Mark Johnson, a former infrastructure engineer turned auditor, conducted a technical due diligence audit of Pulse AI, a firm seeking $18 million in Series B funding. Pulse AI claimed its automated platform achieved 89% production defect detection. During the audit, Johnson discovered that the company's evaluation set of 1,247 defect samples contained 44 exact matches against open-source defect databases and 54 samples that appeared to be manually handcrafted. These 98 samples constituted 7.9% of the total dataset, which had been processed using a tool identified as Apex-Lens-Cleaner v1.0.0.

Upon investigation, Johnson identified that the company's naming convention for its data pipeline, /pulse/ingestion/{env}/{source}, mirrored the architecture from his previous employer. He connected this pipeline design to a former colleague named Caleb and noted that the CTO, identified as Torres, utilized identical IT asset stickers and workspace arrangements to those at Johnson's prior company. When confronted, Torres acknowledged that the evaluation team had sourced samples from public databases and written others to reach a target benchmark of 95% before their Series C funding round.

The audit revealed that Pulse AI's benchmark was not generated from authentic production data as claimed, but rather curated from public sources and internal fabrication. Johnson reported the 7.9% defect overlap to the VC firm while withholding further evidence of the pipeline's origins, which he linked to a former colleague who had disappeared from the industry following layoffs. This engagement served as a covert audit for Johnson, who aimed to track the source of the transplanted system architecture while fulfilling his professional due diligence contract.

Mark Johnson, a former infrastructure engineer turned auditor, conducted a technical due diligence audit of Pulse AI, a firm seeking $18 million in Series B funding. Pulse AI claimed its automated platform achieved 89% production defect detection. During the audit, Johnson discovered that the company's evaluation set of 1,247 defect samples contained 44 exact matches against open-source defect databases and 54 samples that appeared to be manually handcrafted. These 98 samples constituted 7.9% of the total dataset, which had been processed using a tool identified as Apex-Lens-Cleaner v1.0.0.

Upon investigation, Johnson identified that the company's naming convention for its data pipeline, /pulse/ingestion/{env}/{source}, mirrored the architecture from his previous employer. He connected this pipeline design to a former colleague named Caleb and noted that the CTO, identified as Torres, utilized identical IT asset stickers and workspace arrangements to those at Johnson's prior company. When confronted, Torres acknowledged that the evaluation team had sourced samples from public databases and written others to reach a target benchmark of 95% before their Series C funding round.

The audit revealed that Pulse AI's benchmark was not generated from authentic production data as claimed, but rather curated from public sources and internal fabrication. Johnson reported the 7.9% defect overlap to the VC firm while withholding further evidence of the pipeline's origins, which he linked to a former colleague who had disappeared from the industry following layoffs. This engagement served as a covert audit for Johnson, who aimed to track the source of the transplanted system architecture while fulfilling his professional due diligence contract.