Transition: the engine's licence to say nothing yet.

The state

Not a weak reading — an honest refusal to resolve one.

Transition is produced when the drift comparison does not yield a clean conflict type, because the two metrics are moving in ways that do not settle into agreement, divergence, or a leading relationship. It is easy to misfile this as a mild version of a contradiction, a sort of contradiction-in-progress. It is not. The other states are claims about the relationship between efficiency and monetisation; Transition is a statement that the relationship is currently changing and that any claim made now would be a claim about a midpoint. The distinction matters because it changes what a subsequent reading means: a Transition followed by Under-Monetisation is a resolution, and treating the Transition as a weak Under-Monetisation would have made the sequence look like a confirmation of something that was never asserted.

The playbook

Monitor, count, and specifically do not repair.

The instruction is narrow and worth following literally. Monitor: the state goes into the weekly history strip and is watched into the following weeks, because a transition's meaning is entirely in what it becomes. Count: the summary counters tally transition weeks alongside every other state, and a system producing frequent transitions is telling the operator something about its own stability even when no individual transition resolves into anything. Do not repair: this is where the state is most often mishandled. An operator who reads Transition and begins adjusting exits or selectivity is intervening in a system mid-change, which guarantees that the next reading describes a system that has been altered rather than one that resolved. The intervention destroys the evidence it was meant to respond to.

• A transition's information content is in its resolution, and intervening removes the resolution.
• Frequent transitions are themselves a finding, visible only through the counter and not through any single week.
• Waiting is an action here, not the absence of one — it is what preserves the next reading's meaning.

The cost of allowing it

An engine permitted to decline will sometimes decline when you need an answer.

The design tradeoff is real. A classifier forced to always produce a definite state would give the operator something to act on every week, and the price of that would be occasionally confident wrong states manufactured out of genuinely ambiguous data. MARS takes the opposite trade throughout — blanks rather than zeros in the SDE's maturity rule, a QA flag that marks a regime classification provisional, a stability layer that declares a reading untrustworthy before any table interprets it. Transition is the same principle in the contradiction layer. The cost is that a difficult period can produce several consecutive weeks in which the engine's honest answer is that it does not yet know, arriving exactly when the operator most wants a verdict.

Against the leading states

Transition and the leading states both describe change, and they are not the same.

RAER Leading and RAPF Leading also describe systems in motion, and they are distinguishable from Transition in a specific way: they identify which metric is moving ahead, which makes them directional readings rather than indeterminate ones. RAER Leading says efficiency is improving ahead of monetisation, which is a normal and generally constructive pattern as execution improvements work their way through to results. Transition says the pattern is not identifiable at all. An operator who collapses the three into 'something is changing' loses the most useful part — the leading states carry an expectation about what should follow, and a leading state that fails to resolve as expected is itself a finding.

The key idea

The willingness to publish 'not yet' is what makes the other states credible.

If the engine reported a definite state every week, no state would carry much weight, because the operator would know that some proportion of them were manufactured out of ambiguity. Reserving a label for genuine indeterminacy means that when a named state does appear, it appeared because the data supported it. The refusal is not a gap in the engine's coverage. It is what the coverage is worth.