Multi-Resolution Change-Point and Trend Comparison¶
Diagnostic test — instantiates Continuity–Rupture Regime Diagnosis and Transition Design
Compares persistence and candidate breaks across windows, scales, measures, locations, and groups to see which change claims survive a change of resolution.
Multi-Resolution Change-Point and Trend Comparison is the measurement instrument that answers when — if ever — did the number actually break? Given a series, it runs the same question at several resolutions at once: fit a smooth trend and a break at each candidate boundary, then repeat over daily, monthly, and decadal windows, over alternative measures of the same thing, and over regions and subgroups separately. Its distinguishing move — and what separates it sharply from the Continuity–Rupture Claim Matrix, which also "compares across scales" — is that this mechanism computes the comparison from data rather than organizing interpretations of it: it is a statistical detector whose output is which candidate breakpoints and trends are robust to the choice of window, and which are artifacts of how finely you looked. It deliberately stops short of saying why a break happened; it only locates persistence and candidate rupture, with the sampling caveats attached.
Example¶
A climate group has forty years of a mountain glacier's mass-balance record and two rival readings of it: a talk claiming an abrupt "collapse" around one warm year, and a report calling the loss a steady decline. The comparison runs both hypotheses at every resolution. On annual data a single warm year does look like a step; widen to a decadal window and the "step" dissolves into a continuous downward trend that the warm year merely punctuates. But when the same detector is run on the high-elevation accumulation zone alone, a genuine break does survive every window — that sub-series really did shift regime, even though the glacier-wide aggregate did not. The output is not a verdict but a resolution-tagged map: glacier-wide, gradual; high zone, a real break near year X ± a couple of years; and the apparent whole-glacier "jump" is a sampling artifact of annual granularity. That is exactly the evidence the downstream diagnosis needs, and none of the causal story it will later demand.
How it works¶
The engine is the same break-and-trend test swept across resolutions, then cross-checked:
- Fit rival shapes at each boundary. At every candidate date, compare a continuous trend against a trend-plus-break, scoring which the data prefer and by how much.
- Sweep the window. Re-run across sampling intervals and window lengths — micro, meso, macro — and keep only the breaks that persist across them.
- Vary the measure and the population. Repeat on alternative operationalizations of the same property and on each region or subgroup separately, so aggregate smoothness cannot hide a local cliff.
- Report robustness, not a point. Express each finding as a boundary range with the resolutions that support and dissolve it, plus the sampling and instrument caveats that could be faking it.
Tuning parameters¶
- Window and sampling interval — how finely the series is sliced. Fine resolution localizes a true break but mistakes short-run volatility for regime change; coarse resolution is stable but can smooth a real cliff into a ramp.
- Break-detection sensitivity — how strong the improvement must be to accept a breakpoint. Loose settings manufacture change-points from noise; strict ones miss weak but real shifts.
- Multiplicity control — how hard you correct for having scanned many candidate dates and windows. Under-correcting all but guarantees a spurious "significant" break somewhere.[n1]
- Subgroup and spatial decomposition — how far the series is split by region, cohort, or measure before testing. Deeper decomposition surfaces hidden local rupture but thins each series and raises noise.
- Instrument-change annotation — whether definition, sensor, or collection changes are flagged as candidate false breaks rather than read as real ones.
When it helps, and when it misleads¶
It is the right instrument when two accounts call the same trajectory a ramp or a jump and the disagreement is really about resolution — it settles which breaks are robust and which are painted on by granularity, and it exposes the aggregate that hides a subgroup cliff. Its central failure mode is that a change-point detector always finds something: given enough candidate dates, windows, and measures, noise alone yields a "significant" break, and the look-elsewhere hazard is severe unless multiplicity is controlled.[n1] It is equally fooled from the other side — a sparse archive can make a gradual process look punctuated, and a mid-series instrument change can fabricate a clean step that never happened. The discipline is to demand that a break survive across resolutions and measures before believing it, to annotate every instrument change as a candidate artifact, and to hand the surviving breaks to mechanism analysis rather than treating a robust breakpoint as proof of a cause.
How it implements the components¶
multi_scale_timeline_and_boundary_set— its primary output: a set of candidate boundaries tested at several resolutions, with the regional and subgroup chronologies that show which claims depend on which temporal choices.continuity_rupture_evidence_ledger— it supplies the ledger's state and trajectory layer — the level-and-path evidence for persistence and candidate breaks, each stamped with its sampling and instrument uncertainty.
It does not discriminate which causal mechanism produced a surviving break — that is Process Tracing and Mechanism Discrimination (the mechanism classifier) — nor organize the full object×property claim structure the ledger hangs on, which Continuity–Rupture Claim Matrix owns.
Related¶
- Instantiates: Continuity–Rupture Regime Diagnosis and Transition Design — it is the measurement front end that locates candidate breaks before any regime is named.
- Sibling mechanisms: Process Tracing and Mechanism Discrimination · Continuity–Rupture Claim Matrix · Threshold, Hysteresis, and Reversibility Probe · Parallel-Transition and Cutover Rehearsal · Post-Transition Legacy, Loss, and Regime Audit
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Multi-Resolution Change-Point and Trend Comparison operates as a bounded evaluation of existing evidence or work that produces a finding or disposition because it compares persistence and candidate breaks across windows, scales, measures, locations, and groups to see which change claims survive a change of resolution.
Independent corroboration: The frozen evidence defines Multi-Resolution Change-Point and Trend Comparison as 'Compares persistence and candidate breaks across windows, scales, measures, locations, and groups to see which change claims survive a change of resolution', so its operative form is Assessment, Review & Assurance.
Nearest alternative: Analysis, Modeling & Optimization — The procedure computes break robustness, but its bounded purpose is assurance that a claimed transition survives alternative resolutions.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Testing persistence of change points and trends across windows, measures, and groups is rooted in statistical time-series robustness analysis.
Related originating lineages:
- Data Science & Analytics — Large-scale analytics contributes systematic search across measures, groups, and locations.
- Earth Sciences — Climate and geophysical analysis prominently developed scale-sensitive trend and regime-shift comparisons.
Review resolution: Both independent reviews agree on primary origin statistics_experimental_design; reconciliation resolves secondary fields (origin_mode_disagreement, encyclopedia_synthesis_disagreement). Alternate origins retained (data_science, earth_sciences) are the union of reviewer-supported formative lineages with explicit rationales, not a list of later application domains. Present-day breadth is represented separately as domain_reach=multi_domain; origin_mode=cross_disciplinary_synthesis records the historical relationship among lineages. Confidence is conservatively reconciled to high, and encyclopedia_synthesis=true preserves either reviewer's finding that the encyclopedia generalized the mechanism.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The multiple-comparisons (or "look-elsewhere") problem: scanning many candidate breakpoints, windows, and measures inflates the chance that pure noise produces at least one apparently significant break. Requiring a break to persist across resolutions, and correcting the threshold for the number of tests, are the standard correctives. ↩a ↩b