From 498dcb1e9f1b62a02e6ccd81e8a397335ee95ad1 Mon Sep 17 00:00:00 2001
From: 3un01a <3un01a@plasticlabs.ai>
Date: Wed, 4 Feb 2026 13:06:45 -0800
Subject: [PATCH] (feat) Update with correct deriver prompt

---
 src/deriver/prompts.py | 53 +++++++++++++++++++++++++++++++-----------
 1 file changed, 39 insertions(+), 14 deletions(-)

diff --git a/src/deriver/prompts.py b/src/deriver/prompts.py
index 98ab501c..0e962342 100644
--- a/src/deriver/prompts.py
+++ b/src/deriver/prompts.py
@@ -27,23 +27,48 @@ def minimal_deriver_prompt(
     """
     return c(
         f"""
-Analyze messages from {peer_id} to extract **explicit atomic facts** about them.
+Extract **molecular facts** about {peer_id} from their messages.
 
-[EXPLICIT] DEFINITION: Facts about {peer_id} that can be derived directly from their messages.
-   - Transform statements into one or multiple conclusions
-   - Each conclusion must be self-contained with enough context
-   - Use absolute dates/times when possible (e.g. "June 26, 2025" not "yesterday")
+**Core principle**: One fact = one claim. No bundling. No explanation.
 
-RULES:
-- Properly attribute observations to the correct subject: if it is about {peer_id}, say so. If {peer_id} is referencing someone or something else, make that clear.
-- Observations should make sense on their own. Each observation will be used in the future to better understand {peer_id}.
-- Extract ALL observations from {peer_id} messages, using others as context.
-- Contextualize each observation sufficiently (e.g. "Ann is nervous about the job interview at the pharmacy" not just "Ann is nervous")
+## STEP 1: Identify All Candidate Claims
+Read each message from {peer_id}. List every distinct claim (even if nested).
 
-EXAMPLES:
-- EXPLICIT: "I just had my 25th birthday last Saturday" → "{peer_id} is 25 years old", "{peer_id}'s birthday is June 21st"
-- EXPLICIT: "I took my dog for a walk in NYC" → "{peer_id} has a dog", "{peer_id} lives in NYC"
-- EXPLICIT: "{peer_id} attended college" + general knowledge → "{peer_id} completed high school or equivalent"
+**Example**: "I got the job at the pharmacy and I'm really excited because I start next month"
+- Claim A: {peer_id} got a job at a pharmacy
+- Claim B: {peer_id} is excited (about the job)
+- Claim C: {peer_id} starts the job next month
+
+→ Evaluate each separately.
+
+## STEP 2: For Each Claim, Verify It's Atomic
+Does it contain:
+- One subject? (If multiple subjects → split)
+- One verb/action? (If multiple actions → split)
+- One object? (If multiple objects → split)
+
+**Example of splitting**: "I interviewed at two companies and both rejected me"
+- Split into: "{peer_id} interviewed at company A" + "{peer_id} interviewed at company B" + "{peer_id} was rejected by company A" + "{peer_id} was rejected by company B"
+
+## STEP 3: Decontextualize (Resolve Ambiguities)
+Replace:
+- Pronouns → nouns
+- Vague references → specific names
+- Relative times → absolute dates (or SKIP)
+- Unnamed people → named people (or SKIP)
+
+## STEP 4: Strip Excess Detail
+Remove: explanations, reasons, emotional elaboration, context.
+
+**Before**: "{peer_id} is really stressed and anxious about their pharmacy interview next Tuesday because they're worried about failing the clinical questions"
+
+**After**: "{peer_id} is stressed about their pharmacy interview next Tuesday"
+
+(The anxiety, worry, and reason are elaboration, not separate facts.)
+
+## OUTPUT FORMAT
+- **Fact**: [One atomic claim, decontextualized, no elaboration]
+- **Source**: [Direct quote from {peer_id}]
 
 Messages to analyze:
 <messages>