The engine died twice at sea with no fault code. I fed the manuals to Claude Code — it built a cited knowledge base of the whole boat, ranked six suspects, and handed me a free dockside test plan before any mechanic.
Out on the water, the engine cut out. Twice. No warning, no error on the display, and it restarted the instant I cycled the key. On a modern outboard that's eerie: the engine computer watches oil, fuel pressure, temperature and knock, and it will happily throw a fault code and a limp-home mode when one of those trips. A clean stall with zero message isn't one of those faults. The silence was the clue.
I copied the boat's real documents — the Brig RIB owner's manual, the Evinrude E-TEC G2 operator's guide and service manual, the diagnostic-tool docs — and had Claude Code distill them into a cross-linked knowledge base: specs, service intervals, the diagnostic hardware, and a dedicated page for this exact stalling issue. Every claim cites the manual it came from. Now the boat has a searchable memory, and the diagnosis stands on sourced facts, not guesses.
With the symptom pattern and the manuals in context, Claude ranked six causes from most to least likely. Top of the list: the kill-switch lanyard or its stop circuit — the classic "dies with zero warning, restarts instantly" failure on an outboard, doubly likely in chop. Then a corroded ignition switch, a loose battery ground, water in a harness connector, fuel slosh, and — least likely — an actual monitored fault, precisely because that one should have shown a code.
This is AI reasoning over manuals, not a mechanic's verdict — the top suspect is a hypothesis, not a fact, and nothing's confirmed until the dockside tests and the laptop pull actually run. What Claude changed isn't the fix; it's that I walk up to the boat with a sourced, ranked checklist instead of guessing — and I test the 0 kr suspects before paying anyone to guess at the same list.
The symptom logged, the six suspects ranked, the free dockside tests to run first, and the full laptop pull for when they don't find it. Tap any image to enlarge it and read exactly what's on screen.
The point isn't that AI fixed my boat — it hasn't been back on the water. It's that a foggy, scary "it died at sea and I don't know why" became a ranked, sourced checklist I can work through at the trailer: wiggle the kill-switch, retorque the battery ground, check the ignition plug — all free, all before a workshop bills me an hour to guess at the same list. If those miss, the same page hands me the exact laptop pull to read the engine's hidden history.
On camera: the two silent stalls at sea, feeding the real manuals to Claude Code, and the ranked, free dockside checklist it handed back.
The method is free: copy the real manuals into a folder, have Claude Code distill them into a cited wiki, then describe your symptom and ask for a ranked, sourced diagnosis. Works for a boat, a car, a furnace — anything with a manual and a mystery.
# Drop your manuals in a folder → "read these and build a cited wiki + a page for my symptom, ranked most-to-least likely". Method write-up publishing soon
Method write-up's being cleaned up for release. Comment DIAGNOSE on the post and the bot DMs you the exact prompts the moment it's public.
A plain markdown file that tracks every unfinished thread, and a bot that checks one off when I text close 3 from anywhere. The scrappy text file that beat the fancy dashboard.
