Skip to content

Releases

Research skills 0.2.2

Choose a tag to compare

@livejake livejake released this 20 Nov 11:39
668a5d0

0.2.2 (2026-11-19): Kano and Gabor-Granger questions can be created over MCP

  • kano-analysis and gabor-granger-analysis (0.1.2): the QuestionPunk MCP server now creates both question
    types with question_create, so "Field it" gives the payload instead of sending you to the app.

0.2.1 (2026-11-19): third review completed, Van Westendorp checked against R

The third (Gemini) review of survey-bias-check, survey-sample-size, ux-scale-scoring and
survey-crosstabs had failed or stopped early in 0.2.0; it was rerun, each finding was reproduced, and every
confirmed script fix has a regression test.

  • van-westendorp-analysis 0.2.1: --compat r checked against R pricesensitivitymeter 1.3.3. Its price
    points, trial curve and maxima match R in 66 runs (fixtures/r_psm_1_3_3.expected.json). One rule differed
    and now follows R: when curves coincide up to the highest price, --tie max and --tie mean report the
    run's first price. R's revenue price is the script's "highest among the observed prices only" line.
  • ux-scale-scoring 0.1.1: a 7-point SUS (labels or a few 6s and 7s) and reversed agree labels now stop
    instead of being scored; SUS columns must be numbered exactly 1 to 10; --reverse-performance checks the
    column order; negated SEQ anchors ("Not at all easy") no longer trip the reversed-SEQ stop, and
    --seq-reversed is refused on text labels; --target with groups tests the groups, not the pooled line
    or blank rows; --drop-flagged reports and counts correctly; constant item means give no correlation;
    "NaN", "0_5", decimal TLX sliders and trailing empty Excel columns are handled.
  • survey-sample-size 0.2.1: no crash on compare --table near 0% or 100% or on tiny inputs; a one-sided
    alpha of 0.5 or more is refused; a plan that reaches the population is a census; continuity-corrected
    power no longer jumps; nps --diff refuses --population and reports its z values; subgroups raises
    the base by the design effect; power checks state the power; check reports the detectable difference
    that holds in either direction; gap is limited to single-select questions and says its n is a 50%-power
    boundary. Docs: NPS comparisons use nps --diff, opt-in wording, more examples.
  • survey-crosstabs 0.1.1: equal weights other than 1 no longer scale the chi-square; anonymity notes no
    longer split hidden cells into small and complementary counts; withheld tables carry no counts or group
    list in --json; groups with no answers stay listed under anonymity; repeated --hide flags add up;
    "0,750" reads as a decimal; Cramér's V is reported; pipes in labels are escaped.
  • survey-bias-check 0.2.1: numbered verbal scales ("1 - Very dissatisfied") are checked as rating scales;
    one product name with a digit no longer turns a list ordinal; "Very / Somewhat / Not very / Not at all
    likely" is balanced, and a one-sided scale around a neutral midpoint is flagged. Docs: severity column in
    the output, phone-mode length line, no "behaviour before attitude" rule.
  • survey-from-decision 0.2.1: SKILL.md under 300 lines; names the pack's screener, pricing and crosstab skills.
  • kano-analysis and gabor-granger-analysis 0.1.1: the in-app question types are named as they appear
    in the add-question menu ("Feature Priorities (Kano)", "Price Ladder").