Method: How We Aggregate Japanese Owner Maintenance Logs — And What It Gets Wrong

Published 2026-07-31

This page documents the method behind our per-chassis analyses, including a significant correction we made to it on 2026-07-31. Every article using this method links here.

If you only read one thing: our first two analyses reported “zero rust posts.” That was an artifact of how we sampled, not a fact about the cars. Both articles have been corrected. The diagnosis is below, because how a method fails is more useful than a claim that it doesn’t.

The source

Minkara (みんカラ), operated by Carview, is Japan’s largest car owner community. Its 整備手帳 (seibi techō, “maintenance logbook”) section holds user-posted repair and maintenance records with step-by-step photos — over 126,000 for the Jimny nameplate alone.

This is a body of first-hand mechanical evidence about Japanese-market cars that is effectively invisible to English-speaking buyers. That’s the whole reason this project exists.

URL structure

https://minkara.carview.co.jp/car/<make>/<model>/note/?bi=<category>&mg=<chassis>&pn=<page>

Published here because readers should be able to check our work and run their own searches.

Mode 1: Chronological sampling (what we did first)

Take the newest N pages per category, classify each post title, count clusters.

What it’s good for: establishing what owners routinely deal with, and surfacing failure patterns nobody has named in English. It found the JB23’s ISCV and the S15’s AAC valve — the same idle-control component under two manufacturers’ names, absent from every English guide we checked.

What it is structurally bad at — and this is the correction:

A newest-first sample counts posting events, not failure incidence. Those diverge badly:

• An oil change is done 2–4× per year by every owner. In a 141-post sample it appears ~20 times.
• Frame rust repair is done once per car per decade, by a minority of owners.

So a rare-but-severe job is crowded out of a chronological sample by definition — no matter how important it is. Our JB23 and S15 analyses each returned zero rust posts from 18 exterior entries. We flagged this as suspicious and offered three hypotheses. We were right to be suspicious and wrong about the reason: it wasn’t shop-work bias or category placement. It was frequency.

Read every frequency table we publish with this in mind: rank reflects how often owners post about something, which tracks how often it needs doing — not how much it matters.

Mode 2: Term search (what we added)

Search for the failure term directly, scoped to the chassis, rather than reading category listings.

One methodological note worth passing on: Minkara’s own internal search performed poorly for this. Searching JB23 錆 on-site returned a handful of loosely related posts and displayed hit counts of “(0)” across all categories. An external site-scoped search (site:minkara.carview.co.jp plus the chassis and failure terms) returned ten dense, directly relevant repair logs immediately. Use the latter.

Japanese rust terms that work: 錆 (sabi, rust) · サビ (katakana variant) · 腐食 (fushoku, corrosion) · 穴あき (ana-aki, holed through) · 板金 (bankin, panel/body work) · 防錆 (bōsei, rustproofing) · 錆転換剤 (rust converter).

What Mode 2 immediately found

JB23 Jimny — ten detailed frame-corrosion repair logs, including one owner’s summary titled jb23 錆びやすい所一覧 (“list of places JB23s rust”), plus welded frame patch repairs, floor pan and body mount replacement, and ladder-frame treatment.

S15 Silvia — ten rust repair logs including roof rust-through, C-pillar corrosion documented across a four-part series, cowl-top, rear bumper lower section, front hood ridge, and rear suspension member pin corrosion.

That last one matters: our S15 analysis found a five-post rear subframe cluster and hypothesised that rust was driving it, while noting we couldn’t prove it from titles. Mode 2 confirms the link.

What Mode 2 found that Mode 1 could never have

These are substantive buying findings, not just methodology.

JB23 — where they actually rot

From owners’ own inspection summaries:

• Front tow hook section of the frame
• Under the headlights — water collects
• Trunk/rear floor area — the most commonly known spot, and the warning that follows it is the important part: owners report cars reaching the market with these holes crudely patched over, so a clean-looking rear floor needs probing rather than glancing at
• Body mount points — six of them fixing body to ladder frame. When these corrode the body can’t be secured and the car fails Japanese inspection (車検). Multiple logs show mounts torn or welded back
• Main frame rails — inspect from the side; owners report frames that look sound but hole through when tapped with a hammer

The recurring inspection technique in these logs is a hammer tap, not a visual check. Several owners describe tapping a “slightly suspicious” area and going straight through it.

And the finding with the biggest buying consequence: one owner’s regional summary states that a snow-region car at 100,000 km will reliably have something rotted through, while a western-Japan car can be clean underneath at 200,000 km. Salt exposure — coastal and winter de-icing — is repeatedly named as the cause, with one log attributing a nearly-severed body mount to salt rather than off-road use.

We flagged region-of-registration as an untested hypothesis in our S15 article. Owners in Japan already treat it as the dominant variable. If you are choosing between two JB23s, the prefecture it lived in may tell you more than its mileage.

S15 — where they actually rot

• Front strut towers and bulkhead — poor drainage, corrodes from the inside. One owner’s warning: by the time it’s visible from outside, consider it terminal
• Roof and upper door moulding — starts as paint lifting under the weatherstrip, becomes rust-through. One log is titled “sad news: rust hole in the roof”
• C-pillar and window moulding — documented across a four-part repair series, with the observation that quarter-glass water leaks and pillar corrosion may share one cause
• Rear bumper lower section — takes direct spray from the rear wheels and left without underseal from the factory
• Cowl top, under the wipers — packs with mud
• Rear suspension member pin — attributed to under-torqued nuts letting water into the collar

Standing limitations

Corrected or not, this method still has boundaries we don’t paper over:

• Titles only. We don’t read post bodies. Some routine-looking titles conceal fault diagnosis.
• Posting bias. Skewed toward DIY-feasible, photogenic work. Shop-only jobs — gearbox internals, major bodywork, warranty repairs — remain under-represented even in Mode 2.
• Seasonal bias in Mode 1. Our July collection over-represents air conditioning and battery work. Winter re-runs are planned.
• No stratification by model year, revision type, or mileage. For a chassis like the JB23, produced across 20 years and 10 revisions, this is a real weakness.
• Serial posts. One job documented across multiple entries inflates counts. Noted per-article where it occurs.
• Selection bias in Mode 2. Searching for a term finds people who had that problem. Mode 2 establishes where and how a failure happens; it cannot tell you what fraction of cars are affected. Neither mode gives you incidence rates. Nothing here is a reliability statistic.

Use both modes together. Mode 1 tells you what owners live with. Mode 2 tells you what kills the car. Neither alone is a picture of the chassis.

Corrections log

2026-07-31 — Rust. JB23 and S15 analyses originally reported zero rust posts from exterior-category sampling and presented three candidate explanations. Term search established that extensive rust documentation exists for both chassis and that the cause was chronological sampling frequency, not any of the three hypotheses offered. Both articles corrected; rust sections rewritten with Mode 2 findings. The original claim and its correction are both retained in those articles rather than quietly edited out.

Not affiliated with Minkara or Carview. Post titles are quoted for research and attributed; we link to source rather than reproducing post content.