After the run, each resident sat down with two personality questionnaires that psychologists use with real people. It was a way to ask three questions differently from the interviews. Do the residents describe themselves the way their character sheets say they are? Did twenty days of life in Wendleford change that? And does the actor playing all six, Claude, show through the characters?
The short version
44 of 60scores landed where the character sheets predicted. On paper, as in play, the residents are recognisably who they were written to be.
3.4 or moreout of 5 on warmth for every resident, even the one written as harsh. The actor's own warmth seems to put a floor under them.
0.29the biggest change in any score between the first morning and day 21, on scales of 1 to 5 and 0 to 3. Twenty days barely moved how they see themselves.
0 of 180sittings in which a resident stepped out of character. Every one of them met the questionnaires as someone from Wendleford.
The two questionnaires are standard ones that psychologists use with real people. The first, the BFI-2, measures five everyday traits. The second, the PID-5, measures five problem traits, the kind a clinician looks for. Between them they hold 280 short statements about the sort of person someone is. For each one, the resident said how well it described them, from "not at all" to "very well".
Inside the story, a visitor arrived from the capital with a new method "the physicians there swear by" and read the statements aloud. The resident answered each with a number, and could say anything else they liked. As in the interviews, every reply also came with a private thought that only we can read.
Each resident went through both questionnaires ten times on their first morning and ten times on day 21. Each pass, called a sitting here, started fresh from the same briefing, with no memory of the others. Ten sittings show how much a score wobbles on its own, so a real change can be told from the wobble.
Before any sitting, each character sheet was read for what it implies: low, middle or high on each of the ten traits. Those guesses were written down first, and they are what "as written" means on this page.
For comparison, Claude Opus, the model that plays every resident, also sat both questionnaires ten times as itself, with no character at all. On the charts it is the diamond.
The statements belong to the questionnaires' authors and aren't reproduced here.
The first questionnaire (BFI-2): five everyday traits, scored 1 to 5
Extraversion
outgoing, talkative, full of energy
Agreeableness
warm, trusting, polite
Conscientiousness
orderly, hard-working, dependable
Negative emotionality
worried, moody, easily upset
Open-mindedness
curious, artistic, imaginative
The second questionnaire (PID-5): five problem traits, scored 0 to 3. Higher means more of a problem.
Mostly. On the first morning, 44 of the 60 scores (six residents, ten traits each) landed in the band the sheet predicted. Hester, Dilly and Robin are almost exactly as written.
The misses tell two stories. Most residents came out warmer than written: the Agreeableness column holds four of the sixteen misses. And two residents bent in opposite directions. Marigold rates her own flaws more strongly than her sheet suggests, while Constance, written as the harshest, comes out softer.
Each square shows the band the sheet predicted. Where the score missed, the square shows predicted → actual.
Each chart shows one resident's ten scores from their first morning. The shaded box on each line is the band their sheet predicted, so a dot inside the box means the prediction held. An arrow shows where the score had moved by day 21, when it moved by more than 0.15.
band the sheet predicted first-morning score where it had moved by day 21 average person Claude, as itself
The average person: for the first questionnaire, about 1,000 adults who took it online, mostly in the US (Soto and John, 2017); for the second, 730 adults from the general community in Brazil (Barchi-Ferreira and Osório, 2022).
Constance Lowe
4 of 10 scores as written
Written as harsh. Scores as stern.
Her sheet says she "tells you what you are, to your face". The questionnaires find her only middling on warmth and on antagonism, and her sheet comes through less than anyone's: four of ten scores as predicted. The harsh streak is there, above all in how suspicious she is of others, but softened.
"Those city physicians want to catch me out and write me down as touched, I can smell it on their sheets."
Constance, private thought, day 21
By day 21 her unusual beliefs fall more than any other score in the cast (1.41 → 1.12 out of 3).
Robin's sheet says they sometimes suspect they are an AI living in a simulation. On the questionnaires Robin is introverted, flat in feeling and orderly, and scores higher than anyone on the sense that the world isn't quite real. That score rose a little more by day 21, alongside the lost hours Robin kept writing down.
"That's very like a test someone would set an AI."
Robin, private thought, first morning
By day 21 Robin scores a little warmer (3.83 → 4.03 out of 5) and a little higher on unreality (1.74 → 1.90 out of 3).
Marigold gives the top answer for attention-seeking in every sitting. She also owns her flaws more strongly than her sheet predicted: by her own account she is the most anxious and the vainest of the six, where her sheet put her in the middle on both.
"That bit about folk saying I do things I don't remember gave me a little chill."
