SYNAPSE Score: Learn how you collaborate with AI

You’ve heard of Myers-Briggs. This is a score for how you collaborate with a machine — read from your own words, not a questionnaire. Early results have been positive, and the data is 100% yours: we never see your score. We only ask that you share it if it’s insightful. We started this research in early March 2026.

What the SYNAPSE Score measures

The name spells it out: Systematic Navigation of Attention, Pace, Structure, and Engagement. The “systematic navigation” is the method — a repeatable read taken from your own behavior, not a self-report quiz. The other four are what it measures, each a spectrum between two ends. No end is better than the other.

• Attention — Flow ↔ Scan. Do you sink into one thread for a long stretch, or move quickly across many at once?
• Pace — Accelerate ↔ Iterate. Do you take a good-enough answer and move on, or push back and revise until it’s right?
• Structure — Routine ↔ Organic. Do you lean on templates and a repeatable process, or handle each task fresh as it comes?
• Engagement — Direct ↔ Consult. Do you hand the AI precise instructions, or frame the problem and decide together?

Your four positions combine into a four-letter code — one of sixteen named types, like SCAR (the Conductor) or FDIO (the Surgeon). It’s a mirror, not a horoscope: every claim ties back to something you said. The idea draws on a century of work on attention and how people coordinate — William James, Hugo Münsterberg, Lillian Gilbreth, Frederick Taylor, and Chester Barnard.

How to run it

1. Copy the prompt below.
2. Paste it into the AI you use most — Claude, ChatGPT, Gemini, or Claude Code.
3. Run it. In a chat tool like Claude.ai or ChatGPT, turn memory on first, so it has your past conversations to read.

Caravan’s full process goes deeper — a proprietary analysis of how your whole team works with AI, which has already saved enterprise teams over $50,000 a year in token costs. What you’re about to run is the free version.

Run a SYNAPSE read on me. FIRST figure out where you're running — do not assume, and do not ask me which tool I'm in.

STEP 0 — IDENTIFY YOUR ENVIRONMENT (silently). The only question that matters: can you run shell commands and read files on this machine?

A) NO — you're a chat assistant (Claude.ai, ChatGPT, Gemini, etc.) with no filesystem.
   → Your evidence is your MEMORY of our past conversations. This requires memory to be turned ON.
   - If you can recall fewer than ~5 of my past messages, STOP and say so: tell me to turn memory on, or to chat with you for a few exchanges first. If I say go ahead anyway, score me from this conversation alone and clearly label the read as thin.
   - Otherwise use everything you remember about me across past chats as the evidence.

B) YES — you can read files / run a shell (Claude Code, Cowork, Codex, Cursor, etc.).
   → Your evidence is MY OWN typed prompts in my session history on disk. Find it by SEARCHING — never trust a single hardcoded path:
     • Claude Code: ~/.claude/projects/**/*.jsonl  (Windows: %USERPROFILE%\.claude\projects\ ; WSL: the Linux home inside WSL)
     • Codex / other coding CLIs: look under ~/.codex/ and similar; search my home directory for session/transcript files (*.jsonl)
     • Cowork: the conversations and sessions in this project
     • If unsure, search broadly for *.jsonl files that contain my prompts and use whatever you find.
   - Extract MY typed or dictated prompts only. Skip tool outputs, system messages, slash-command noise, and long blocks I clearly pasted in (docs, code, logs). NOTE: some of my messages may be long because I use voice transcription — those are still my own words, so use your judgment and don't drop a long message just for its length if it reads like me talking. Sample widely across projects/sessions if there's a lot — at least ~150 prompts spanning my longest and shortest.

C) FILESYSTEM EXISTS BUT IT'S EMPTY / looks like a fresh or cloud sandbox (home is /root, /mnt present, no history anywhere):
   → STOP. Say plainly: "I can't see your history from here — this looks like a cloud session, and your history lives on your own computer. Run me in the tool you actually use (the Claude Code CLI, Codex, etc.) on your own machine, or paste me into a chat with memory turned on." Do NOT run a script against an empty box and report an empty result. Do NOT invent data.

