What Happens When Two "Correct" Numbers Disagree

The most frustrating disagreements in finance aren’t between right and wrong.

They’re between two correct numbers that don’t match.

Same issuer. Same reporting period. Different answer.

The Comparison Scenario (and why it keeps happening)

Two teams deliver numbers that are both defensible in isolation:

• One uses a data vendor’s normalized feed.
• The other parses primary filings and applies its own mapping.
• One runs a model on Monday; the other on Wednesday.
• One includes amendments; the other excludes them.

Each output is “correct” given its definition. But those definitions are different, and nobody realizes until a meeting asks, “Which number do we use?”

This is where data reconciliation best practices tend to break down:

• The comparison is ad hoc.
• The logic lives in spreadsheets or private notebooks.
• The answer changes next time because the context changed again.

The Internal Politics Layer

This is the moment where internal trust gets stressed.

PMs, Heads of Data, and engineering leads are forced into a build vs buy data pipelines debate they didn’t ask for:

• “We should trust the vendor.”
• “We should trust our internal pipeline.”
• “We should trust whoever can explain it fastest.”

The conversation shifts from data to politics. And the cost of reconciliation shows up every time another discrepancy appears.

Why This Matters for AI and Automation

If humans argue, agents fail.

When you start asking “can you audit AI outputs?” or explore explainable AI for finance, you quickly hit the same core issue:

If the system can’t explain a delta between two outputs, it isn’t safe to automate.

Automation demands a repeatable comparison layer, not just a delivery mechanism.

Reference Systems as Neutral Arbiters

The only reliable way to resolve competing “correct” numbers is to introduce a reference system.

A reference system:

• Runs in parallel with your existing process.
• Captures the exact inputs, timing, and methodology used.
• Produces a comparison artifact whenever outputs differ.
• Creates a durable audit trail you can reuse later.

In practical terms, it enables parallel run vendor comparison without forcing a full migration. You can keep your existing pipeline while a neutral reference system checks it.

What the Resolution Actually Looks Like

When two correct numbers disagree, a reference system gives you answers like:

• “Both are correct; one used an amended filing published 48 hours later.”
• “Both are correct; one used a different definition of ‘total debt.’”
• “Your vendor is lagging the primary source by 72 hours; the delta is timing.”

Instead of debating whose number wins, you can explain why they differ—and decide which definition is right for the decision at hand.

The Goal: Fewer Debates, More Confidence

The point isn’t to stop discrepancies. It’s to make them explainable.

When you have a reference system, disputes get resolved faster, decisions get defended more easily, and internal trust goes up.

If you want to see how this works in practice, here's how we structure verification and reference execution.

---

Related: What is Regulatory State? — Why state matters more than filings

How we make it defensible: Evidence Packs — The artifact behind every outcome