Ward coverage

1. Location precision in the published file

Unit: service requests. Source: 311 Service Requests 2025, all 500,269 rows, 2025-01-01 to 2025-12-31. Query location_precision_partition in sql/10_evidence.sql. The coarse bucket is identified by an A0A-form value; the fine bucket carries the literal string Intersection in the postal-code column.

2. Raw ranking versus deduplicated ranking

Only 8 of 25 wards move, and no ward moves more than 2 places. This is not a reordering.

The arithmetic is the reason, and it is not subtle. Deduplication can only touch the 15.68% of requests that carry a corner, and it removes 13,395 of 500,269 rows, 2.68% of the file. The gaps between adjacent wards in the raw ranking are larger than that. A 2.68% perturbation cannot reorder a list whose neighbouring entries differ by more.

Deduplicated unique-issue count for a ward = requests in that ward with no corner (each its own unique issue, because we cannot prove otherwise) + clusters whose earliest report is in that ward. Anchor attribution avoids double counting clusters at an intersection sitting on a ward boundary; only 8 of 64,438 clusters span more than one ward.

Units: requests, issues, rank positions. Source: query ward_ranking in sql/10_evidence.sql, over all 500,269 rows. Ward boundaries: 25-ward model, 2018 federal riding boundaries.

One further caveat that cuts against the thesis rather than for it: two of the eight moving wards in this table move only because of a data artifact. Removing the automated Bathurst St submitter and the 311 contact-centre feedback records, both documented on the Concentration page, drops the table to 4 wards moving, maximum 1 place. The honest reading of this dataset is that there is no reordering at all.

3. Income validation

Income is not a model input. It is joined after every count above is final, and it is read by nothing upstream of any allocation decision. Its only job is to test whether the under-reporting tracks disadvantage. If income were a feature, the system would be allocating on who lives somewhere rather than on what is broken there.

Units: Pearson r (dimensionless), t statistic, two-tailed p. Sources: request counts from sr_raw (311 Service Requests 2025); median total household income 2020 and population 2021 from City of Toronto Ward Profiles (25-Ward Model), 2023 release, sheet 2021 One Variable → Statistics Canada 2021 Census. Correlations computed by Postgres corr() in query income_correlation; p values by two-tailed t test in scripts/part1_evidence.py.

Reading these honestly. Higher-income wards do report more, and the sign is the one the reporting-bias literature predicts (Kontokosta & Hong 2021, Kansas City; Minkoff 2016, NYC). But r = +0.3827 at p = 0.059 does not clear the conventional threshold with 25 wards. It is consistent with the literature; it is not independent confirmation of it. And the duplicate rate does not track income at all, r = −0.0987, p = 0.6388, a flat null. Whatever the loud-ward effect is in this data, it is in report volume, not in redundancy per problem.

Units: CAD (median total household income, 2020 income year); residents; service requests; issues; requests per 10,000 residents; percent. Source: query income_table, joining sr_raw to ward_income.

4. Subgroup analysis, located requests only

Subgroup analysis. This is not the headline. It covers 15.68% of the file.

The full-ward table in section 2 is dominated by the 84.32% of requests that carry no corner and therefore cannot be deduplicated at all. Those rows pass through both columns unchanged and dilute any movement. This narrower table asks a different question: among requests where deduplication is even possible, does the ranking move?

Units: requests, issues, percent, rank positions. Source: query ward_ranking_located_only in sql/10_evidence.sql, over the 77,833 intersection-located requests.

The −10 on York Centre is one street corner, and that corner is a data artifact.

The confound, named. York Centre's 48.91% duplicate rate and its 10-place drop are produced almost entirely by BATHURST ST & CANYON AVE, a single intersection carrying 1,948 Moving Motor Vehicle Noise reports, 1,929 redundant, 14.4% of all redundancy in the city. Those reports arrive at 43–46 second intervals; 30.1% of the gaps between consecutive reports at that corner fall in the 30–60 second band, against 14.7% for every other noise report in Toronto. That signature appears at exactly two places in all 500,269 rows and both are on Bathurst St. It is one automated submitter, not 1,948 residents. Remove it and York Centre's movement disappears. The evidence is set out in full on the Concentration page.

Every other ward in the subgroup table moves by 3 places or fewer, and 10 of 25 do not move at all. Even in the subset built to favour the mechanism, the effect is modest once the artifact is removed.