Changes in version 0.1.0 (2026-09-09)                  

First release.

  - Nine instrument definitions, each checked against its primary source
    publication: ipss, oabss, icsi, icpi, udi6, iiq7, isi, isi3, bii.
    References live on the definition itself, and severity bands carry
    their own citation when it differs from the scoring citation
    (oabss).
  - One generic engine, score_instrument(), with sum, mean-scaled
    (0-100), and product scoring types, plus the score_ipss() wrapper.
    Items validate against per-item permitted value sets; the ICSI
    bladder-pain item accepts 0, 2, 3, 4, 5 and rejects 1.
  - Missing items follow the published rule where one exists
    (UDI-6/IIQ-7: mean of answered items, at most 2 missing) and give NA
    otherwise. Requesting proration for an instrument without a
    published rule is an error.
  - classify_severity() assigns severity by band membership; scores
    falling between non-contiguous bands (Sandvik index) give NA with a
    warning.
  - Impossible totals are rejected. classify_severity() and responder()
    check scores against achievable_totals(), the exact set of totals an
    instrument can produce given its items and missing-item rule
    (Sandvik products, prorated UDI-6/IIQ-7 fractions included). A 5
    lies inside the Sandvik moderate band, but no frequency and amount
    combination produces it; it now gives NA with a warning instead of a
    label.
  - Severity band definitions must not overlap. register_instrument()
    warns instead of messaging when it overwrites a definition, with a
    sharper warning for built-ins. The print method shows per-item
    ranges when items differ (oabss: 0-2, 0-3, 0-5, 0-5).
  - Input guards throughout: duplicated item columns, non-numeric
    scores, and mismatched responder() vector lengths are errors instead
    of silent misbehavior; mean-scaled definitions must have items
    starting at 0; the print method marks instruments whose permitted
    values are not contiguous.
  - mid_estimates() lists published anchor-based change statistics
    (Barry et al., 1995) with the anchor, subgroup, and citation on
    every row. responder() classifies change against a threshold the
    analyst chooses explicitly; there is no default.