What changed in the data?

Frequency: compare raw counts over declared windows

Crime and 311 compare the most recent 12 complete covered months with the preceding 12. Business licenses compare the latest 24 with the preceding 24, requiring 48 consecutive complete months, because two-year renewal schedules can distort annual comparisons. Every business category, community and ward condition uses this rule, not just the motivating example. Missing or provisional months make the comparison unavailable; there is no annual fallback.

The unchanged magnitude gates require at least a 1.5-fold change, an absolute monthly-mean difference of at least five records, and at least 120 raw records in the larger comparison window. Means divide totals by the actual window length (12 or 24); gates use unrounded values. Zero onset or cessation retains the raw zero rather than inventing an infinite ratio.

Keep window sensitivity visible

Business pages retain the annual comparison beside the two-year comparison. Annual-only flags are unranked review context, not qualifying frequency candidates. Opposing directions, two-year-only flags and insufficient history are labeled explicitly. The revised scan has 293 qualifying candidates and 70 additional annual-only contexts, whose original URLs remain available. The context count overlaps the no-finding/insufficient-evidence dispositions below; do not add it to the total conditions searched.

This revision was motivated by the observed Archer Heights case and remains exploratory. Fixed two-year windows reduce a renewal-calendar mismatch; they are not seasonal adjustment, proof that renewals explain every condition, a detected breakpoint or independent statistical validation. A longer window can dilute a genuinely recent change. No source observations were removed or corrected, and the same dated source snapshots were used.

Celeration: compare fitted multipliers

The shared Analyze engine fits at most three disjoint phases. Each monthly phase has at least 12 observations. A turn requires at least a 1.5-fold change between adjacent six-month factors and at least 120 raw records in each adjacent phase. Native selection rules and serial-pattern withholding remain in force. A withheld suggestion is not promoted to a turn or a finding of stability.

Charts retain the same six-log-cycle, 120-month frame. Confirmed zeros become one only in log-space fitting and diagnostics; observations, totals and plotted zero markers remain zero. Missing or provisional interior observations are not silently removed. Read about Standard Celeration Charts.

A declared search, not proof of significance

The scan searched named communities across the five feeds, using the baseline plus individual category, flag, time-band, ward or status conditions as applicable. It did not search arbitrary subsets or synthesize citywide totals. 9,944 conditions had no finding and 688 had insufficient evidence. Neither disposition proves stability. There were no pending conditions or errors in this saved run.

Turn priority combines the log of the factor change with supporting volume; frequency priority combines relative count difference with total volume. Priorities are compared only within a dataset and signal. Conditional factor intervals and near-optimal boundary ranges do not correct for searching thousands of dependent conditions. No calibrated false-discovery rate, peer-review claim or causal inference is offered.