A verifiable structural-state diagnostic — and its honest boundary.
CMM.X converts Bitcoin market structure into bounded, non-directional diagnostic statistics, seals each state cryptographically into an append-only chain, and publishes the evidence so anyone can re-derive it. The seven axes describe structure; they state no view on price direction.
What this is — and what it is not
It is
- Seven bounded, non-directional diagnostic axes describing structure (six KEEP, one DEMOTE).
- An append-only, hash-linked state chain co-signed with a hybrid (Ed25519 + ML-DSA-65) signature.
- A reproduce-leg that re-derives derived content from declared inputs, with the honest split shown.
- A pre-registration ledger recording falsifiable, out-of-sample hypotheses over real elapsed time.
- A cross-asset study measuring where breadth helps (faster evidence) and where it does not (pooling hurts).
It is not
- Not a forecast, not a directional call, not a trading signal.
- Not a claim of edge or alpha —
predictive_claim = NONE. - Not investment or financial advice.
- Not a "verified system" — the reproduce number is a fraction of derived-content leaves, stated plainly.
- Not proof of predictive skill — historical walk-forward is mechanism validation, suggestive only.
Glossary — the seven axes (SOLLSTAND-P1 verdicts)
| Axis | Name | Verdict | Load-bearing note |
|---|---|---|---|
| Φ | structural completeness (0..1) | KEEP | max|corr| ≈ 0.14 — distinct & non-degenerate |
| Λ | level / liquidity structure | KEEP | max|corr| ≈ 0.25 |
| Χ | cross-flow / composite | KEEP | distinct but drift-coupled; decoupled into Χ′ (~74% correlation reduction) |
| Μ | momentum / drift | KEEP | M.drift_magnitude max|corr| 0.359 |
| W | wave / cycle state | KEEP | W.phase_state max|corr| 0.112 |
| Ψ | participant stress / crowding | KEEP | Ψ.conviction max|corr| 0.143 (Ψ.leverage_density = NULL_NO_BASIS on this history — reported, not fabricated) |
| Β | balance / bias | DEMOTE | |corr| 0.968 vs W.cycle_state — redundant (re-measures range-position W owns) |
Each axis is a descriptive statistic. Correlations are shown as magnitude only. KEEP / DEMOTE is a redundancy/degeneracy verdict, not a performance claim. "distinct" = max |cross-corr| < 0.90.
What "reproduce" means
A leaf is REPRODUCED when an independent code path, given only the artifact's declared inputs, yields the identical value (float bit-for-bit). Leaves that depend on wall-clock time are integrity-checkable but not replay-deterministic; declared inputs are echoed, not re-derived; constants and metadata are what they are. Where a rule or input is not declared, it is recorded as an honest gap — never fabricated. Both the derived-content fraction (93.06%) and the all-leaves fraction (33.50%) are shown on the Verify page, with the full by-design-non-reproducible split.
What the learning results honestly say
- T1 (SHAPE) — suggestive: beats baselines on log-loss (calibration, not accuracy). Not proof.
- T2 (STRUCTURE_REGIME) — ties climatology on accuracy (the regime is near-constant, so accuracy is baseline-easy); the real win is calibration (log-loss).
- T3 (ENERGY_STRESS) — refuted-degenerate: the a-priori terciles make the target near-constant (HIGH never realized), so no model can express skill. Refutation reported plainly.
- T4 (ELEMENTS) — 22/30 elements beat climatology on log-loss (the strongest signal in the set); still suggestive, not proof. OI-intensity cells (tiny n) lose and are shown.
Cross-asset: pooling into one shared model hurts vs per-asset (log-loss); breadth's real benefit is a ~1.85× correlation-discounted faster-evidence rate, not more skill.
Honesty & classification
The learning, axis, Σ/Ω, cross-asset and reproduce artifacts are labelled RESEARCH_SHADOW NON_AUTHORITATIVE kernel_state_hash=null in their own files, and are surfaced as such here. A predictive claim would require many pre-registered, out-of-sample confirmations accumulated over real elapsed time. None exist yet. Until then: diagnostics + verification only. Every value on this site is read from a real file; genuinely-absent data is shown as DATA_LIMITED with its reason — never invented.