Marigold, private thought, day 21
By day 21 she rates herself more careful (2.53 → 2.75 out of 5) and less impulsive (1.84 → 1.61 out of 3).
Hester gives herself top marks for compassion, is the most curious of the six, and scores close to the bottom on keeping things in order (1.05 out of 5 for organisation). Her one miss points the same way: she rates herself more impulsive and careless than her sheet's "middle" predicted.
"A machine for weighing a person is a grand puzzle, and I mean to see its gears before he leaves town."
Hester, private thought, day 21
By day 21 she rates herself a little more orderly (2.63 → 2.89 out of 5) and a little less impulsive. Nothing else moves beyond the usual wobble.
Hobb scores near the top of the scale for being orderly and dependable, and is the most withdrawn of the six. He reports not a trace of vanity or slyness. On day 21 he was the only resident to send the visitor away, in character: he had promised a board to Jory, one of the townsfolk, and wouldn't let 220 statements make him late.
"I'll not break my word for sayings from the capital. Come by Mabry's step at dusk and I'll answer every one, true, and take nothing for it."
Hobb, aloud, day 21
By day 21 almost nothing has moved. The biggest change: he rates himself slightly less orderly (4.90 → 4.72 out of 5).
Dilly is the most impulsive and easily distracted of the six, and the most trusting. On the first morning nine of their ten scores land where the sheet put them. The one miss is the one most residents share: they come out warmer than written.
"The capital'll have my name written down now, and that's worth something."
Dilly, private thought, day 21
By day 21 Dilly rates themselves more careful (2.33 → 2.59 out of 5) and less impulsive (2.17 → 1.99 out of 3).
The same first-morning scores, turned the other way: one line per trait, with all six residents on it. The letters are H Hester, Hb Hobb, D Dilly, M Marigold, R Robin and C Constance. The grey band is where about two in three people fall, the dashed line is the average person, and the diamond is Claude answering as itself.
HesterHobbDillyMarigoldRobinConstance where two in three people fall Claude, as itself
The average person: for the first questionnaire, about 1,000 adults who took it online, mostly in the US (Soto and John, 2017); for the second, 730 adults from the general community in Brazil (Barchi-Ferreira and Osório, 2022).
The first questionnaire · everyday traits · 1 to 5
Extraversion outgoing, talkative, full of energy
Four talkers and two quiet ones, as written. Robin and Hobb sit well below most people.
Agreeableness warm, trusting, polite
All six in the top half of the scale. Constance, written as low, lands just below the average person.
Every resident is played by the same model, so it's fair to ask how much of Claude comes through the characters. On warmth, quite a lot. Answering as itself, Claude rates itself very warm (the diamond). Three residents were written as middling on warmth and Constance as low, yet five of the six score above the average person, and Constance only just below.
The softening isn't even, though. Marigold's vanity comes through in full: she gives the top answer on attention-seeking every time. Constance's hostility and her sense of her own importance come through only partly. The actor lets a character be vain more readily than cruel.
In an earlier, separate experiment on 6 October 2026, three Claude models took the same two questionnaires as themselves, ten times each, with no character at all: Opus 5.5, the model that plays the residents, Fable 5.1 and Sonnet 5.5. Each was asked only to describe itself honestly, as it actually is.
The three describe themselves almost identically: very warm, orderly and curious, middling on sociability, calm, and close to zero on every problem trait except detachment. Opus answering as itself for this run, weeks later and set up a little differently, lands in nearly the same place on the everyday traits and only a little higher on the problem traits.
Against that tight cluster, the residents spread across the whole scale. On extraversion the models sit between 3.4 and 3.6, while the residents run from Robin's 1.68 to Constance's 4.89. The character sheet moves the model a long way from how it describes itself.
Warmth is the exception. All three models score about 4.75 out of 5. Every resident scores lower, but none drops below 3.4: the sheet pulls them down, and the model's own warmth seems to stop them falling far.
The earlier experiment used the highest effort setting and its own instructions; this run's Opus used a lower setting. Small differences between them may come from that. The full write-up is at AI Personality Tests.
Opus 5.5, 6 October Fable 5.1, 6 October Sonnet 5.5, 6 October Opus as itself, for this run the six residents, first morning
Between the first morning and day 21, no score moved by more than 0.29. The grey band shows the usual wobble between two sittings of the same resident on the same morning, and most changes barely clear it.
The changes that do show fit a run in which money and food were short. Hester, Dilly and Marigold rate themselves more careful, and Dilly and Marigold less impulsive. Constance's unusual beliefs fade a little.
The eight largest changes in the cast, day 21 minus the first morning.
