> ## Documentation Index
> Fetch the complete documentation index at: https://dev.jolts.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Understand your results

> Know what happened, why it might matter, and what the data cannot tell you.

A result is a reason to investigate a company, not a promise that it will buy.

## Read a result in this order

| Field                                 | Question it answers                                                   |
| ------------------------------------- | --------------------------------------------------------------------- |
| `observed_fact`                       | What activity does the source describe?                               |
| `evidence[].source_uri` and `excerpt` | Where can I check the supporting information?                         |
| `occurred_at`                         | When did the reported activity happen?                                |
| `observed_at`                         | When was it observed?                                                 |
| `why_it_may_matter`                   | Why might this activity be relevant to the search?                    |
| `demand_class`                        | Is the need explicit, indirect, or structural?                        |
| `confidence`                          | How confident is the observation? This is not a purchase probability. |

## Facts are not inferred demand

**Illustrative example:** a company announces a new warehouse. That announcement is an observed fact.
It might need warehouse equipment, but the announcement does not prove an open budget or a purchase.

| Demand class       | Meaning                                                                |
| ------------------ | ---------------------------------------------------------------------- |
| `explicit_request` | Evidence of an expressed request or need. Still not a guaranteed sale. |
| `proxy`            | An activity that may suggest a need, such as a facility expansion.     |
| `structural`       | A possible need inferred from the company's circumstances.             |

Company results contain up to three supporting observations. Ranking is not a calibrated fit score.
Several evidence references can report the same original source; reference count is not independent
verification. `differing_source_claims` flags differing claims among the returned observations for
you to inspect. It does not choose which claim is true.

## Coverage has limits

Company search currently returns `coverage_status: "partial"`. Read `coverage.stop_reason` and
`coverage.limitations` alongside the results. An empty list does not prove no matching company exists.
The implemented observation kinds are procurement requests and facility expansions, in Canada and
the United States. A broad natural-language question does not expand that coverage.

`next_cursor` continues the existing search snapshot. It does not start a new round of research.
At most 100 companies enter the shortlist. A new search may produce a different list.

## Results can change when you replay them

Detail reads and operation replays recheck current evidence. Expired or withdrawn evidence may no
longer be returned. `withheld_units` reports previously delivered companies that no longer have
eligible support; the historical usage settlement does not change.

Use [company-list workflows](/workflows) to filter and investigate results, or the API Reference tab
for the complete response schema.