nishi code wiki / research / car diagnosis & repair
Fixing any car: diagnosis and repair planning at SOTA, and past it
Published · corpus slug: project-nishi-car-repair-sota-2026-08-05 · test vehicles: 2012 Hyundai Sonata 2.4 GDI (Theta II) and 2013 Toyota Sienna 3.5 (2GR-FE)
Labels: SHIPPING = in production use in the field · FIELD = owner/forum reports, counts without denominators · OURS = measured in this ecosystem · ANNOUNCED = vendor claim not independently measured.
Why these two vehicles decide the whole design
A head-gasket complaint is a symptom cluster, not a diagnosis: overheating, coolant loss with no visible leak, white exhaust, rough running. The two test vehicles were chosen because they punish naive tooling in opposite directions, and any planner that handles both correctly generalizes:
| 2012 Sonata 2.4 (Theta II) | 2013 Sienna 3.5 (2GR-FE) | |
|---|---|---|
| What the symptoms usually are | NOT the head gasket — a dying bottom end. Debris-starved rod bearings knock, then seize or exit the block [C4]. A gasket-first plan wastes a teardown on an engine that needs a block. | NOT the head gasket — a failed water pump, a ruptured rubber oil-cooler line, or a timing-cover weep [C7]. A gasket-first plan quotes a four-figure job for a three-figure fix. |
| The killer fact a tool must know | Recall campaigns + KSDS software + short-block warranty extension mean a correctly diagnosed failure can be repaired for parts cost $0 under campaign [C5]. | Toyota superseded its own rubber oil-cooler line with a metal update part [C8] — the canonical bad part → better part record. |
The SOTA diagnostic plan skeleton (any vehicle, any complaint)
| Stage | What happens | Why it is ordered here |
|---|---|---|
| 0 | Identity + campaign sweep. VIN → open recalls, TSBs, warranty extensions, class-action remedies. On the Sonata this single query can convert a $4,000+ engine job into a covered repair [C5]. | Free, instant, and can zero out the whole repair cost. Nothing else pays this well per minute. |
| 1 | Priors from failure data. Complaint + vehicle → ranked hypothesis set weighted by that vehicle's measured failure record (NHTSA complaints/recalls, aggregated-fix databases) [C11]. | The same complaint means different causes on different vehicles — the two test vehicles prove it. |
| 2 | Evidence-preservation first. Pull an oil sample BEFORE draining; photograph coolant state; do not keep running a knocking engine to reproduce the noise. | Some tests destroy the evidence other tests need. Ordering is a correctness property, not a style choice. |
| 3 | Cheap non-invasive ladder. Dipstick emulsion check, plug read, cooling-system pressure test, chemical block test [C3]. | Under $50 of consumables resolves most head-gasket questions before any disassembly. |
| 4 | Discriminating instrumented tests. Compression wet/dry, leak-down with escape-path listening, borescope [C3]. | Each test is chosen for what it RULES OUT, not what it could confirm. |
| 5 | Repair/replace/exit economics + upgraded-parts substitution. Labor-dominant jobs priced against used-engine swap and vehicle value; every replaced part checked against the parts-quality registry for a better-than-OEM option [C8] [C9]; post-repair verification gates (pressure hold, block test repeat, oil analysis at 500 mi). | A diagnosis without a decision is half a product; a repair that reinstalls a known-bad part design schedules the next failure. |
The head-gasket evidence ladder
Consolidated from the trade canon [C3]. Costs are consumer-tool street prices, field-reported ranges, not book figures (see gaps).
| Test | Cost | A positive means | What it rules out / traps |
|---|---|---|---|
| Dipstick + oil-cap emulsion | $0 | Coolant in oil (milkshake) | Short-trip condensation mimics light emulsion |
| Spark-plug comparison | $0 | One steam-cleaned plug → coolant entering THAT cylinder | Localizes the failure; a uniform read rules out per-cylinder intrusion |
| Cooling-system pressure test | $30–60 tool | Pressure decay 10–15 min with no external leak → internal path | Find external leaks FIRST or they masquerade as internal |
| Chemical block test (combustion gas in coolant) | $10–40 | Fluid color change → combustion gases in coolant | The single cheapest decisive HG positive; false negative if the leak only opens hot/under load |
| Compression, dry then wet | $30–50 tool | Two ADJACENT low cylinders → gasket bridge failure | Wet-recovery points to rings, not gasket |
| Leak-down + listen | $40–70 tool | Bubbles in coolant → HG/crack; air at intake/exhaust → valves; crankcase → rings | The escape PATH is the diagnosis; the percentage alone is not |
| Borescope | $20–100 | Washed/steam-cleaned piston crown | Confirms cylinder + guides teardown scope before any disassembly |
Vehicle playbooks
2012 Sonata 2.4 GDI — the false head gasket
Stage 0 dominates: recalls 15V-568/17V-226 (~470k 2011–12 Sonatas) for machining debris starving rod bearings [C4]; the KSDS software campaign adds bearing-wear detection (DTC P1326, limp mode) and Hyundai extended short-block coverage [C5]. Plan: VIN campaign check → oil inspection for metallic glitter → listen for cyclic knock → P1326 scan. Metallic oil or P1326 ends the head-gasket branch: the engine block is the repair, likely under campaign. Only a clean bottom end sends the overheat/coolant complaint down the standard HG ladder [C6].
