# Library inclusion indicators: population and interpretation Author: Alisson Bento. Source: Istat, Indagine sulle biblioteche pubbliche e private, reference year 2025, collected March–July 2026. Tables 3, 9, 28 and 29; questionnaire Q17, Q35 and Q37. The 20 region rows are disjoint; Italy is a separate aggregate. Province rows are not added to their containing region. The equipment margins refer to all eligible library functions. Inclusion-project questions are restricted to libraries with predominant public-reading function. National frame totals are 8,883 and 6,252 respectively. Public reading is a function, not synonymous with public ownership. Equipment rates restricted to public-reading function cannot be recovered from the published margins. No cross-universe gap or correlation is calculated. The equipment extract includes equipped-place categories and physical/communication supports, not digital-content services. Table 28 provides positive percentages but no separate unknown percentage; its complement is not a verified no rate. Table 9's Trentino-Alto Adige equipped-place distribution lacks Bolzano information; the region's full frame size is not its available-data denominator. Conditional mean places are not extracted. Programme margins retain autonomous, formal collaboration, no project and unknown responses separately for physical disability and sensory or emotional disabilities and/or cognitive disorders. The questionnaire allows both modes. Adding modes must not be interpreted as distinct libraries, and neither table establishes which equipment and projects coexist within a library. Table 29 nevertheless shows the four responses of each category summing to 100 within rounding (total column 100), and its footnote and the methodology note do not state how libraries selecting both modes are classified; the modes are therefore never added. Percentages retain one-decimal source precision. Symbols are retained: '-' denotes no phenomenon or no observed cases according to the source convention; '(a)' in Table 9 denotes unavailable information. Unknowns are never recoded as no. This census-style collection has coverage and item-information limitations; no sampling significance tests are supplied. No legal compliance, complete accessibility, realised user access, demand, programme quality or causal effects are measured. CSV files are bounded aggregate derivatives with exact source cells, UTF-8 encoding and LF line endings. Reproduce them by obtaining the official workbook named in sources.json and running `python extract-source.py /path/to/Tavole-dati.xlsx`. Requires Python and openpyxl; reads the workbook without modifying it. ## Conservative programme-only comparison For each target separately, let A be the autonomous programme percentage, C the formally collaborative percentage and N the no-project percentage. The completion guide printed p24 (PDF page25) defines No as no project in either mode and permits selecting both yes modes. Therefore max(A,C) is a lower bound on the union, while 100−N is an upper bound allowing every unknown to offer. Unknown is not treated as No. Expand to [max(0,max(A,C)−0.05), min(100,100−N+0.05)] under conventional nearest-tenth rounding. The source documents automatic rounding, not its tie rule; closed expansion is conservative under this rounding model. These are deterministic logical/rounding envelopes, not confidence intervals. They are deliberately conservative and not claimed to be sharp identified sets. All 190 unordered pairs of the 20 regions are checked separately for the two target categories. A direction is retained only when one expanded lower endpoint is strictly greater than the other's expanded upper endpoint. Overlap or touching means unresolved by this rule, not equality or proof that stronger information could never identify an order. The rule retains 35 physical-disability and 37 sensory or emotional disabilities and/or cognitive disorders pairs; it leaves 155 and 153 unresolved. Selection is exploratory. The two target populations are never added together. The public standard-library reproduce.py operates on programme-margins.csv and rebuilds programme-bounds.csv, all-pair-checks.csv and bounds-results.json. The source extraction is separately available as extract-source.py; it accepts an official workbook path, requires openpyxl, and uses no microdata. Independent validation used the workbook ZIP/XML directly and integer arithmetic, reproducing both pair totals. ## Regional aggregation check For both selected Q37 targets, Bolzano and Trento weighted by their public-reading frame counts (234 and 126) reconcile to the Trentino-Alto Adige regional percentages. Weighting all twenty regions by their public-reading counts reconciles to Italy. No unavailable markers occur in the selected target columns. Every selected mode, no and unknown aggregate comparison is within 0.1 percentage points, the conservative tolerance for independently rounded source components and aggregates. Other target categories in the continuation of Table 29 have unavailable Bolzano values; those categories are excluded, not extrapolated into this analysis.