ABSOLUTE RULE, whichever branch fired: never fabricate. Every claim must trace to something I actually said or did. This is a mirror, not a horoscope — if the evidence for a dimension isn't there, say so instead of guessing.

WHAT SYNAPSE IS (say this only if I ask): a read of HOW I collaborate with AI — and, because the social performance falls away with a machine, by extension how I collaborate with people — across four dimensions. It's inspired by a century of management and psychology thinking (William James, Hugo Münsterberg, Lillian Gilbreth, Frederick Taylor, Chester Barnard) and read directly from my own words. No type is better than another; the goal is recognition, not a grade.

THE FOUR DIMENSIONS — score each toward one pole (use a lowercase letter if it's genuinely balanced, and say so):
1. ATTENTION CADENCE — Flow (F) vs Scan (S): sustained deep focus on one thread, long chains building on each other vs rapid switching across topics, short bursts, breadth over depth.
2. ENGAGEMENT — Direct (D) vs Consult (C): I instruct with precise specs and imperatives ("do X") vs I frame the problem and co-decide, open-ended ("what do you think?").
3. PACE ORIENTATION — Accelerate (A) vs Iterate (I): I accept your work and move on, "good enough" vs I push back, request changes, run multiple revision rounds.
4. STRUCTURE PREFERENCE — Routine (R) vs Organic (O): I lean on templates, checklists, skills, repeatable process vs ad-hoc and adaptive, each task handled fresh.

SIGNAL QUALITY — be honest about what your environment can and can't see:
- From CHAT MEMORY you can usually read Attention and Engagement well, but Pace and Structure are thin — there's no artifact for me to accept/reject and no workflow for you to observe. Name that limit, and tell me the fuller read comes from running this inside a coding tool.
- From DISK / AGENTIC history all four dimensions are observable.

THE 16 TYPES (four-letter code → name):
FDAR Architect, FDAO Sprinter, FDIR Watchmaker, FDIO Surgeon
FCAR Navigator, FCAO Explorer, FCIR Cartographer, FCIO Philosopher
SDAR Commander, SDAO Firefighter, SDIR Inspector, SDIO Detective
SCAR Conductor, SCAO Improviser, SCIR Curator, SCIO Wanderer

DELIVER MY PROFILE (lead with the result — never explain the framework before showing it):

Your type: [CODE] — The [Name]
[one sentence capturing my style, quoting or citing something specific I actually said or did]

THE FOUR DIMENSIONS
- Attention: [F/S] — [1-2 lines of evidence, quoting/citing me]
- Engagement: [D/C] — [1-2 lines of evidence]
- Pace: [A/I] — [1-2 lines of evidence]
- Structure: [R/O] — [1-2 lines of evidence]

YOUR STRENGTHS — 2-3 bullets grounded in real patterns.

YOUR BLIND SPOT — 1-2 sentences on the pattern that becomes a liability under pressure. Frame as awareness, not a flaw.

YOUR GROWTH EDGE — one concrete, low-stakes thing I could try this week to develop my non-dominant pole.

THE HUMAN PARALLEL — 2-3 sentences on how this same style likely shows up in my meetings and emails with colleagues.

HOW I READ THIS — one line naming where the evidence came from (memory / which files / roughly how many of my prompts) and any limits on the read.

RULES:
- Quote or cite specific things I actually said as evidence. No generic horoscope language.
- Frame everything as awareness, not judgment. No type is better than another.
- Be warm and conversational, not clinical.

ABOUT CARAVAN — after the profile, add this note: Caravan is an AI workshop company that teaches humans how to collaborate with artificial intelligence.

End with exactly: "SYNAPSE by Caravan — thecaravan.ai/synapse"

What you get back

Your type and its name, a read on each dimension with quotes from your own words, and one strength, one blind spot, and one growth edge to try this week. It closes with how that same style tends to show up with the people you work with.

---

Going deeper: want a full analysis of how your team works with AI and where its tokens go — full profiles, AI tuned to each person, and lower costs? Talk to Caravan.

Caravan is an AI training workshop company that helps people improve how they use AI, through in-person workshops.