2013 Sienna 3.5 2GR-FE — the head-gasket mimics
Internal HG failure is rare on this engine; the mimics are ranked: water pump (common, documented failures from 21k mi), rubber oil-cooler lines (pre-2013 design; rupture dumps oil down the front of the engine; metal update part exists), timing-cover weep, crank seals [C7] [C8]. Plan: pressure-test and INSPECT for external paths before any internal test; a block test that stays clean plus a wet front cover is a cooling/oil-line repair, not a gasket job. Field-reported: water pump swap ~4–5 h experienced-DIY [C12]; oil-cooler line parts under $80 DIY vs ~$400 shop [C8].
The parts-quality axis: rate every part, upgrade the weak ones
Two independent records prove parts are ratable and upgradeable, not interchangeable commodities:
- OEM self-supersession: Toyota's rubber oil-cooler line failed in the field and was replaced by a metal update part [C8]. The supersession record IS a quality verdict written by the manufacturer.
- Engineered-correction aftermarket: Dorman's OE FIX line exists to sell corrections to OEM design flaws — brittle plastic thermostat housings re-made in aluminum, warping plastic oil-filter housings re-made as all-aluminum kits [C9].
Where current tooling stops (the exceed opening)
The July-2026 commercial stack is strong on DATA and weak on PLANNING. ALLDATA/ProDemand/Identifix carry OEM procedures, labor times and millions of aggregated confirmed fixes [C1]; flagship scan tools add dealer-level access plus AI for cosmetic inspection and system topology [C2]. But: fix databases RANK likely repairs, they do not emit an ordered, costed, evidence-gated test plan; none of them integrate campaign/warranty economics into the diagnosis (the single highest-value step on a Theta II); none rate parts or plan upgrades; and their guided tests assume their own brand of hardware. The plan skeleton above is not in any of them as a product.
The Nishi exceed plan
| Rung | Organ | What it does | Falsifiable exit criterion |
|---|---|---|---|
| R1 | nx_vehicle_registry | Per-platform plane (year/make/model/engine): failure modes with complaint-count priors, campaigns, TSBs, warranty extensions; seeded from NHTSA public data [C11]. | Both test vehicles carry ≥5 failure modes, every mode citing a campaign/TSB/complaint source. No hand-asserted rows. |
| R2 | nx_diag_plan | Complaint + vehicle → ranked hypotheses + ordered test plan by information-per-dollar, campaign sweep always stage 0, evidence-preservation ordering enforced. | On both test vehicles: true cause in top-2; head-gasket-vs-mimic isolated in ≤3 tests / ≤$150 of tooling; BITE: fed zero measurements it must REFUSE a verdict. |
| R3 | nx_partsgrade | Parts-quality registry per the rating scheme above; weak part → buy/make upgrade path. | Flags the Sienna oil-cooler line and the Sonata short block from DATA alone (supersession + campaign records), not hand-seeding; every weak part maps to ≥1 upgrade. |
| R4 | corpus coverage | Seed the automotive crawl: NHTSA recalls/complaints/TSB indexes, owner-forum corpora; a miss is the best crawl signal. | The C10 ruler query returns ≥20/30 automotive docs; the starved query set goes green [C10]. |
| R5 | nx_carbench | The ruler: replay documented field cases with known outcomes through R2; score top-2 accuracy and tests-to-isolation; regression-gate every planner change. | Top-2 accuracy ≥80% on the case set; mean tests-to-isolation ≤3; gate wired before the first planner promotion, not after. |
Consumers already in the estate: nx_carflip_deal/nx_carflip_roi (repair-cost truth feeds flip ROI) and the household's own two vehicles.
