{"path":"research/how-independent-reasons-add/his-readings.md","content":"# His readings of the independence cases, one case at a time\n\n*Started 2026-10-23, when the independence reading (`deliberus/independence.py`, ledger Q40) and its\nquality control passed to the session that owns machine passes. The method is the global rule\n*Case review starts one case at a time*: the first case chosen on purpose, his verdict and reason in\nhis words before the next, each reason turned into a candidate rule tried on the remaining cases.\nThe cases are `independence_pass_cases.py`; the machine's answers are the JSON files beside this.*\n\n## Case 1 · markets and experts\n\n**Why it was chosen first.** The machine was least steady on it: across five runs (four on Sonnet, one\non Haiku 5.5; § 10 of the note) it answered *one story* four times and *partly* once, while the note's expected answer was *different*\n(`how-independent-reasons-add.md` § 6 point 2, resting on Graefe et al. 2014's three elections,\nwhere markets and experts together cut the error as much as two independent methods would).\n\n> Claim: the incumbent will win the election.\n> Reason one: *Prediction markets give the incumbent a 70% chance of winning.*\n> Reason two: *A panel of political scientists expects the incumbent to win.*\n\n**His reading, 2026-10-23, verbatim:**\n\n> The experts are part of the \"market\" generally. \"Prediction markets\" doesn't name a single such\n> platform but is more broad, so hints at almost naming \"the whole market\", kind of. That makes it\n> feel like there's some obvious overlap because everyone who is part of society kind of loosely\n> either could or should participate in prediction markets, the idea behind them is that markets\n> self-correct and have an intelligence of their own because everyone who sees an opportunity to\n> make money off good bets goes for it etc... But that's ideological and I don't think more than a\n> handful are actually interested, aware of the existence of prediction markets and well-read enough\n> etc etc, a bunch of selection criteria there for ppl to even engage and make bets. Those experts\n> might very well not be a part of this crowd in actuality. In any case, the inferential distance\n> between these two reasons is large. But ofc there is a POSSIBLE overlap because they're about\n> multiple sets of people out there in the world - the political scientists and the users of all (?)\n> prediction markets. And generally I guess it's well-read people who we could expect to have a good\n> grasp on the matter that participate on prediction markets, but again that's just an idea I hold\n> that becomes a premise in how I make sense of this. It might not hold up, maybe a majority of users\n> in prediction markets are opinionated, ideologically stubborn or misinformed or miscalibrated\n> epistemically and the smart minority take advantage of that, maybe that's part of the business\n> model, etc etc. Maybe there's a norm for political scientists to stay off prediction markets, I\n> wouldn't know in particular because I'm not in pol-sci academia. How would we know that for sure?\n> Who would we trust on that matter? I could go on and on. There's so many thoughts to be had about\n> this.\n\n> The inferential distance between the claim and the reasons here feels like a bit too much of a\n> leap. We need to be responsibly aware at least of all the inferences between them.\n\n**What his reading settles, and what it leaves.** No verdict of *different* or *one story*: a\npossible overlap through shared people, resting on premises he holds and names as premises (who\nbets on prediction markets, whether political scientists do, whether markets are well calibrated),\neach of which he says he could not settle and would need someone to trust on. And past the question\nasked: the distance between each reason and the claim is itself too large to judge the pair over.\nHis answer reaches past the question, so the question was too narrow (the case-review rule): the\ncase is widened in [../folded-routes-and-shared-premises.md](../folded-routes-and-shared-premises.md).\n\n**Candidate rule from his reason** (Claude's wording, unruled): *two reasons are read as different\nonly when every channel by which they could share an error has been written out and examined; a\nplausible channel left unexamined caps the reading at partly.* Its channels for this case: shared\ninputs (both read the same polls), shared people (experts who trade), one reading the other\n(traders read the experts, experts read the markets).\n"}