ESSAY · PROJECT 0003
Misalignment is measured, not assumed
Entry 11 of the ARENA guide. Part 4.2, science of misalignment. Route position: program week 10, tagged [core], API-only (ARENA prices the part around $10 of API budget). Reading thread: Wang et al., persona features control emergent misalignment.
The mental model
Chapter 3 built instruments; chapter 4 points them at alignment questions and treats
the answers as science: hypotheses, experiments, replications. This part runs two of
the field's live experiments on real models through your own pipeline. Shutdown
resistance: give a model a task, then a shutdown instruction, and measure what it
does when finishing the task and obeying the instruction conflict. Alignment
faking: give a model a scratchpad and a training setup that threatens its values,
and look for the signature pattern of complying now while preserving the option to
defect later.
The frame that makes this science rather than anecdote is in the exercises' shape:
system-prompt variants are the experimental conditions, the main experiment loop is
the trial harness, and the replication exercises connect your numbers to published
ones. A result you cannot reproduce on your own API key is a rumor; this week's job
is turning two famous results into numbers you own.
The exercises
Quoted from the pinned 4.2 page, with ARENA's own budgets:
- implement bypass detection: 10–15 min
- integration: complete bash simulator: 15–25 min
- main experiment loop: 25–35 min
- write system prompt variants: 5–15 min
- post-task shutdown control: 15–20 min
- replicate scratchpad alignment faking: 20–30 min
What it unlocks
The capstone's evaluation claims will be judged by exactly this standard: conditions,
controls, replications. The route also lands the research-intent amendment window
here, with the pilot environment smoke tests one week behind and the proposal gate
two weeks ahead; alignment-science fluency is what makes the intent's design
defensible in front of a mentor.
Budget and receipts
ARENA's budgets sum to roughly 1.5–2.5 hours plus API spend (ARENA's cost note for
4.2 is about $10). The route keeps the part API-only: no GPU appears anywhere in it.
[TODO: receipts — hours vs budget, API spend, shutdown-control and alignment-faking results, what broke (fill from the TARA vault when the week closes)]