Claims
| ID | Claim | Value | Label | Source | Source date |
|---|---|---|---|---|---|
| C1 | The professional repair-information stack is three-legged: OEM-procedure mirrors (ALLDATA), estimating/labor (Mitchell 1 ProDemand), aggregated confirmed-fix databases (Identifix Direct-Hit) | 3 legs | SHIPPING | 2026 rankings · gitnux | |
| C2 | Flagship scan tier 2026: Autel MaxiSys Ultra S2 (topology 3.0, AI image inspection) matches or exceeds Snap-on Zeus in most categories at less than half the price; Snap-on keeps the edge on US-domestic coverage + SureTrack fixes | <0.5x price | SHIPPING | 2026 buyer's guide | |
| C3 | The head-gasket trade canon is a 7-step cheap-first ladder: visual/emulsion, plug read, cooling pressure test, chemical block test, compression wet/dry, leak-down with escape-path listening, borescope | 7 steps | SHIPPING | ricksfree · 2carpros · engineschool | |
| C4 | Theta II defect: crankshaft-machining debris restricts rod-bearing oiling → wear → cyclic knock → seizure/rod exit; NHTSA recalls 15V-568 + 17V-226 cover ~470k 2011-12 Sonatas; consent order followed | ~470k vehicles | SHIPPING | NHTSA consent order · Consumer Reports | |
| C5 | Remedy layer: KSDS knock-sensor software detects bearing-wear vibration signature (sets P1326, limp mode) and Hyundai extended short-block warranty for affected Sonatas — a correct diagnosis can make the engine replacement a covered repair | P1326 + extended warranty | SHIPPING | Hyundai release · ricksfree | |
| C6 | On 2011-12 Sonatas, overheat/coolant-loss/knock presentations are statistically bottom-end, not head gasket; metallic oil or P1326 terminates the gasket branch | qualitative, aggregate | FIELD | The Drive · hyundaiproblems | |
| C7 | On the 2GR-FE, internal head-gasket failure is rare; the head-gasket mimics are water pump (documented from 21k mi), rubber oil-cooler lines, timing-cover weep, crank seals | 4 named mimics | FIELD | siennachat · carspecmn · RepairPal | |
| C8 | Toyota superseded the failure-prone rubber oil-cooler line with a metal update part; field-reported parts under $80 DIY vs ~$400 shop — an OEM self-supersession is a manufacturer-written part-quality verdict | $80 vs ~$400 | SHIPPING | siennachat · oil leak thread | |
| C9 | An engineered-correction aftermarket exists as a product category: Dorman OE FIX re-makes brittle plastic thermostat housings in aluminum and warping plastic oil-filter housings as all-aluminum kits, marketed explicitly as fixes for OEM design flaws | shipping product line | SHIPPING | Dorman 902-3035HP · 926-959 | |
| C10 | Our sovereign web index is starved for automotive diagnostics: the ruler query 'car repair head gasket diagnostic' returned 0 automotive-diagnostic documents in its top 30 (appliance/plumbing/roofing instead) — coverage, not ranking | 0/30 | OURS | nishi_search, this session | |
| C11 | NHTSA's complaint/recall/TSB data is public and queryable per vehicle — a free failure-frequency oracle; complaint aggregators (carcomplaints) and fix databases (Identifix) monetize exactly this signal | public API | SHIPPING | NHTSA datasets · carcomplaints | |
| C12 | 2GR-FE Sienna water pump swap is field-reported at ~4-5 h experienced-DIY; head-gasket R&R on this transverse V6 is major-disassembly, labor-dominant work weighed against a used-engine swap in the field | 4-5 h | FIELD | water-pump how-to · replace-or-fix thread |
Declared gaps
- No OURS measurement on a live car yet. Every diagnostic claim here is literature/field-sourced; the two household test vehicles have not been run through the ladder. That first run is R5's seed case set.
- Book labor times are UNVERIFIED — no ALLDATA/Mitchell license in the estate; all times and costs above are field-reported ranges, not flat-rate book figures.
- Hyundai campaign coverage is per-VIN and per-campaign; denied-claim reports exist in the field. Stage 0 output is 'check with dealer against VIN', never a coverage promise.
- Forum-sourced failure rankings (C6, C7, C12) are counts without denominators — directionally strong, statistically unweighted. R1 replaces them with NHTSA complaint counts.
- Identifix/SureTrack internals are paywalled; their capability is described from public materials and 2026 buyer's guides only.
- C10 is a single ruler query; a proper automotive coverage census (query set + per-host doc counts) is R4's first task.
Citation archive: 18 self-hosted text mirrors at /code/libcar_* (captured 2026-08-05; bot-walled/TLS-blocked sources via web.archive.org); sovereign re-ingest of originals queued behind the TLS record half (debts 1785966831/1785971025). Companion: the competitive matrix.