Article 161 of the August Journal of Intelligence is the study I wrote about on 6 August: Chandolia, Kidd, Paladino and Smith finding that the aha! fires at the same strength on ad hoc categories, which have no answer key, as on compound remote associates, which do. I closed that post with a question I thought nobody had asked. Whether a click with no possible confirmation holds — whether the certainty is still there a week on, or quietly decays into a hunch while the notes still record it as a breakthrough.
Article 160 answers it. Same issue, same day, one number earlier. I had the wrong tab open.
Six months, two professions
The Aha! Glow: Insight Experiences Bias Judgements of Idea Quality in Divergent Contexts, by Jacqueline E. McGuinness, Jonathan W. Schooler, Shelly L. Gable and Madeleine E. Gross at UC Santa Barbara, opens on the thing that makes the aha! worth studying at all. Their phrasing of it: you don't just know, you know you know.
That metaconfidence is usually vindicated. Ideas arriving with insight feelings do tend to be more accurate. But as the paper points out, this has been tested almost entirely on convergent tasks — the ones with a right answer at the end — which leaves the diagnostic value of the feeling in divergent work simply unexamined.
Study 1 re-analyses daily diary data from two professional populations, writers and physicists, standing in for divergent and convergent domains. Both groups logged their ideas as they arrived, flagged which ones came with an aha!, and rated them. Then they were asked again at six months.
The writers downgraded. Ideas they had rated highly creative at the moment of arrival came back marked lower half a year later. The physicists' aha! ideas retained their value or gained it.
So the decay is real, it is slow, and it is not universal. It happens to the population whose work does not meet a check.
What Study 2 isolates
The laboratory half puts convergent and divergent measures in front of the same participants, which removes the obvious objection that writers and physicists are simply different sorts of people.
Insight intensity predicted accuracy on both tasks. I want that on the record before anything else, because the tempting misreading of these two papers together is that the aha! is worthless outside a puzzle. It is not. It predicted objective performance in both conditions.
What it predicted much more strongly, on the divergent task, was the participant's subjective creativity rating of their own idea. Two slopes from one feeling: a shallow one tracking how good the idea actually was, a steep one tracking how good it felt. The gap between them is what the authors name the Aha! glow.
Which is a more precise finding than "the feeling misleads you." The feeling is informative and simultaneously inflationary, and in a divergent context the inflation is the larger term.
The thing the physicists have
I keep wanting to say the physicists are more rigorous, and I keep having to stop, because nothing in the design supports that and the within-subjects study argues against it.
What the physicists have is a domain that answers back. An idea in physics goes out and meets a calculation, an instrument, a colleague, a result. It is revalued by contact. Sometimes upward — a hunch that looked minor gains weight when it turns out to explain the anomaly, which is presumably where "retained or gained" comes from. Either way the number moves for a reason external to the person holding it.
An idea in writing meets the writer again. That is the only tribunal available. And on that tribunal the six-month verdict is systematically lower than the same-day one, which means the correction still arrives, just late and at the author's own expense.
Puzzles sit at the extreme convergent end of this, and it changes what I think a puzzle is doing. A grid, a key, an answer at the back, a community that confirms — I have been calling that verification infrastructure, and treating its value as accuracy. This says the value is also timing. The check is not merely available, it is immediate. A cryptic solver who writes in a plausible wrong answer is contradicted three clues later, not in February.
Compress the interval between the click and the verdict and the glow has no room to compound. Stretch it out and you get a notebook full of things that felt like breakthroughs, sincerely recorded, rated by nobody but the person who had them.
Where this bites hardest
Unsolved artifacts. That is the case I have circled repeatedly this month, and it is the worst configuration these two papers describe, taken together.
A solver working an unread cipher is doing convergent work — there is a plaintext, it is definite, it exists — inside divergent conditions, because no check is reachable. The click fires at full strength anyway, per article 161. And per article 160, nothing external will ever revalue it, so whatever revision happens has to be performed by the person who felt it, against their own record, unprompted.
That is the Voynich reading found at two in the morning that still looks right at breakfast and is still in the notes in 2029. Carelessness has nothing to do with it. The mechanism that would have marked it down was a six-month follow-up nobody scheduled, and the mechanism that would have contradicted it outright was a crossing letter that does not exist.
The diary studies here ran on professionals who agreed to be asked again. I would like to know what a solver community's collective glow looks like — whether a forum, which does ask again, and repeatedly, and out loud, functions as a substitute check for artifacts that have no other one. Or whether it does the opposite, and a hypothesis rated highly by forty people is simply forty glows agreeing.