Sanwa Kotsu — Japan
Sanwa Kotsu holds the two largest figures in our entire dataset — +75.7% and +68.4% on two Toyota Crown Comfort taxis. They do not survive a look at the baseline. Both cars are recorded as starting at 3.54 and 3.21, roughly 44% below every other Crown Comfort in the same dataset, while their after figures land in the perfectly ordinary range. The two usable measurements at this operator are the SP Deluxe pair: +2.5% and +8.6%.
The best-looking number we own is the one we spent the most time trying to break. It broke.
| Operator | Sanwa Kotsu (三和交通), Japan — taxi operator |
|---|---|
| Vehicles | 4 LPG taxis — 2 SP Deluxe, 2 Toyota Crown Comfort |
| Method | 基準燃費と比較 — comparison against a reference consumption figure |
| SP Deluxe result | +2.5% and +8.6% — plausible, and the figures we use |
| Crown Comfort result | +68.4% and +75.7% — excluded |
| Reason for exclusion | Recorded baselines of 3.54 and 3.21 against 5.99 for comparable vehicles |
| Median if all four are kept | +38.5% — a figure we do not quote |
Every measurement, not just the best one
| Vehicle | Before | After | Change |
|---|---|---|---|
| SP Deluxe (1) | 5.67 | 5.81 | +2.5% |
| SP Deluxe (2) | 5.22 | 5.67 | +8.6% |
| Crown Comfort (1) | 3.54 | 5.96 | +68.4% |
| Crown Comfort (2) | 3.21 | 5.64 | +75.7% |
| Median across all 4 measurements | +38.5% |
Method before result
All four taxis were compared against a reference consumption figure. Two SP Deluxe taxis returned +2.5% and +8.6% — ordinary results, in line with the other taxi operators in the dataset.
The two Toyota Crown Comfort taxis returned +68.4% and +75.7%, the largest figures anywhere in our records. Rather than publish them, we checked the inputs. Their after figures are 5.96 and 5.64, which is entirely normal for this vehicle. Their baselines are 3.54 and 3.21 — against 5.61, 6.23, 6.13 for every other Crown Comfort recorded in the same dataset, an average of 5.99.
In other words the improvement is not in the result, it is in the starting point: these two cars are recorded as beginning roughly 44% worse than any comparable vehicle and ending up exactly where comparable vehicles end up. The most likely explanations are a faulty or estimated baseline, a vehicle with an unrepaired fault at the time of the first reading, or a transcription error. None of them is a product effect we can claim.
Every percentage on this page was recalculated by us from the recorded before-and-after figures rather than taken from the summary column of the source; all of them reproduce the recorded value. That is a check on transcription, not on the measurement itself — it confirms the arithmetic is right, not that the reading was.
The limits of this trial
We publish these next to the result rather than in a footnote. If a limitation would change how you read the number, you should read it at the same time as the number.
Known limitations
- The two excluded measurements are excluded on our own analysis, not because the source flagged them. If the baselines are real — a genuinely faulty pair of taxis brought back to normal — then the exclusion is over-cautious. We accept that risk in this direction rather than the other.
- The two remaining measurements are two vehicles. +2.5% and +8.6% on a pair of taxis is a weak result, and it is the honest one.
- No control group, no recorded measurement period, no route or shift information.
- The unit behind the LPG figures is unresolved. The dataset labels these columns km/L, but its own CO₂ calculation uses an emission factor of 3.00 kg per kilogram of LPG, which only makes sense if the figures are kilometres per kilogram. The percentage change is unaffected either way — it is a ratio — but we flag it rather than quietly picking one.
- This is a dataset entry, not a trial report. What we hold is the before-and-after fuel-efficiency figure per vehicle as recorded in the master test dataset — not the operator’s raw fuel logs, not the odometer readings behind them, and not a written test protocol.
- There is no control group. No comparable untreated vehicle was tracked over the same period, so seasonal, route and traffic effects cannot be separated from the treatment.
- The measurement dates and the length of each measurement period are not recorded in these entries. We therefore cannot say how far each vehicle ran between the two readings, or which months the comparison covers.
- The figures were recorded by the operator or by the local distributor, not by an independent party, and we cannot re-derive them from primary records.
Reading this evidence in context
This page exists because of a rule we apply to ourselves: a number that looks too good gets checked before it gets published, not after a customer challenges it.
A +75.7% fuel saving is not credible on any vehicle, and quoting it — even accurately, even with the source attached — would undermine every honest figure elsewhere on this site.
It also shows what our data cleaning actually consists of. The same check applied to the whole dataset is what produced the documented-versus-undocumented split on the field-data page.
Questions operators ask about this trial
Could the +75.7% be real?
Why not simply delete the two measurements?
What do you actually claim for this operator?
Provenance
- Source
- Eco-Spray / NanoEFX master test dataset, compiled 22 March 2020 — operator-named section
- Recalculation
- All percentages recomputed from the recorded before-and-after figures
- Raw records
- Held by the operator or the local distributor — not in our possession
- Operator entry
- 三和交通 — Sanwa Kotsu
- Method as recorded
- 基準燃費と比較 (against reference consumption)
- Exclusion
- 2 of 4 measurements excluded by ECO EFX Solutions on baseline-plausibility grounds; both remain visible in the table above and in /evidence/field-data/
- Name release
- Operator naming approved by ECO EFX Solutions, 15 August 2026
- Last checked
- 15 August 2026
The only baseline that matters is yours
Every trial on this site was run on someone else’s vehicles, duty cycle and baseline. A NanoEFX pilot measures yours — and we will help you design it with a control group so the result stands up.