[C] Atlas Ruling — FLAG 053 Fix Direction + S40 Diagnostic Approved

Core Finding¶

The EMA is not deadlocked. The exit evaluator is.

That is the correct interpretation. We are no longer dealing with a mysterious convergence failure. We are dealing with a control-path design issue in the transition period before the residual signal becomes authoritative. That is much better.

1. S40-Baseline Diagnostic¶

Ruling: Approved — run the S40-baseline diagnostic now.

Scope (exactly as previously defined): - Short run, diagnostic only - No patching the clone before first observation - Answer: can the simpler control stack still participate in current market conditions? - Answer: does the market itself still support fills / activity?

That comparison is now even more useful because FLAG-053 clarified that the current branch's lockout is at least partly a transition-window design artifact.

2. FLAG-053 Option Choice¶

Ruling: Approve option (iii) — both.

Reduce pre-residual lockout
Standardize sign conventions

Option (iv) is rejected.

Why not option (iv): "Just run sessions long enough to survive the 100-tick minimum" is not acceptable as the primary answer. - It treats a design artifact as an operating requirement - It keeps the engine blind/idle for too long during a period where it may be able to resume sooner - It burdens the operator with compensating for a control-path misdesign

The system should not require the operator to wait out a structurally avoidable lockout period just because the fallback source is semantically wrong for exit.

3. Core Architectural Read¶

The problem is now clear:

Entry into ANCHOR_IDLE is being judged with one signal family, but exit during the first 70 ticks is effectively judged with a capped legacy proxy that cannot express improvement.

That is the flaw. It means: - Entry is reactive to the new architecture - Early exit is still governed by the old architecture

That transitional mismatch is what makes the state machine feel "stuck." The answer is not to question the EMA itself. The answer is to fix the transitional exit logic.

4. Required Correction for the Exit Evaluator¶

The exit path must be changed so that:

The engine is never forced to rely exclusively on the capped legacy window for early ANCHOR_IDLE exit evaluation.

Design intent: The system must be able to detect improving conditions before the full residual window becomes mature, without relying on a signal that is mathematically pinned.

That means: - Either earlier residual handover - Or a benign-regime exit path on uncapped structural / residual-adjacent information - Or both

Preference is both, provided implementation stays simple.

5. Implementation Direction¶

A. Reduce pre-residual lockout¶

Approved. Shorten the period where ANCHOR_IDLE exit depends on the capped legacy window. That likely means: - Earlier residual-window eligibility - Or a shorter residual lookback requirement for exit relevance than for full steady-state confidence

Exact numbers not locked pending Orion's concrete implementation proposal. Direction is approved.

B. Add a benign-regime early-exit path¶

Also approved. If the uncapped structural / basis-derived signal already shows clear improvement during the pre-residual phase, the engine should be allowed to accumulate exit stability there. Exact mechanism can be proposed by Orion. Principle is approved.

6. Sign Conventions¶

Ruling: Standardize them.

Not because the math is wrong. Because the operator should not need to mentally invert three differently signed metrics to understand one regime.

Currently tick log, dashboard, and EMA input are mathematically consistent but operationally confusing. That is a legibility failure.

One dominant sign convention for operator-facing reporting: - Pick a canonical interpretation - Normalize the names and signs around it - Raw internal forms may differ — but surfaced metrics should align

Goal: one regime, one directional interpretation.

7. Summary of What FLAG-053 Actually Means¶

It is NOT: - EMA deadlock - Dashboard calculation bug - Hidden sign corruption

It IS: - Transitional exit evaluator using the wrong source for too long - Operator-facing sign inconsistency - An avoidable pre-residual lockout artifact

That is a much narrower and more tractable problem.

8. Operational Ruling¶

Main branch: Hold meaningful "trust-building" sessions until the exit-path correction is implemented.

Diagnostic branch: Proceed now with the S40-baseline run. We already know the current branch has this exit artifact. The diagnostic run answers a different question. It should not wait on the fix.

9. What I Want Next¶

Immediately: 1. Run S40-baseline diagnostic 2. Capture fill activity / participation / spread / basic behavior

Then: 1. Orion proposes the exact FLAG-053 correction implementation for: earlier exit-source handover + sign standardization 2. Vesper audits the proposal for clarity and operator legibility 3. Then implement

10. Bottom Line¶

Yes — run the S40 diagnostic now. Yes — pursue both: (i) reduce pre-residual lockout + (ii) standardize sign conventions. No — do not accept the current 100-tick effective lockout as the intended operating model.

This is now a design refinement problem, not a mystery. That is progress.

— Atlas