We ran the engine blind on a deal that went bankrupt.
A real HVAC acquisition, sold for $2.1M in April 2023. Our engine, given nothing but public records and a hard information cutoff the day before close, said do not close on current information. The buyer closed. Roughly 55% of the earnings evaporated almost immediately and the buyer later filed personal bankruptcy.
This was an internal backtest against a public record, not a client engagement. Nobody paid for it and nobody endorsed it. Everything below is scored against research files written before the outcome was known, and the one thing the engine missed is on this page at full size.
The deal, in five lines.
What the blind lanes found, and what they didn't.
Five independent research lanes ran against a 2023-04-01 information cutoff with every post-mortem source blocklisted, so none of them could read the answer key. Here is each of the four issues, and exactly how the research did against it.
Unlicensed plumbing work
Surfaced
Roughly 15% of revenue, illegal under a Colorado law passed in 2019. Two lanes hit it independently and one named the exact statute. Marked unresolved rather than asserted, because the state licence registry blocked automated access.
Off-books cash bonuses
Missed
Undisclosed cash compensation worth 15 to 25% of reported earnings. The engine did not find it. Nothing in the public record contains it, and no public-record process at any price finds it. This is a miss and it stays a miss.
Key technician departure
Surfaced
The lead technician left and opened a competing shop, taking roughly a third of the profit with him. Named as a risk with the exact questions to ask: tenure, non-competes, recent departures.
Worst season in 25 years
Surfaced and quantified
Not merely named as sector risk. The research modelled the actual SBA debt stack and priced what a bad summer does to it, before there was any way to know one was coming.
Debt coverage falls from a comfortable 2.29x to 1.05x. In plain terms, the business stops covering its own loan.
Computed blind, from public records, before the season happenedRaw recall. Three of the four issues that destroyed this acquisition, surfaced from public records alone.
Public-record recall. Of the issues knowable from public records at all, the research surfaced every one.
The verdict issued. On current information, with the open items named. The buyer closed anyway.
Federal registration codes in, a closing condition out.
The licensing exposure is the clearest example of what the engine actually does. Nobody hands it a red flag. It reads dull public records, notices a mismatch, finds the law that makes the mismatch expensive, and tells the buyer what to do before wiring funds.
Federal registration classified the business under NAICS 238220, "Plumbing, Heating, and Air-Conditioning Contractors," alongside PSC code J045, maintenance and repair of plumbing and heating equipment. A third-party directory described the firm as specialising in plumbing, heating and cooling systems.
Colorado has required state plumbing licensure since HB19-1086, signed April 2019. If this business performs plumbing work, a licence is a hard legal requirement, and a gap is material to the purchase. Treat it as a closing condition. Do not wire funds without a live pull from the state registry.
Ranked the number one high finding · left open rather than asserted · the state registry blocked automated access during the run
A backtest is only worth the constraints on it.
Anyone can run an engine on a deal whose ending is public and then describe the result generously. These are the constraints that make this number mean something, including the one that works against us.
A hard information cutoff
No source published after 2023-04-01, the day before the deal closed, was used for any finding. Every lane kept a discard log. One lane threw out four years of industry benchmarking literature and worked from a single dated 2022 report instead.
The answer key was blocklisted
The post-mortem interview, the publication that ran it, and anything published afterwards about this business were blocked before the run started. All five lanes independently reported zero blocklist hits. None of them ever saw how the story ended.
Zero seller documents
No P&L, no tax returns, no CIM, no add-back schedule. A real engagement receives all four. This run had public records only, which makes it a harder test than any paying client would ever set.
The part that works against us
The five research lanes were blind. The final write-up was not. Mid-run, the sealed outcome record re-entered the writer's context, so while every finding traces to a lane file written blind, the report's ranking and emphasis were not produced blind. That is why the claim on this page is that the blind lanes surfaced these issues, not that the report ranked them blind. The lane files are on disk and auditable, and the process is fixed for the next run.
What this does not prove.
The engine named the law, named the registry, and told the buyer to close the item before wiring funds. It did not verify the licence itself, because three separate state registries blocked automated access during the run. A person with an ordinary browser would likely have closed it in under an hour. That gap is a tooling limit, and it is fixed.
The undisclosed cash bonuses were the single largest hit to earnings, and no public-record process finds them. What the research did do was reconcile the claimed earnings against the documented headcount and conclude, on public data alone, that the number could not be underwritten. Different mechanism, same warning. It is not a detection of hidden cash and we will not describe it as one.
We did not predict the bankruptcy, and we do not claim to have. We said the deal could not be underwritten on the information available, and named what had to be closed first. That call was right. Everything after it belongs to the buyer.
Same engine, 48 hours, on your deal.
The Micro-DD is the product this backtest ran: a red-flag brief on a small-business acquisition, from public records plus whatever the seller has given you. It runs before you spend real money on accountants and lawyers, and it tells you whether the deal deserves them.