Meaning of the Lock

This document fixes every methodological degree of freedom before a single empirical case is scored. Once locked, no parameter, definition, threshold, or hypothesis below may be altered during case work. Any change requires a new version (v2, v3, …), dated and justified, and applies only to cases scored after that new version. Cases already scored under v1 are never re-scored under a later version. This rule is the sole protection against hindsight-driven parameter drift.

Parameters below are tagged [PRINCIPAL] (confirmed by the author as a research commitment) or [DEFAULT] (set by the methodology at lock time; changeable only by versioned amendment, like any other locked value).

1. Purpose and Scope

This protocol tests one claim of the Theory of Asymmetric Capital Accumulation (TACA):

Asymmetry emerges when structural reality evolves faster than collective interpretation.

It operationalizes that claim as a measurable, point-in-time divergence between what the market prices and what a locked evidence model implies, and specifies how that divergence is to be validated against realized outcomes.

This protocol does NOT claim:

·         to measure a Structural Probability Transition directly (SPTs are latent);

·         predictive validity in advance of the empirical test;

·         that any single case validates the theory.

It claims only that the methodology can be executed consistently, auditably, and point-in-time, and that its central hypotheses are falsifiable on a pre-registered sample.

2. Canonical Definitions (Name-Collision Resolution)

Within this protocol and all cases scored under it, "CRF" has exactly one meaning:

CRF (Cognitive Risk Friction) = an information-theoretic divergence between a market-perceived outcome distribution and an evidence-model outcome distribution, computed at a fixed point in time.

The doctrinal "Cognitive Reflexivity Function" (epistemic-legitimacy construct, declared orthogonal to probability distributions) is explicitly excluded as an operational definition here. It may be cited only as qualitative motivation, never as the measured quantity. This resolves the corpus-wide collision in which "CRF" carried incompatible meanings.

·         X — the locked outcome space (Section 4).

·         Q_t — the market-perceived distribution over X at time t (Section 5).

·         P_t — the evidence-model distribution over X at time t (Section 6).

·         R — the realized outcome, observed only at t+h (Section 8).

3. The Metric

CRF_t = JSD( Q_t || P_t )

·         JSD = Jensen–Shannon divergence, log base 2, range [0, 1]. [DEFAULT]

·         Chosen over KL because JSD is symmetric, always finite (even with disjoint support), and bounded — hence comparable across cases. [DEFAULT]

4. Outcome Space X (LOCKED)

·         Variable: forward total return of the object over horizon h = 24 months. [PRINCIPAL]

·         Binning (return space), 10 bins: [DEFAULT]

(-100%,-50%], (-50%,-25%], (-25%,-10%], (-10%,0%], (0%,10%], (10%,25%], (25%,50%], (50%,100%], (100%,200%], (200%,+inf)

Bins are deliberately asymmetric to preserve right-tail resolution (the theory lives in the tail; symmetric bins would blur it).

·         All three distributions (Q_t, P_t, R-as-point) are expressed over these identical bins.

5. Q_t — Market-Perceived Distribution (LOCKED)

Q_t is read from the market, never constructed by the analyst. Source precedence: [DEFAULT]

1.       Option-implied risk-neutral density (Breeden–Litzenberger), when listed options exist with maturities bracketing h and open interest above the locked liquidity floor. The RND is mapped onto the Section-4 bins.

2.       Fallback — analyst target distribution: if options are insufficient, use the cross-section of 24-month analyst price targets (consensus + dispersion) fitted to the same bins.

3.       If neither exists: the case is flagged Q-unmeasurable and excluded from H1/H2 (logged, not silently dropped — Section 13).

6. P_t — Evidence-Model Distribution (LOCKED): Reference-Class Method

P_t is the smoothed empirical histogram of the forward-h returns of the k historical analogues most similar to the object, by a locked structural feature vector. P_t is the output of a locked computation, not the analyst’s opinion. Six locked components:

6.1 Feature set f (declared once, program-wide) — 6 features [PRINCIPAL]

No per-case feature selection. Ambiguity in any feature is a leakage-log item (Section 10).

6.2 Similarity metric & class size [DEFAULT]

Features z-scored on the pre-t cross-section; distance = Mahalanobis using the pre-t cross-sectional covariance; take the k = 75 nearest analogues. [PRINCIPAL: k=75]

6.3 Strict point-in-time rule [DEFAULT]

An analogue qualifies only if its own forward-h (24m) window closed before t. The feature→outcome relationship is therefore entirely historical relative to t. No future information enters P_t.

6.4 Survivorship-free universe [PRINCIPAL]

Analogues drawn from US listed equities meeting the locked size/data filters, including delisted, bankrupt, and acquired firms. A survivorship-biased universe would poison the left tail of P_t — the very bias the theory claims to fight.

6.5 Binning & smoothing [DEFAULT]

Analogue returns binned per Section 4; Dirichlet (add-α) smoothing with α = 0.5 (Jeffreys) so empty bins do not destabilize small-sample estimates.

6.6 Measurability floor [DEFAULT]

Require at least n_min = 40 qualifying analogues. If fewer, the case is flagged out-of-measurable-domain; P_t is not constructed and the case is excluded from H1/H2 (logged, Section 13). Novel structural transitions with no analogue base are declared unmeasurable rather than fabricated.

6.7 Robustness cross-check [DEFAULT]

An alternative P_t′ from a locked quantile regression on the same f and same pre-t data is computed; JSD(P_t || P_t′) is reported as a model-dependence diagnostic. Large disagreement flags low confidence for that case.

