

Rate and breadth

Measures

Rate of change, and whether participation is broadening or narrowing beneath it.

Cannot tell you

Exhaustion from continuation. The same reading precedes both, which is why it is never read alone.

Institutional activity

Where size traded

Measures

Block and dark-pool participation: how much size is being transacted away from the lit tape, and where it is concentrated.

Cannot tell you

The aggressor. Participation magnitude records that size traded, never which side initiated it, and inferring direction from it is a mistake the read does not make.

Options positioning

What dealers must do

Measures

Gamma and dealer positioning, where flow is concentrated, and whether dealers dampen moves or amplify them.

Cannot tell you

Intent. Large flow can be a directional bet or a hedge against something you cannot see.

Risk context

What could break it

Measures

Scheduled events, correlation, and deterioration in breadth or credit ahead of price.

Cannot tell you

The unscheduled shock. Nothing prices what has not been announced, and no method claims otherwise.

CORE · Priority read

The synthesis

Produces

A ranked read of what deserves attention, with the chain above it visible and the level at which it is wrong.

Is not

A recommendation to you. It is analysis output; what you do about it is your decision.

Market data in The same seven layers, running
Priority read out Source · window · timestamp
The weighting

Six answers rarely agree

The hard part is not collecting the layers. It is deciding what to do when structure says one thing and positioning says another, which is most days.

Evidence-weighted

Sources are weighted by what they are worth under current conditions, not averaged equally. A layer that has been unreliable in this regime is discounted rather than allowed to vote at full strength.

Probability, not verdict

Conviction is expressed as a degree, never as a binary. A read at low confidence is published as a read at low confidence, rather than rounded into certainty for the sake of a cleaner interface.

Recalibrated

Weightings update as outcomes land, session after session. The method is not a fixed rulebook; it is a system that gets better at weighing evidence the longer it operates.

What is always shown

Three things every read carries

Confidence
Stated, not implied

Every read says how sure it is. A confident read and a marginal one look different, because presenting them identically would be the single most misleading thing an instrument could do.

Staleness
Flagged, not hidden

Data has an age. When a feed lags or a source is two days behind, the read says so rather than serving it as though it were live. Source, window and timestamp sit on every figure.

Invalidation
The level it breaks

Every read states what would make it false. It is the difference between analysis you can hold to account and an opinion you cannot.

The boundaries

What this method cannot do

Stated plainly, because a reader who finds these out later will trust nothing else on the site, and because they are true.

  • It does not forecast price. Deskscale classifies conditions. It does not predict where a market closes, and any read framed as a forecast is a read being misused.
  • It does not know your position, size, or risk tolerance. Nothing here is personalised, which means nothing here can be suitable for you specifically. That judgement is yours and cannot be outsourced.
  • It cannot price the unscheduled. A surprise headline, a halt, a liquidity event: no model weighs what has not happened yet.

None of this is hedging. An instrument that claims no error bars is not more accurate, it is less honest, and a reader who trades real size already knows that.

See it on a live tape

The method is easier to judge running than described.