Positions

Judgment stated as judgment. Every call I make in public goes here with its date, where it was first stated, what would change my mind, and a review date. Later it is scored: held, revised or withdrawn. Nothing is quietly deleted.

4 positions · 3 open · 1 held · 0 revised · 0 withdrawn

Dated	Position	What would change my mind	Review	Status
25 Sept 2026	A completion number read off a connected-television account measures a room's tolerance, not one viewer's choice. A completion figure that is not split by device is an average of a person and a room. Stated in Finding No. 3, On the phone, 72 percent watch alone. On the television, 19 percent.	One platform publishing completion split by device, with the phone curve and the connected-television curve aligned on the same titles. If a room and a person finish the same shows at the same rate, the split does not matter and this position is withdrawn.	1 Mar 2027	open
21 Sept 2026	The commissioning spreadsheet divides in two: columns that are facts, and columns that are opinions wearing a fact's clothes. Neither kind can tell you whether the next show will work, because before anything is made there is nothing to measure. The objective columns are useful only for settling what the words mean. Stated in What Your Commissioning Spreadsheet Cannot Tell You, part three of the commissioning series	A published pre-greenlight model that predicts title-level outcomes out of sample on any streamer's slate — not a ranking of titles already made, a prediction made before the greenlight and scored afterwards.	30 Mar 2027	open
30 Aug 2026	Completion and retention come apart. A slate optimised against captured signal gets very good at holding attention inside a title and has no purchase on whether someone is still a subscriber next month. Stated in Stated in part one as an observation about measurement; tested against RevenueCat's short-drama data in part two Outcome: RevenueCat's 28 August 2026 data: short-drama apps convert about three and a half times better than other entertainment apps and retain about four times worse at six months, on the smaller subset that reports six-month data. The claim held on the first public test.	Resubscription data for micro-drama showing that the person who lapses after six months comes back for the next title that catches them — then 1.8% six-month retention is describing something less alarming than it sounds; nobody has published it, and that case stays open until somebody does. Or evidence that price dominates cancellation, in which case slate composition is a second-order effect and I am over-weighting it.	1 Dec 2026	held
30 Aug 2026	Data does not tell you what to make. It tells you how to make it and how to be less wrong, because the part of the outcome that matters never enters the log. Stated in Data Won't Tell You What to Make, part one of the commissioning series Outcome: Restated on 21 September 2026 in part three, What Your Commissioning Spreadsheet Cannot Tell You, as the argument of the whole series; nothing has tested it yet.	Evidence that the part you can't capture is a rounding error — if it is, you can ignore it and get on with modelling. If it's where the outcome actually lives, then a system built only on the captured part is measuring the shadow rather than the thing.	30 Oct 2026	open