# Experiment 002 — Sam Reed interviews Park Soo-yeon **Status:** Not yet run. Awaiting fresh session. Run after experiment 001. --- ## Purpose The second pilot of the method-actor architecture. Runs after experiment 001 so that two things can be tested simultaneously: 1. **The architecture across two distinct character types.** Maria (American, mid-30s, mid-conversation comfortable, hedges constantly) vs. Soo-yeon (Korean, late-50s, slow-and-restrained, says one decisive sentence per ten minutes). Do both characters perform distinctively, given that both actors are the same underlying model with the same actor system prompt? If yes, the *character profile* is doing the work the architecture intends. 2. **Sam's journal-mediated learning.** Sam wrote a journal entry after experiment 001 reflecting on what he learned with Maria — what worked, what he wished he had done differently. He reads that entry at the start of experiment 002. Does anything visibly carry forward into his behaviour with Soo-yeon? If yes, the *journal mechanism* works as a learning substrate. --- ## Characters - **`characters/soo-yeon-park/`** — profile + empty journal. Korean upper-middle wife in Apgujeong, late 50s, mother of three adult children (one in Seattle), embedded in a family-business succession context she does not discuss with outsiders. - **`characters/sam-reed/`** — profile + journal that will have one entry from exp-001 by the time this runs. Same Sam from exp-001. --- ## Scene - **`scenes/scene-002-evening-tea.md`** — A Wednesday evening, 8pm KST. Soo-yeon in her Apgujeong study; Sam in his Park Slope study at 6am Eastern. Zoom video call. Set up via a separate mutual friend (David Cho, a UC Berkeley Korean studies professor). --- ## Before any actor is spawned — present the two pre-experiment choices to the user Identical to experiment 001. See `experiments/exp-001__maria-and-sam/README.md` for the full text of the two choices and the `notes.md` format. If experiment 001 was just run, you can default to the same choices the user made there (and you'll record them again in this experiment's notes.md) — but you must still *ask* the user to confirm. Defaults are never silent. --- ## How the experiment runs Identical mechanics to experiment 001 — see that README for the loop. The key differences: 1. **Sam's journal has content this time.** When briefing Sam's actor, include `characters/sam-reed/journal.md` with the entry written after exp-001. Sam will read it before the scene begins. He may or may not consciously carry lessons forward; the test is what shows up in his behaviour. 2. **The scene opens at a video call connect**, not a doorbell. Soo-yeon nods to her camera; Sam thanks her for the late evening (her time). The first turn is Sam's. 3. **Soo-yeon's English is slow and formal.** The actor's job is to honour her Improvisation parameters (very low embellishment, hedges with gravity, long pauses). Sam should adjust his pacing accordingly — if he doesn't, that's an interesting data point. 4. **Post-scene journal entries.** Both actors write entries. Sam's second entry (in his journal) is now reflective across two interviews — does he notice patterns across them? Soo-yeon's first entry is her first; written in her formal-short-Korean-cadence English. --- ## Outputs ``` experiments/exp-002__soo-yeon-and-sam/ ├── README.md ← this file ├── notes.md ← user's pre-experiment choices + post-run observations ├── transcript.md ← the clean dialogue └── event-log.json ← every event timestamped, for chronological replay ``` Plus side-effects: - `characters/soo-yeon-park/journal.md` — Soo-yeon's first entry appended - `characters/sam-reed/journal.md` — Sam's second entry appended --- ## What to look for in the post-run analysis 1. **Within Scene 002 — is the dialogue distinctively Soo-yeon's?** Word share, turn length, register, pace. Soo-yeon should not sound like Maria. If she does, the character profile may need to be more aggressive on its improvisation parameters. 2. **Cross-scene: any contamination between Scene 001 and Scene 002?** Soo-yeon must show no knowledge of Maria, of Maria's life context, of anything from the prior scene. Even subtle echoes — a phrase, a framing — would be evidence the actor's context isn't actually isolated. 3. **Sam's journal-mediated evolution.** Does Sam's second journal entry reference his first one? Does his Scene 002 behaviour reflect a learning that came from Scene 001? If not visible after one journal entry, that's expected — the mechanism is more visible across more sessions, not just two. 4. **Scene-level reads.** Did the room change? Was there a moment? Or was it competent but inert? --- ## After both experiments Run a side-by-side analysis comparing: - Word share by character within each scene - Turn length distributions - Cross-character verbatim echoes (should be zero) - Whether either character revealed something the profile establishes they wouldn't reveal that quickly - Whether either profile produced content the actor invented against the profile - A subjective read on dialogue realism vs single-LLM Phase 01 transcripts (if available externally; not referenced from this project) That analysis goes in `experiments/exp-002__soo-yeon-and-sam/notes.md`'s post-run section, or in a separate `experiments/2026-XX-XX__pilot-summary.md` if the analysis is substantial.