7. Direction / Quadrant Rule (LOCKED) [DEFAULT]

JSD is unsigned. Direction is assigned by comparing P_t and Q_t and their shift over the trailing window [t−12m, t]:

"Higher/lower" = comparison of distribution central tendency (median bin). "Moved" = shift in central tendency over the trailing window beyond one bin.

8. Three-Layer Evaluation (LOCKED)

·         Layer 1 — Detection (at t, observable): CRF_t = JSD(Q_t||P_t); quadrant per Section 7.

·         Layer 2 — Validation (at t+h): realized R observed. Compute L_Q = D(R||Q_t), L_P = D(R||P_t), and Gap = L_Q − L_P (D = KL; R as the realized-bin indicator with the same α-smoothing).

·         Layer 3 — Payoff link (at t+h): record whether realized R landed in the right-tail bins (top 2 bins) in the direction the quadrant predicted.

9. Pre-Registered Hypotheses & Decision Rules (LOCKED) [DEFAULT thresholds]

Define high-CRF = CRF_t in the top tercile of the scored sample. Define positive-direction = True-positive-asymmetry quadrant.

H1 (cognitive-lag claim):

Among high-CRF positive-direction cases, the share with Gap > 0 exceeds 0.5 (one-sided binomial test, α = 0.05). Refutation: if high CRF does not predict that the evidence model beat the market, the cognitive-lag thesis fails.

H2 (asymmetric-payoff claim):

High-CRF positive-direction cases realize right-tail outcomes (Layer 3) at a higher rate than a matched low-CRF control (two-proportion test, α = 0.05). Refutation: if not, CRF does not locate asymmetry.

H0 (null):

CRF_t has no relationship to Gap or to realized asymmetry.

Minimum sample before any H1/H2 inference: N ≥ 60 scored cases spanning all four quadrants. [DEFAULT] No inference is reported on fewer.

10. Controls (LOCKED)

·         Independent dual scoring: every case scored by ≥ 2 independent evaluators who do not deliberate. Agreement metric = JSD(P_t^A || P_t^B). Tolerance τ = 0.10. Disagreement beyond τ = protocol failure for that case (flagged, not reconciled by discussion — reconciliation would contaminate independence on later cases). [PRINCIPAL: dual scoring; DEFAULT: τ=0.10]

·         Author exclusion: the theory’s author may not be a sole scorer of any case; if participating, that score is flagged.

·         Leakage Control Log (per major score): (a) Could this be influenced by future knowledge? (b) Would it survive if the object had later failed? (c) What contemporaneous evidence alone supports it? Logged into the audit record.

·         Blinding (preferred where feasible): present the contemporaneous evidence dossier de-identified (name redacted, dates shifted) so the scorer does not know the object. Where the object is too famous to blind, this is recorded as a known contamination risk and is an argument for preferring obscure cases (Section 11).

11. Case-Selection Frame (LOCKED) [DEFAULT]

Cases are sampled from a pre-registered frame, never cherry-picked by known outcome.

·         Frame: firm-quarters from the Section-6.4 universe over 1995–2020 (so that the t+24m window has fully resolved and predates the lock date).

·         Sampling: stratified random draw designed to populate all four quadrants; selection uses only point-in-time data available at each candidate’s t.

·         Outcome valence (winner/loser/false) must not enter selection. Famous cases are permitted but flagged for contamination risk; obscure cases are preferred for clean process tests.

12. Forbidden Practices (LOCKED)

4.       No hindsight numbers. P_t comes from a locked model on pre-t data — never assigned by recollection of the outcome (forbids the historical-CRF-table practice).

5.       No hand-assigned probabilities or price targets as drivers of the metric. Q from market, P from model (forbids the ARE 50/50 example practice).

6.       No hidden aggregation. CRF is one defined operator (JSD); no multiplicative index of unitless quantities (forbids the AXI-PRO practice).

7.       One CRF definition (Section 2). The doctrinal/orthogonal definition is barred as the operational quantity.

8.       No silent caps. Every excluded case (Q-unmeasurable, out-of-domain, τ-failure) is logged with reason; coverage is reported as "M of N candidate cases."

9.       No mid-study parameter change. Section 0 governs amendment.

13. Measurable-Domain Boundary (Honest Limitation)

The most transformative asymmetries (genuinely novel structural transitions) are precisely where both Q_t (no liquid options) and P_t (no analogue base) are hardest to build. The methodology is therefore most reliable on moderate, recurring asymmetries and is silent outside its measurable domain. Results pertain only to cases with sufficient Q-liquidity and ≥ n_min analogues. This boundary is reported, not hidden. CRF measures divergence from the evidence model, not from truth; all conclusions are conditional on the model being a reasonable, inspectable evidence aggregator.

14. Role of Amazon 2004–2005

Amazon is a Protocol Execution / Feasibility Case, not a validation case and not a calibration (parameter-fitting) case. Its sole purpose: demonstrate that the protocol can be executed, that the evidence trail reconstructs, that two evaluators follow the same process. A clean Amazon run validates the process, never the theory. Because Amazon is outcome-loaded and hard to blind, it carries a flagged contamination risk and cannot enter H1/H2 inference.

15. Locked Parameter Appendix

End of TACA Protocol Lock v1. Locked 2026-06-22. Amendments only by dated, versioned re-issue applying to cases scored after re-issue.