Sources3
Newest value60 min described moment
Oldest value60 min described moment
Without timestamp0
Page refreshesevery 20 s
Tip75473
PhaseNORMAL
Coherence0.568
P(large move)0.0470
The model's own verdict on itself:
WORSE_THAN_PERSISTENCE
It predicts the next state
worse than simply assuming the state stays the same. This is not a caveat we
added — it is the number the system computes about itself, every minute,
and it is published because it is measured.
model accuracy
2869 / 10000 · 28.7 %
Baseline (always the majority class): 0.5325. Lower is worse.
Where it comes from, where it is, how near the next event is
| last regime shift | 2026-09-04T12:00Z · -2.25 % |
| now | 2026-09-11T03:57Z · 76835.7 |
| in sync | YES |
| sync wave phase | TRANSITION (slope) |
| proximity to next shift | 0.035 (4h) |
| out-of-sample AUC | 0.6877 |
This layer states TIMING and PROXIMITY to the next regime shift. It states no direction — a shift is a large move, up or down is not part of the measure. Provenance warning, measured: the fields in this block do not all come from the same model. The probability and the timeframe come from the per-timeframe risk gate; the AUC comes from a separate precision model with a one-bar horizon. Both are shown elsewhere on this site; neither is suppressed, and they are not merged into one number here.
Measured just now: the two disagree on the warning level — the risk gate reads CALM on 4h, the precision model reads LOW. cmmx_orientation.json:wohin.warnung · cmmx_unified_state.json:transition_warning.level
Two models, two warning levels — at the same instant the per-timeframe risk gate reads CALM on 4h, while the precision model reads LOW on a one-bar horizon. That is not an error in either — they measure different scales. It is shown because a single "warning level" would have to pick one and hide the other.
Segment geometry
| transition points | 6142 |
| sign autocorrelation (lag 1) | 0.017 |
| flip rate | 0.491 |
| magnitude autocorrelation | 0.100 |
| duration autocorrelation | 0.226 |
A sign autocorrelation near zero with a flip rate near one half means no direction is inferable from the previous segment. Magnitude and duration carry a little structure, the sign carries none. The system publishes this against its own interest.
Point order
| self_excitation_x | 2.700 |
| powerlaw_alpha_mle | 3.483 |
| powerlaw_ks | 0.046 |
| powerlaw_favored | power-law |
| interarrival_cv | 1.912 |
| bars_since_last | 15.000 |
| in_excited_window | NO |
Information clock
| state | MEDIUM |
| expected activity vs daily average | 0.928× |
| peak hour (UTC) | 14 |
| trough hour (UTC) | 4 |
| confirmed out of sample | YES |
Deliberately withheld
| Artefact | Reason |
|---|
| cmmx_future_distribution.json | PRE_GATE_FORECAST_LAYER |
| cmmx_guidance_scorecard.json | PRE_GATE_FORECAST_LAYER |
Both artefacts exist, are
live, and are not shown. They belong to the forecast layer, which has
not passed its gate; neither carries the flag by which the other artefacts
declare themselves free of a price claim. Omitting them silently would be the
same failure as a silent status change — so the omission is stated.
Provenance
| unified | 2026-09-11T03:59:19Z 60 min described moment |
| orientation | 2026-09-11T03:58:36Z 60 min described moment |
| transition | 2026-09-11T03:55:32Z 60 min described moment |