The thinking you used to do before reaching for AI.

Is it still there?

Zanshin is a reasoning benchmark for engineers who use AI constantly — built to notice if independent thinking is quietly getting weaker underneath the speed.

[ FAULT_DETECTION_MATRIX ] STATUS: CRITICAL

You reach for AI before you've even thought about it. You use its output without checking it. You don't remember what it wrote because you never really processed it. None of this feels bad in the moment — it feels like speed.

What it measures

[ SEC_03 // SYS_DIAGNOSTICS ]

DOMAIN_01 // REF.Y2K

Syllogistic reasoning

deductive logic

DOMAIN_02 // REF.SIGMA

Fermi estimation

sizing problems under uncertainty

DOMAIN_03 // REF.DIFF

Argument evaluation

spotting weak reasoning, like code review

DOMAIN_04 // REF.CUBE

Causal reasoning

root-cause thinking, debugging logic

EXT_MODULE // REF.LIVE

Browser extension

lives inside ChatGPT, Claude, Gemini, Cursor — asks one quiet question after AI answers, skippable in one tap

See it in action

[ SYSTEM_PREVIEW // LIVE_BUILD ]

SCREEN_01 // BENCHMARK SESSION ACTIVE

Active benchmark session — a timed multiple-choice question testing syllogistic reasoning

5-minute benchmark

A real reasoning question — timed, no hints, no AI. Four domains tested: logic, estimation, argument analysis, causal thinking.

SCREEN_02 // DASHBOARD TRACKING

Reasoning dashboard showing domain correlation network with four score nodes and domain status list

Your reasoning dashboard

Four domains tracked separately. You see exactly which area is weakening — not a single blended score that hides what's actually happening.

SCREEN_03 // SCORE_REVEAL PATTERN FOUND

Score reveal screen showing performance trend chart, 98/100 risk score, and identified pre-delegation behavioral pattern

When the pattern lands

Only fires when a behavior pattern and a dropping score line up across sessions. No false alerts. When it says something changed — something actually changed.

SCREEN_04 // EXTENSION EXT ACTIVE

Extension activity panel showing AI dependency risk score, offload rate, pushback rate, and conversation history signals

Extension signals

The browser extension runs silently inside your AI tools. It tracks how often you offload vs. push back — adding real-world weight to your benchmark score.

SCREEN_05 // EXT_UI EXT LIVE

Zanshin browser extension UI showing the in-context reasoning prompt inside an AI tool

Inside your AI tools

One quiet question, right after the AI answers. No interruption — just a tap to engage or skip. It lives where you already work.

How it works

Take a 5-minute benchmark, three times a week

See your score and trend by domain, not one vague number

Get a specific insight when a pattern and a dropping score actually line up — never a guess

[ EXT_MODULE // OPTIONAL ]

Optional, not required. The benchmark works fully on its own. Install the extension and it gets sharper — it sees what you actually do with AI, not just how you perform in a 5-minute test.

Zanshin measures. It doesn't diagnose, and it doesn't promise to fix anything.

It's a place to keep your reasoning active — the same way a gym doesn't promise to make you healthy, it gives you somewhere to do the work.

When a specific domain slips, you get one targeted practice challenge — not a lecture, an exercise.