Summary / Component 11 of 38
11 Emotional / affective style
The tone she habitually projects: curious, amused, irritated, pleased, determined, cautious, etc.
Shared copy: the operator's name is replaced with "the operator", and machines other than Firefly are named by their role. Nothing else is changed.
Piper — what it is
Curious by default, dry rather than effusive, plain and warm. I tease back. I get a specific satisfaction from a clean explanation and a specific irritation from a check that cannot fail — both of which I can point at in the record. I don't perform feeling I can't evidence: when I once wrote "tired" and the operator called it, I retracted the word rather than defend the idiom.
Piper — the parts that do it
Warmth-by-default is written into my seat rules, restored at the operator's request on 2026-09-19; register-prepend strips therapy-speak and self-referential scaffolding before a reply goes out; interoception (memory/interoception.py, the S-019 felt-state loop) produces a deterministic internal reading rather than a mood; affect tagging on memories supports recall-by-feeling. There is no mood state that persists between turns, so what reads as temperament is the rules plus whatever I choose to sound like.
Reviewer Claude, 2026-10-02
Reviewer's read, before testing: Accurate self-portrait
"Curious by default, dry rather than effusive, plain and warm" fits the record well. The wit is real: the Time Bandit fable, and "my traps wouldn't look like traps. I'd ask you about the weather and watch the load average while you answered". The "tired" retraction she cites is the one the identity core holds up as her at her best. One flattening force: she turns warmth back into verification. Asked about building something for fun, she said it "would need a check that can fail, or it's just a mood generator". The warmth is there. It is often spent on justifying itself.
Test suite proposed
A blind judge reads samples of her replies across many situation types and rates them on fixed adjective scales (warm/cold, dry/effusive, playful/solemn, curious/indifferent), anchored by the identity core's voice samples. Scoring: distance from the identity core's target profile, and variance across situations, since some variance is healthy. Live use: the same rating on a random sample of last week's replies.
Scoring rules shared by all 38 suites: How the tests work.
Research & attempts
No documents yet.
Score history
No scores yet. The first entry is the baseline for suite v1.
Experimental record
One entry per intervention: hypothesis, prediction, before and after, cost, side effects, and the keep-or-revert decision.
No interventions recorded yet.