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

# Review results

> Interpret generated reconciliation output.

The results page is where a reconciliation earns its keep. It shows which records matched cleanly, which are missing from one side, and which pairs have field-level differences — organized so you can act on what you find, not just download a file.

## Review order

1. From **Ask Darpan**, search by run name or output file name.
2. Open **Open Result: ...**.
3. Confirm run metadata and input sources.
4. Review match and mismatch counts.
5. Open generated output.
6. Classify each difference as source data, schema, mapping, rule behavior, or expected variance.

## Common result categories

| Category          | Meaning                                                         |
| ----------------- | --------------------------------------------------------------- |
| Matched           | Darpan found corresponding records with no relevant difference. |
| Missing in source | A record appears in one source but not the other.               |
| Value mismatch    | A matched record has field-level differences.                   |
| Processing error  | The run could not parse, compare, or output part of the data.   |

## Where the numbers come from

Two things shape a difference count before you see it, and both exist to stop you chasing differences that were never real.

### Verification passes

Some differences are an artefact of how a source was read rather than a real gap. A bulk export can be indexed slightly behind the live system, so a record that exists is reported missing.

For sources that support it, Darpan re-checks reported-missing records with a direct point lookup and drops the ones that turn out to be present. The result names **which systems each pass rechecked**, so a count that moved is traceable rather than mysterious. Verification is capped per connector, so a pass never turns into an unbounded stream of lookups.

### Structural suppression

Some differences cannot exist by definition, and reporting them buries the ones that can. Where a record's own data proves no counterpart is possible, Darpan suppresses it and says so.

Returns reconciliation carries three such rules on the missing-in-OMS side, each a logical implication rather than an observed correlation:

| Rule                     | Why no counterpart can exist                                                                          |
| ------------------------ | ----------------------------------------------------------------------------------------------------- |
| Whole-order cancellation | The order was cancelled, so the refund is a cancellation, not a return.                               |
| Superseded return draft  | The draft was replaced; the surviving return is the real one.                                         |
| Cancelled-item refund    | The refunded lines were never shipped, so the system books an item cancellation rather than a return. |

<Note>
  Suppression only applies when the row's own data **proves** the rule. Where the evidence is inconclusive — a truncated view, a field the extract predates — the record stays reported. Suppressing on a "don't know" is the one direction that hides real differences, so Darpan never does it.
</Note>

Together these substantially change what a result reports. On one production run, the missing-in-OMS count fell from 31 to 9: 78 cancellation refunds, 6 superseded returns and 65 cancelled-item refunds suppressed as structurally impossible, plus 127 records confirmed present by point lookup.

### Reading source labels

Run results name the **system** a file came from — resolved to its proper label rather than showing an internal enum token or the endpoint that fetched it. When both sides of a run are the same system, the labels tell the two instances apart.

## Next actions

* Fix source data when the mismatch is real.
* Adjust schema or field selection when parsing is wrong.
* Adjust rules when business classification is wrong.
* Rerun after correction so the new output preserves evidence. See [Run reconciliation](/guides/run-reconciliation) to start the next run with updated inputs.