Written Feedback Quality Assessment for Managers, Teachers and ReviewersThey read it. Could they act on it?

Methodology: Thirty-five original exercises across five formats: fourteen binary calls on real feedback lines, ten single-choice rewrites, five select-every-that-applies exercises, four matching exercises and two written exercises. One response instruction is declared for the whole keyed instrument and it is a KNOWLEDGE instruction - what is most likely to be true, or most useful to the person receiving it - never what you would do. The two written exercises carry a production instruction, are rubric-graded, and are reported apart and never averaged into any keyed figure. Construct statement: it measures whether somebody can tell feedback a receiver could act on from feedback that only tells them how they did, and whether they can move a comment off the person and onto the work. It does not measure writing ability, warmth, subject expertise, how well liked somebody is, whether their feedback is kind, or whether the people they manage are performing. Scoring is signal detection. The fourteen binary calls are a balanced set - seven that can be acted on and seven that cannot - so discrimination (d-prime) and threshold (c) are computed separately and reported separately, with the log-linear correction applied so that an extreme rate cannot make d-prime infinite. Two readers with the same discrimination and opposite thresholds are a different problem from each other and a single accuracy percentage destroys that distinction. Accepting every line and rejecting every line both land at zero discrimination as arithmetic rather than as a threshold. The keyed strands beside it are corrected against an authored answer prior per option rather than against a uniform draw, because a uniform draw is not what anybody answers. Reliability is McDonald's omega estimated from item count and a stated assumed inter-item correlation; it is printed with what it permits, and where it will not carry a number the refusal is printed rather than performed silently. Every reported figure carries its standard error band. Constructs and sources: the four-level model of feedback - task, process, self-regulation and self - and the finding that the self level is the one that does not help (Hattie and Timperley 2007); the meta-analytic finding that about a third of feedback interventions REDUCED performance, and that directing attention to the self is the mechanism (Kluger and DeNisi 1996); the three conditions for formative feedback - know the standard, see the gap, act to close it (Sadler 1989); specificity, timing and elaboration in formative feedback (Shute 2008); the finding that information-rich feedback substantially outperforms reinforcement and punishment feedback (Wisniewski, Zierer and Hattie 2020); feedback literacy and uptake on the receiver's side (Carless and Boud 2018); implementation intentions and the if-then form, meta-analytic d = 0.65 across 94 studies (Gollwitzer and Sheeran 2006); goal specificity (Locke and Latham 2002); the speaker's overestimation of their own clarity (Keysar and Henly 2002); signal-detection theory and the separation of discrimination from criterion (Macmillan and Creelman 2005); the log-linear correction for extreme rates (Hautus 1995); and the Barnum effect as the reason a report that names only attractive things reads as a horoscope (Forer 1949). All items are original works. No commercial feedback instrument, 360 questionnaire, competency vocabulary or report layout is reproduced or implied, and no affiliation with any assessment publisher or training provider exists or is claimed.