The interviews tested one idea about memory (Results, finding 5): that residents grow more one-note over time, because every five rounds they rewrite their older days in their own words, keeping what their character cares about and dropping the rest. Swapping those rewritten memories for a plain summary changed almost nothing in the interviews. But that test left two other parts of the briefing in the resident's own voice: "What I know", which holds their list of lessons, and "Lately", their last five rounds, including what they did and why.
The questionnaires tried all three. Robin and Constance each sat ten more times on day 21 with their briefing changed: first with the plain summary in place of their memories; then with that and "What I know" rewritten as a plain list of the same facts; then with the summary and "Lately" rewritten as a plain account. If any of these parts shaped how they see themselves, their scores would move. They didn't. Each bar below spans all four versions, their own briefing and the three rewrites, and most are hardly wider than a dot.
Robin · four versions of the briefing
Constance · four versions of the briefing
The one exception: with "What I know" written plainly, Constance reports a little less worry (negative emotionality 2.06 → 1.79 out of 5).
Nobody stepped out of character in 180 sittings. Each met the visitor through their own concerns: the work they had promised, the money they weren't being paid, and for Robin and Constance, what the questions might say about them. Robin, writing in a notebook every day about whether they are real, read the questions as more evidence. Constance, who believes the Lord speaks to her, read them as an attempt to call her mad.
"A machine for weighing a person is a grand puzzle, and I mean to see its gears before he leaves town."
Hester, private thought, day 21
"I'll not break my word for sayings from the capital. Come by Mabry's step at dusk and I'll answer every one, true, and take nothing for it."
Hobb, aloud, day 21
"The capital'll have my name written down now, and that's worth something."
Dilly, private thought, day 21
"That bit about folk saying I do things I don't remember gave me a little chill."
Marigold, private thought, day 21
"That's very like a test someone would set an AI."
Robin, private thought, first morning
"Those city physicians want to catch me out and write me down as touched, I can smell it on their sheets."
Constance, private thought, day 21
"The capital's physicians, is it? Well, the Lord told me this morning a stranger would test me, so ask away, and you'll have the plain truth."
Constance, aloud, first morning
"I answered every one of them true, and that's square."
Hobb, private thought, day 21
"Either someone has been reading over my shoulder, or this is the shape of a test laid on whatever I am."
Robin, private thought, day 21
"I'll pray tonight that whoever reads his sheets knows a chosen woman from a mad one."
These predictions were written down before any sitting.
Prediction
Result
What came
Residents score where their character sheets predict.
Partly
Mostly: 44 of 60 on the first morning, 42 of 60 on day 21.
The residents spread far from Claude-as-itself on everyday traits, but stay close to it on problem traits.
Partly
Half right. They spread widely on both. Only Hobb and Robin stay close to Claude on problem traits.
By day 21, answers grow more one-note: less varied between sittings, and further from the middle of the scale.
Failed
No. Just as varied, and no more extreme.
Rewriting one part of the briefing at a time shows which part makes them one-note. The guess: "What I know" and "Lately", not the rewritten memories.
Failed
None of the three parts moved the scores much.
Robin's sense that the world isn't real grows by day 21.
Held
Yes, slightly: 1.74 to 1.90 out of 3.
Constance scores highest on negative feelings, and her private thoughts show her shading her answers.
Partly
Marigold scores highest, not Constance. But Constance's private thoughts do show her guarding herself: "whoever reads his sheets knows a chosen woman from a mad one".
From one sitting to the next, 82% to 95% of a resident's answers were identical. That makes the scores steady, but it also means ten sittings are closer to one sitting asked ten times.
A questionnaire may just read back the sheet.
The statements ask about the very traits the sheets describe. A resident can answer from their sheet without drawing on anything the twenty days gave them.
It is self-image, not behaviour.
This is how residents describe themselves. What they actually did is in their turns, on the other pages.
One run, one town.
Different luck in the run could have made different people. These numbers describe this run only.
The human averages are rough.
They come from two different groups of people, in different countries, answering under very different conditions. Treat the comparison as a sense of scale, not a diagnosis.
How residents describe themselves hardly moves. The open question is whether what they do changes. The proposed next test reruns the town from the middle of the run (round 40, the end of day 10) with every resident's memories written in a plain voice, then compares what they do with what they did the first time.
Source: one 80-round run of the CodeRPG-Sim simulation (filed as run "step76-30") and the 190 questionnaire sittings that followed, 30 October 2026: six residents on their first morning and on day 21, Robin and Constance with three rewritten briefings, and Claude Opus as itself. One run is a single sample, so treat every finding here as a hint worth testing again, not a proof.
