Structural Diagnostic
Signal Coherence Index (SCI)
“Do the system’s signals agree with each other?”
The Signal Coherence Index is a weighted agreement score: it checks whether the major parts of the trading system are all telling the same story. Think of several doctors reviewing one patient — one reads long-term health, one reads current symptoms, one checks whether the test results agree, one checks for contradictions — and the SCI combines those opinions into a single number. Critically, it measures diagnostic AGREEMENT, not quality: it does not measure profitability, drawdown, risk, or expectancy directly. It measures whether those diagnostic areas are consistent with one another — which means a system can be coherently strong, or coherently bad. All metrics agreeing the system is weak is still high coherence; the score expresses confidence in the diagnosis, never whether the diagnosis is positive.
The formula, derived
Each component is a 0–1 helper score from the SDE pipeline, and the result is rounded to two decimals: 0.00 reads as highly fragmented, 0.50 as mixed or partially aligned, 1.00 as fully coherent. The weights are doctrine, not tunable knobs — they descend from the metric's main purpose (agreement) toward its noisiest input (current-period data).
How MARS reads it
Technical explanation
Alignment carries 40% because it is the metric's main purpose: are RAER, live RAPF, EV, hit rate, and risk-deployment quality broadly pointing in the same direction? EV strong with RAER strong, RAPF strong, hit rate reliable, and deployment appropriate is a coherent picture. EV strong while RAER is weak, RAPF is deteriorating, hit rate is unstable, and deployment is aggressive is exactly the split the Alignment component exists to detect — one part says the edge is healthy while several others say the system is not converting it.
Divergence Control carries 25% and measures the gap between the system's structural condition and its current tactical condition. Structure reflects slower, established evidence; tactics reflect what is happening now. A small gap means live performance agrees with the longer-term record; a large gap means the two are telling different stories. The weight is deliberately moderate because some short-term disagreement is normal — a strong system can have a weak week, and a weak system can have a lucky one. This component is what distinguishes temporary noise from genuine confirmation.
Structural Condition carries 20% and answers what the deeper, longer-horizon evidence says — typically RAER and benchmark RAPF, the underlying quality of risk conversion and monetization. A strong structural score means the performance foundation is solid; a weak one means current results may not be trustworthy even when recent PnL looks good.
Tactical Condition carries 15% and reads live or recent-period measures — live RAPF, EV, hit rate. It gets the smallest weight because tactical data is the noisiest input: the design deliberately prevents one strong week or one weak week from dominating the total interpretation.
Worked example: Alignment 0.90, Divergence Control 0.85, Structural 0.80, Tactical 0.88 → (0.90×0.40) + (0.85×0.25) + (0.80×0.20) + (0.88×0.15) = 0.36 + 0.2125 + 0.16 + 0.132 = 0.8645 → SCI = 0.86. Reading: the major metrics are strongly aligned, and the diagnosis deserves high confidence.
The metric exists to stop decisions being made from one attractive number. A trader who sees high EV and assumes everything is healthy may be looking at weak RAER, deteriorating RAPF, excessive deployment, and an unstable hit rate. The SCI forces MARS to ask whether the supporting metrics confirm the headline metric — and a sagging SCI flags internal disagreement before any single metric breaches its own threshold.
Interpretation bands
Strong. Neutral. Weak.
Strong
Metrics agree across windows — the machine’s state is knowable and posture decisions are well-founded.
Neutral
Partial disagreement consistent with transition — hold posture, increase review frequency.
Weak
Persistent contradiction across the spine — no confident risk decision is justified until the story resolves.
Use cases
Where it earns its place
- ▸Gating confidence in the weekly diagnosis: a strong scorecard read with a weak SCI is treated as provisional until the signals reconcile
- ▸Catching the classic headline trap — strong EV masking weak conversion, deteriorating monetization, or aggressive deployment
- ▸Distinguishing temporary tactical noise from genuine structural confirmation via the divergence component
- ▸Trend-watching coherence itself: an SCI decaying across review cycles is an early warning that the system's signals are coming apart
- ▸Prioritizing review time — low coherence tells the operator WHERE to look (the disagreeing signals) before anything breaches
Edge cases
Where it can mislead
- !Coherently bad: all signals agreeing the system is weak produces a HIGH SCI — the score expresses diagnostic confidence, so it must always be read alongside the condition metrics themselves, never as a health grade.
- !High coherence from shared inputs: components that lean on overlapping evidence can agree for structural reasons rather than informational ones — prolonged near-1.00 readings deserve an input-independence check.
- !Mixed-zone ambiguity: readings near 0.50 mean partial alignment, not mediocrity — the correct response is decomposition into the four components to find WHICH signals disagree, not averaging the ambiguity away.
- !Transition lag: after a genuine regime change, structure updates slower than tactics, so SCI temporarily sags even when the system is adapting correctly — the divergence component is doing its job, not misfiring.
Example scenarios
The metric in the wild
Four green, one red
EV, RAPF, RAER, and ACCEL all read constructive while DD deteriorates. Coherence flags the split: growth quality is real but bought with expanding risk — the exact pattern that precedes lock events.
Monte Carlo connection
Benchmark envelopes assume a coherent system. Incoherent live signals mean live behavior cannot yet be cleanly classified against the simulation — resolve coherence before trusting band comparisons.
Monte Carlo Lab →Live benchmark comparison
Coherence is the gatekeeper for benchmark attribution: only a coherent system’s outperformance can be confidently labeled alpha.
7-Tier MC Benchmark →Go deeper
Operator briefs on this territory.
Deep dive — 01
Expectancy arithmetic: what a trade is worth before it happens.
The EV formula walked end to end — why per-trade worth beats realized P&L as the master metric.
Read the full brief →
Deep dive — 02
Reading the fan: Monte Carlo without self-deception.
What the percentile bands actually promise, why P10 governs sizing, and where simulation authority ends.
Read the full brief →
Deep dive — 03
MAE/MFE: what your trades did while you weren't looking.
Reading maximum adverse and favorable excursion to audit stops, targets, and exits with evidence.
Read the full brief →
SCI is calculated for you — automatically.
Every reading on this page is produced, tracked, and interpreted inside the MARS workbook ecosystem.