Current EV and rolling EV are separate claims, and both have to hold.

Why four numbers

One expectancy figure cannot say whether edge is recent, persistent, or fading.

A single expectancy reading is a point estimate with no temporal information attached, and the three questions an operator actually needs answered are all temporal. Is this edge new? Has it been there throughout? Is it on its way out? The windowed views answer these by construction: the current reading captures the most recent state, the twelve-week view is the slowest and least excitable, and the four- and six-week views sit between them where change becomes visible first. Reading them together converts a number into a trajectory. The engine's stated purpose for this block is exactly that — it shows whether edge is recent, persistent, or fading — and those three descriptions carry three different deployment implications despite being compatible with the same current figure. A branch whose edge is persistent has earned a claim on the risk budget. A branch whose edge is recent has earned observation.

Agreement

The windows agreeing is a stronger statement than any of them individually.

When current, four-week, six-week and twelve-week expectancy all read positive, the claim being made is not merely that the branch is profitable — it is that the branch has been profitable across overlapping samples of different lengths, which is a far harder condition to satisfy by chance. Short windows are noisy and long windows are slow, so a favourable stretch that would dominate a four-week reading is diluted in the twelve-week view, and a genuine structural edge is one of the few things that survives both. This is why agreement carries more weight than magnitude. A branch reading modest expectancy consistently across every window is a better deployment candidate than a branch reading twice that figure in the current window and nothing in the twelve-week, because the first is describing a property of the branch and the second may be describing a property of the last fortnight.

Disagreement

The direction of the divergence is the diagnosis.

Divergence between windows is not a defect in the data; it is the most information the block produces, and it points in a specific direction depending on its shape. Current strong with rolling weak is the recency pattern — the branch has had a good run against an unremarkable base, and the honest reading is that nothing has yet been established. Current weak with rolling strong is the opposite and demands a different response: the branch has a demonstrated history and is currently underperforming it, which is a question about whether conditions changed or whether this is ordinary variance around a real edge. Long windows decaying while short windows hold is the most serious of the three, because it describes an edge that is being consumed slowly enough to remain invisible in any single window. In each case the pattern names the next question, which is more than a single expectancy figure has ever been able to do.

In deployment

The engine is a confidence input, and confidence is not authorisation.

The instruction attached to this tab is deliberately restrictive: it is used as a risk-confidence input rather than as a licence to override the trade plan. The distinction matters because a strong window stack reads like permission and is not. What the block establishes is how much weight the branch's expectancy deserves in a decision — whether the number should be treated as a property of the branch or as an observation about a period. That weighting then enters a decision that is governed elsewhere, by the gate, the throttle, and the plan's own rules about which branches are eligible under which conditions. An operator who reads four agreeing windows and increases exposure on that basis alone has used a measurement of confidence as though it were a measurement of capacity, and those are different quantities held by different parts of the system.

The honest failure

Recency is not a bias the operator can decide not to have.

The reason this block exists in a workbook rather than as advice is that the failure it prevents is not a knowledge failure. Every operator knows in the abstract that a good fortnight is not a new edge. The difficulty is that recent evidence is more vivid, more emotionally available, and arrives attached to memories of specific trades, while a twelve-week average arrives as a number in a cell. Under that asymmetry, intention loses reliably. Rendering all four windows side by side removes the need to remember to check, because the divergence is visible in the same glance as the current reading. This is the same structural logic that runs through the rest of the panel — a protection that depends on the operator recalling a principle at the moment of temptation is not a protection, and the fix is to put the counter-evidence in the same field of view as the temptation.

• Windows agreeing: a property of the branch, and it can carry weight.
• Current strong, rolling weak: a property of the period, and it carries none yet.
• Long windows decaying under holding short windows: the pattern that costs most.

The key idea

The block turns a number into a claim with a stated confidence.

Every expectancy figure is implicitly a claim about the future, and claims without confidence attached are unusable for deciding how much capital to commit. The window stack attaches it. What the operator ends up with is not a better estimate of expectancy — the current reading is already the best estimate of the current state — but a defensible statement about how much that estimate should be trusted, which is the quantity that ought to govern the size of the bet placed on it. Confidence and magnitude are different axes, and the persistent error in discretionary allocation is reading a large number as a confident one.