Skip to main content

Where answers come from

Every answer SixDegree gives can be traced back to the systems it came from. This page explains the proof you see in the product and how to use it.

The sources line​

Under every answer that came from your data, you'll see a line like:

✓ Checked against Salesforce HubSpot · 261 records · data as of Jul 16

This line is built from the query's recorded execution, not from the AI's text, so it cannot be invented. It tells you three things: which systems the answer drew on, how many records were consulted, and how fresh that data is.

The answer itself ends with a matching Sources: line. The AI is only allowed to name systems from the recorded list. It cannot cite a system that was not actually consulted.

When an answer is not backed by data​

If an answer contains figures but no data source was consulted, SixDegree says so:

⚠ Not checked against your data. Treat any figures here as unverified.

This warning is added by the platform, not the AI, and the AI cannot suppress it. Every answer is either visibly backed by your data or visibly not. There is no silent middle.

Verifying a single value​

Anywhere you see an entity's fields, hover a value and click verify. You'll see:

  • Which system the value came from, and the exact source record
  • When that data was last true (its as-of date)
  • What other systems said, when your tools disagreed. For example: industry is "Retail" from Salesforce, but HubSpot called it "Consumer Goods"
  • How the entity was assembled: how many source records were combined into it

If a value was brought in from a related record (a lookup), or calculated by SixDegree rather than pulled from a system, the verification says that too. If a lookup could not find its target, the field is left empty and the reason is recorded. SixDegree never shows a stale value whose source is gone.

Asking the AI directly​

You can also just ask: "where did that number come from?" The AI answers using the same recorded lineage that powers the verify button, so the conversational answer and the click-through answer always agree.

Answer quality reports​

Under Build → Answer quality, you can see the results of verification runs: a set of standard questions asked against your own environment, checked for correct facts, honest citations, and grounded execution. Each run produces a scorecard, so answer quality is measured on your data, not claimed in the abstract.

Why this matters​

Merged data is only trustworthy if you can see the seams. SixDegree records, at merge time, which system won each field and what the losing values were, and keeps each record's own timeline. The proof you see in the product is that record, surfaced. Nothing is reconstructed after the fact.