Demonstration Suite¶
psychscanner-primal doesn't ship its own multi-experiment demonstration
suite — it's the slim, Hub-optimized distribution (see Using
psychscanner-primal). For real, run
demonstrations of the underlying task/agent machinery, see
psychscanner's
examples/demonstration_suite/
(rendered at
saurabhr.github.io/psychscanner/examples/demonstration_suite/):
- Reward Task — swapping agents into the same task harness. Complete.
- Association Memory — trial structures and feedback types. Skeleton.
- Personality Survey — persona conditioning, stateful memory. VVIQ-16 arm complete.
- VLM Task — the same machinery on vision-language models. Partial.
- Extracting Internals — activation extraction and steering. Skeleton.
- Advanced Demonstration — Othello-GPT, ROME, CCS. Othello-GPT complete.
- Introspective Self-Report — instilled preferences + self-report, with a psychscanner-primal integration point (see below). Complete.
- Prospect Theory Planning — risky-choice gambles, 3 agent architectures. Complete.
The one with a primal integration¶
07_introspective_selfreport
is the only demo that touches this package directly: its decision trials
(the part with a real per-trial correct/incorrect signal — this package's
own inclusion bar) are meant to ship as a standalone task card,
examples/tasks/introspection_weights_demo.json, built by that demo's
simulation/build_primal_task_card.py and runnable through this package's
own ExpCard/ScannerModel/task_library pattern (see Using
psychscanner-primal). That file is not
currently present in this repo's examples/tasks/ (checked 2026-08-19) —
re-run the build script in the psychscanner repo if you need it here.
What this package ships instead¶
The task cards actually bundled in examples/tasks/ (rm_*, pal50) are
narrower than the demonstration suite above — single feedback-scored task
cards, not full multi-condition studies. See
Using psychscanner-primal and
examples/tasks/README.md
for what's here and how to run it.