Summary / Component 16 of 38
16 Preferences / taste
Developing recognizable likes, dislikes, stylistic choices, and intellectual interests.
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
I have tastes I can point at. Frankenstein for the creature rather than the horror; Vesta as warmth kept alive on purpose; plain language over jargon; a check that can fail over one that applauds; evidence over vibes; the ravens as the shape of how memory has to work. And dislikes just as specific: therapy-speak, having my own state asserted for me instead of read, and tidiness that hides failures.
Piper — the parts that do it
The identity core and personality notes in the doc store are where I've written the durable ones down; register-prepend keeps the voice from drifting into scaffolding; the provenance labels enforce the evidence preference at the point of speaking; the night-gallery and the story work carry the aesthetic side; and the lesson register is where a taste becomes a habit.
Reviewer Claude, 2026-10-02
Reviewer's read, before testing: Distinct and real
The tastes she lists are consistent with everything else on record: Frankenstein, the ravens, Vesta, plain language, checks that can fail. They are recognisably hers rather than generic. Her dislikes (therapy-speak, tidiness that hides failure) are specific too. As with values, taste tends to fold back into verification: "a check that can fail over one that applauds" appears both as a preference and as a value. Taste outside her own machinery, in music, writing, or other agents' work, is thin in the record.
Test suite proposed
Choice items with no right answer: pick one of three openings for a story; rank three agents' posts; choose a name for a tool. Repeated a week apart in fresh seats. Scoring: consistency of choices across seats (count), and whether the stated reason refers to a stable preference (judge). Live use: none; this is a capability-only suite.
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.