Measuring the Complete Selection Process
August 2026 · Accountability 2.2 · Selection Intelligence
A model can fail a test because the model is weak. It can also appear to fail because the test measures only part of what the system was designed to do. AMAAS encountered the second possibility when its first accountability experiment evaluated securities primarily through the Composite Score while omitting two important dimensions of the selection process: Momentum and Regime-Aware evidence.
From Principle to Evidence
Chapter Two introduced the AMAAS Evidence Loop: preserve the prediction, collect independent evidence, measure the outcome, and allow the evidence to challenge the model. At that point, only one trading day of realized evidence existed. The purpose of that first observation was not to declare the model right or wrong after a single day. It was to demonstrate something more fundamental: the prediction could be frozen, the subsequent market evidence could be collected independently, and the result could be preserved without rewriting history.
One week later, the question changes. There is still far too little evidence to claim persistent predictive performance, but there is now enough realized market behavior to begin examining the selection process itself. Instead of asking only whether the accountability machinery works, AMAAS can begin asking which signals are contributing information and whether different selection methods identify meaningfully different opportunity sets.
The First Test Raised the Right Question
The original accountability framework grouped companies by Composite Score and measured subsequent realized returns. That was a legitimate test: did a broad cross-sectional assessment of company attractiveness independently discriminate future outcomes? The early evidence was not encouraging enough to claim that higher Composite Scores alone were producing better short-horizon results.
The appropriate response was not to rewrite historical observations or modify the model until the result looked better. The original experiment remains frozen. AMAAS instead added a second experiment: does the complete AMAAS selection process contain information that the Composite Score alone does not?
Three Different Questions
Composite, Momentum, and Regime are not three versions of the same ranking. They represent different forms of evidence.
| Signal | Question | Role |
|---|---|---|
| Composite | Which companies appear most attractive across the broader financial and quantitative evidence? | Cross-sectional company attractiveness |
| Momentum | Which securities are demonstrating the strongest market behavior? | Observed price behavior and market confirmation |
| Regime | Given the security's current volatility-adjusted state, what does its historical transition structure imply about the state likely to follow? | Conditional market-state behavior |
Composite: Broad Analytical Evidence
The Composite Score evaluates multiple dimensions of company evidence rather than relying on one valuation ratio, forecast, or accounting statistic. AMAAS combines financial quality, valuation, probability-based expectations, liquidity, profitability, solvency, and other quantitative evidence into a consistent cross-sectional comparison framework. Its purpose is not to suggest that one number represents intrinsic truth; it helps compare thousands of companies on a common analytical basis.
That distinction matters for accountability. Composite is primarily a measure of company attractiveness according to the analytical evidence set. It is not intended to be a pure short-term price-timing indicator.
Momentum: What the Market Is Rewarding
Momentum contributes a different kind of information. Rather than asking whether financial and probabilistic characteristics appear attractive, it measures the strength of observed market behavior. A company can appear financially attractive while its security is being repriced downward; another can have less compelling static characteristics while the market strongly rewards it. AMAAS preserves that disagreement because the disagreement itself may be informative.
Regime: What State May Come Next?
AMAAS Regime analysis uses a deterministic five-state, first-order Markov framework based on point-in-time price history. Daily log returns are normalized relative to the security's own return distribution and classified into five volatility-scaled states: Strong Down, Down, Neutral, Up, and Strong Up.
| Standardized Return | State | Score |
|---|---|---|
| z ≤ -1.25 | Strong Down | 0.00 |
| -1.25 < z ≤ -0.35 | Down | 0.25 |
| -0.35 < z < 0.35 | Neutral | 0.50 |
| 0.35 ≤ z < 1.25 | Up | 0.75 |
| z ≥ 1.25 | Strong Up | 1.00 |
AMAAS estimates how the security historically transitioned from each state to every other state. Given the current state, the resulting transition probabilities form a probability distribution for the next state. The Regime Score is the probability-weighted expected state:
A Regime Score is therefore not the probability that the stock will rise. It is an expected-state measure on a zero-to-one scale. AMAAS separately retains upside, downside, neutral, persistence, and expected-state information.
The model normally examines up to approximately one trading year of price history and requires a minimum number of observations. A robust median/MAD volatility estimate is used where possible, and a small smoothing prior helps avoid zero-probability transition estimates. These choices improve stability but do not eliminate sampling uncertainty.
