Geek+, the warehouse robotics maker listed in Hong Kong, published interim results at 14:28 UTC on 28 August. The announcement's second section is headed "Strong Improvement in Operating Performance" and lists as its third achievement a "remarkable improvement in profit." The accounts in the same document show the loss nearly quadrupling.

The reported numbers

Revenue of RMB1,283.9m, up 25.3%. Loss from operations of RMB175.7m against RMB82.2m. Loss for the period of RMB176.5m against RMB48.0m a year earlier — a widening of roughly 268%. The company attributes it to foreign exchange losses and increased embodied-intelligence spending. Overseas sales are above 75% of revenue at a 46.2% gross margin, which is the genuinely strong part of the release.

How the improvement is constructed

The "improvement" refers to a non-IFRS adjusted net loss, itself stated excluding foreign exchange gains: RMB60.6m against RMB89.3m, a 32.1% narrowing. The headline figure is a further step. Strip out the RMB44.5m spent on embodied-intelligence R&D and the adjusted loss becomes RMB16.1m — an 81.9% improvement. That is the number leading the announcement and the coverage.

What the common framing gets wrong

Two things, and the second is the interesting one. "Improvement in profit" describes a company with no profit; every figure in the sequence is a loss, and the statutory one grew. But the deeper problem is what the 81.9% removes. The same document opens by describing the company's philosophy as leading the transformation of embodied intelligence in warehousing — the AI programme is the stated strategy. Excluding its cost to reach a flattering number means the best result is obtained by subtracting the thing the company says it is for. Total R&D rose to RMB187.9m from RMB147.2m; the embodied-intelligence share is the growth.

The claim to date-check

The release states the company has ranked first in global autonomous mobile robot market share for seven consecutive years, footnoted to a market research firm. The footnote is to a published study; readers should check what period that study covers before treating the streak as current.