Seven entities: what the SDE is actually allowed to diagnose.

The universe

Four branches that exist, three views that are constructed.

The four primitive entities are the branches themselves: Normal, the core branch and simulation centre, representing static-regime execution and serving as the baseline for stable edge quality; Trend Partial, the hybrid fat-tail accelerator, representing trend participation with partial monetisation; Trend No-Partial, the true fat-tail accelerator, representing trailing behaviour with higher variance and higher acceleration potential; and Overflow, the stability branch, representing supplemental higher-timeframe flow that must be watched closely because its volume and cost profile can distort expectancy. The three composites are not branches at all. They are constructed views: Normal + Trend, which combines the three core branches and deliberately excludes Overflow to read core plan behaviour; Trend Blend, which combines the two trend branches to isolate the trend engine; and All Blended, which combines all four for the total system view.

Why both kinds are needed

Localised diagnosis and blended behaviour are not substitutes.

Branch-level diagnostics identify where strength and weakness actually live. If Trend No-Partial is degrading while Normal holds, that is a specific, addressable finding attached to a specific exit architecture. Composite diagnostics answer a different question: how the blended system is behaving, which is what the operator's equity curve actually experiences. Neither read subsumes the other, and the failure modes run in both directions. A branch in genuine trouble can be invisible at the All Blended level, its deterioration absorbed by three healthier branches — the aggregate stays acceptable while a component rots. Less obviously, the reverse also happens: a composite can weaken while no individual branch does, because the composite reflects the mix as well as the parts, and a shift in how much flow each branch is receiving changes the blend without changing any branch's own behaviour.

• Aggregate health can hide a dying branch — the classic case, and the reason branch-level pipelines exist at all.
• Composite health can shift purely from a change in mix, with every underlying branch unchanged.
• A finding is only actionable once you know which of the two it is.

The exclusion that defines each composite

Normal + Trend is defined by what it leaves out.

The three composites are best understood by their exclusions rather than their contents. Normal + Trend exists specifically to exclude Overflow, because Overflow is supplemental flow with a different cost structure and different volume characteristics, and including it means the reading describes the plan plus its supplement rather than the plan. Trend Blend excludes Normal in order to isolate the trend engine, which is the right view when the question concerns trailing behaviour and fat-tail capture rather than baseline execution. All Blended excludes nothing, which makes it the honest total and also the least diagnostic of the three — it tells the operator what the account experienced without telling them why. Reading All Blended first and stopping there is a recognisable habit, and it is how composite analysis degrades into a second, more elaborate equity curve.

The closed universe

Seven, and not extensible on a whim.

The universe is fixed for the same reason branch identity is fixed at entry: an analytical dimension that can be added mid-stream produces history that cannot be compared with itself. Every SDE pipeline — six tables across five metric families — is instantiated against these entities, and the rolling windows, stability tests, drift comparisons, and z-scores all assume a continuous series for each one. Introducing an eighth entity does not add a column of data; it adds a column with no history, which will read as unstable for as long as its window takes to fill and will contaminate any composite it is folded into. If the universe ever needs to change, that is a deliberate structural revision with a known cost, not a convenience.

The key idea

The entity you choose is half the diagnosis.

Two operators can run identical pipelines on identical data and reach opposite conclusions purely by reading different entities — one looking at All Blended and seeing stability, the other looking at Trend No-Partial and seeing decay, both correct. The entity is not a filter applied to a finding. It is part of the question, and a structural reading that does not state which of the seven it came from has not actually said anything.