Turnmetry model changelog
Material rule changes receive a new model version. Earlier signals should remain attributable to the rules that actually produced them.
Last updated: September 18, 2026
September 18, 2026 — v3 evaluation correction
Evaluation daily-close-v1 measures top signals against the highest daily close and bottom signals against the lowest daily close inside the existing historical reference windows. The earlier report used the close on the original anchor date, which was not always the closing-price extreme. Both the corrected reference and original anchor remain visible.
The model remains cycle-v3: its signal rules and historical transition dates are unchanged. Timing and price-distance results have been recalculated, and the report now distinguishes time matches from price accuracy. Counts within 10% and 15% are descriptive checks, not a newly fitted success standard.
The report also makes explicit that the reference set mixes local and cycle extremes and was selected with historical knowledge. A replay with potentially revised provider data is not an immutable live track record or proof of out-of-sample predictive performance.
July 2026 — cycle model v3
Model v3 is the active version. It narrows the primary top path to a causal, price-derived state machine and keeps broader valuation and miner indicators as context. This reduces the risk of treating several correlated transformations as independent confirmation.
Peak Candidate
A new daily closing high must coincide with either late-cycle halving age or a Pi Cycle ratio near its crossover.
Top Turn
After a candidate, price must retreat at least 3% from the tracked high and close below its seven-day average.
Top Confirmed
A 30% decline from the tracked high is treated as structural confirmation, which necessarily arrives after the top.
Bottom path retained
The multi-family bottom watch and rebound confirmation introduced before v3 remain in place.
The top cooldown and state transitions also prevent repeated daily alerts from being presented as independent calls. Historical evaluation uses the same sequential state machine as live processing.
Previous — cycle model v2
Model v2 established category-aware readiness, explicit watch states and a persistent bottom process. Valuation, momentum, miner stress and cycle location were grouped so that multiple related price indicators could not inflate readiness simply by moving together.
Research on historical cycle-level behaviour motivated the simpler v3 top path. That change does not prove v3 will perform better in future cycles; it records a design choice made before accumulating a forward track record.
Earlier experimental versions
Earlier iterations explored threshold voting, divergence and additional market inputs. These were useful research branches, but not every experiment belongs in a live signal. Disabled rules are not described as active evidence, and retrospective experiments should not be mixed with live v3 outcomes.
Historical comparisons may be recomputed for research. A replay uses the available data and rules; it does not by itself establish which signals were originally issued live. Original assessments must be preserved separately before claiming an immutable forward track record.
How changes are evaluated
A proposed change should be causal, reproducible and assessed across complete cycles where data permits. Evaluation must include unmatched signals and references, timing error, price distance and subsequent adverse moves. A match in time alone does not establish a useful forecast, and known historical cycles do not become unseen test data by replaying them.
Read the full methodology, review the data definitions, and use the historical results to understand what the active version did and did not capture.