Skip to main content
Reality Oracle
OverviewCitizensReality FeedVerified ActorsIntel SubscribersMarket PredictionInvestment FundsHow It WorksGates
Products
Backtest ForensicsLiving MapDrift AlertsPath QueriesCustom AxesRisk Maps
Solutions
Compliance OfficersRisk Managers / CROPortfolio ManagersCTOs & Engineering
Enterprise
Enterprise OverviewReality Oracle · SuperappThe AppsArchitectureAudit ManifestTrust CenterPricing
Developers
DocumentationAPI ReferenceSDKsArchitectureAudit Manifest
Resources
BlogUse CasesCustomer StoriesPressChangelog
Company
AboutFoundersCareersContact
Trust
Trust Center
More
PricingJoin Waitlist
··12 min read

What an acquirer's diligence actually looks for

From the v2 audit — the questions a buyer asks that a manifest answers and a dashboard cannot.

What an acquirer's diligence actually looks for

An acquirer buying an AI company does not buy the model; they buy the right to defend the model's decisions after the founders leave. The diligence questions that matter are the ones a manifest answers and a dashboard cannot — provenance, reproducibility, boundaries, oversight. A company that cannot answer them is a company the acquirer cannot keep.

When an acquirer runs diligence on an AI company, the questions that kill the deal are rarely about the model's accuracy. They are about whether the model's decisions can be defended after the team that built it is gone. The founders leave; the model stays; the regulator still asks questions. An acquirer buying the company is buying the right to answer those questions, and the answers have to live in the substrate, not in the founders' heads. That is the diligence that matters, and it is the diligence a manifest answers and a dashboard cannot.

The four questions a dashboard cannot answer

A dashboard shows what happened. An acquirer needs to know four things a dashboard cannot show: provenance (which data, which model, which versions produced this output), reproducibility (can a third party rerun the computation and get the identical bytes), boundaries (what the system does not do — the scope that was stated), and oversight (who can override, when, and how the override is recorded). A dashboard has none of these; a per-output hash-chained manifest (BR-842) has all of them. The acquirer is buying the manifest, not the screen — because the manifest is what survives the founders.

The acquirer reads the substrate, not the slide. The manifest answers the four diligence questions a dashboard cannot — and the answers outlive the team that built them.
The acquirer reads the substrate, not the slide. The manifest answers the four diligence questions a dashboard cannot — and the answers outlive the team that built them.

Why reproducibility is the asset

An acquirer who buys a model without reproducibility is buying a black box they cannot defend. The model produces outputs; the regulator asks how; the acquirer cannot answer, because the team that knew is gone and the record is not in the substrate. A model with a per-output manifest is a model the acquirer can defend on day one — the manifest carries the model version, the input NodeVersions, the agent trace, and a replay anchor. The acquirer is buying the right to answer the regulator, which is the right that has value. Reproducibility is not a compliance feature; it is the asset.

An acquirer does not buy the model; they buy the right to defend the model's decisions after the founders leave. The right lives in the manifest, not the dashboard — because the manifest is what survives the team.

  • Provenance: which data, which model, which versions — the manifest fields an examiner reads.
  • Reproducibility: a third party reruns the computation and gets identical bytes — the replay anchor.
  • Boundaries: what the system does not do — the stated scope that makes the system credible.
  • Oversight: who overrode, when, how it was recorded — AxisOverride, sealed in the chain.

What the manifest refuses, and why the acquirer wants the refusal

The manifest refuses to certify truth (Certified ≠ True), and the acquirer wants the refusal — because a manifest that refused to overclaim is a manifest the acquirer can stand behind without fear of the overclaim surfacing in their own audit. A target company whose manifest says "reproducible, not correct" is a target with a clean boundary. A target whose manifest says "certified correct" is a target with a liability the acquirer will inherit. The diligence question that matters most is the one the target should refuse to answer: "do you certify this is correct?" The refusal is the diligence pass.

The v2 audit, and what it found

The v2 acquisition audit (at docs/audit/) walked the questions a buyer's diligence team actually asks and mapped them to the manifest fields that answer them. The finding: the questions a manifest answers are the questions a dashboard cannot, and the questions a dashboard answers are the ones an acquirer does not ask. The acquirer is not buying a screen; they are buying the right to defend the decisions — and the manifest is the right. We are pre-revenue, building the POC; the manifest is built, the calibration is gated (Gate 1b), and the diligence posture is the posture of a company whose claims outlive its founders.

The acquirer is buying the right to answer the regulator after the team is gone. That right lives in the manifest, not the dashboard. A target that cannot hand over the manifest is a target the acquirer cannot keep — and a target that overclaims in the manifest is a target the acquirer should not.

— v2 audit note
§Keep reading