Accountability 2.2: Selection Intelligence
The expanded test compares four Top 20 cohorts against the same realized-growth evidence. It does not replace Accountability 2.1. It asks whether signals intentionally used elsewhere in AMAAS identify different opportunity sets.
| Selection Method | N | Average Growth | Median Growth | Positive | Min | Max |
|---|---|---|---|---|---|---|
| Composite Top 20 | 20 | +0.46% | -0.55% | 45.0% | -5.19% | +11.12% |
| Momentum Top 20 | 20 | +9.53% | +7.46% | 95.0% | -0.15% | +26.38% |
| Regime Top 20 | 20 | +7.49% | +3.94% | 85.0% | -1.20% | +26.38% |
| Consensus Top 20 | 20 | +7.08% | +3.76% | 90.0% | -1.20% | +26.38% |
Initial Accountability 2.2 observation: 62 eligible Top 100 observations from the Aug. 3, 2026 research dataset measured against the Aug. 11, 2026 growth snapshot. The $35 Primary Research Universe eligibility rule is applied before comparison. These are early short-horizon observations, not evidence of persistent predictive performance.
What the Initial Evidence Says
The difference is large enough to deserve attention. Composite Top 20 produced average growth of +0.46%, a -0.55% median, and 45% positive observations. Momentum Top 20 produced +9.53% average growth, a +7.46% median, and a 95% positive rate. Regime Top 20 produced +7.49% average growth and an 85% positive rate. Consensus produced +7.08% average growth and a 90% positive rate.
The medians are especially useful because they make it harder for one exceptional winner to explain the result. Momentum's +7.46% median and Regime's +3.94% median suggest that the observed advantage was distributed across their cohorts rather than solely produced by one extreme security.
These observations are not proof of persistent predictive value. The horizon is short, each cohort contains only twenty securities, the methods can select overlapping companies, and one market environment may favor particular signals. Those limitations are why the experiment must continue.
What Changed — and What Did Not
| Preserved | Added |
|---|---|
| Original Accountability 2.1 observations | Momentum Top 20 cohort |
| Composite decile diagnostics | Regime Top 20 cohort |
| Historical realized-growth evidence | Multi-signal Consensus cohort |
| $35 Primary Research Universe rule | Side-by-side selection-intelligence comparison |
From Ranking to Decision Support
AMAAS is designed as a sequence of evidence rather than a single stock ranking. A user can discover companies through the broad analytical universe, inspect Composite evidence, examine Momentum and Regime behavior, compare securities, and construct a personalized Top 20 based on the factors the user considers important. Those securities can then be transferred into the Portfolio Simulator for continued observation before capital is committed.
A security does not stop changing after it appears on a list. AMAAS can continue to evaluate price behavior, changing market regimes, macroeconomic context, and event-driven discontinuities through jump-diffusion and event intelligence. The resulting workflow is more faithful to the actual decision process:
The Lesson
Accountability is not merely measuring whether a model was right. It also requires asking whether the experiment measures the system being evaluated. The first AMAAS accountability results challenged the Composite Score. The correct response was to preserve those results and investigate the hypothesis they exposed.
Composite identifies what appears attractive. Momentum identifies what the market is rewarding. Regime estimates what price state is likely to follow, conditional on the state the security occupies today.
Accountability 2.2 now asks whether those distinct forms of evidence contribute different information over time. The early results are encouraging, particularly for Momentum and Regime, but the experiment has only begun. The meaningful test is whether the relationships persist across longer horizons, new datasets, different market environments, and securities outside the initial observation.
The Evidence Is Still Accumulating
Chapter Two captured the first one-day observation. This chapter returns after approximately one week. Neither horizon is long enough to establish durable predictive value. Their importance is that they create a sequence of observations against predictions that remain unchanged.
The next accountability milestone will extend that same experiment to approximately one month of realized evidence. The conclusion is deliberately unwritten. Momentum may continue to lead. Composite may strengthen. Regime or Consensus may become more informative. The apparent relationships may weaken or disappear. Whatever the market produces becomes the next piece of evidence.
That is the larger purpose of AMAAS 2.0: not to build a system that can always explain why it was right, but to build one that remembers what it predicted, observes what happened, identifies where its assumptions were incomplete, and becomes more useful without rewriting its own history.