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

# Evaluating an enrichment waterfall

> How coverage, identity, acceptance rules, and source overlap determine whether an enrichment route is useful.

An enrichment route is useful when it establishes a fact the next process can
trust. Raw match rate measures a narrower thing: whether a source returned a
candidate. The gap between those measures is where many waterfall comparisons
become misleading.

The [waterfall overview](https://deepline.com/docs/features/waterfall-enrichment.md) explains the
execution mechanism. This page explains how to evaluate its business quality.

## Coverage depends on the population

A provider's result on one geography, company size, or identity input does not
establish its performance on another. A profile URL and an ambiguous name give
sources different evidence to work with. Mixing those inputs into one coverage
number can hide the segment that actually matters.

An evaluation therefore needs a defined population and acceptance rule. The
useful comparison is how each source performs on the same records, not which
vendor reports the largest aggregate match rate.

## Identity and field validation answer different questions

An address can pass an email check while belonging to the wrong person.
A correctly identified person can have an address that is stale or unsuitable.
Identity confidence and field validity are separate dimensions.

For example, two people may share a name at related companies. A source can
return a syntactically valid email without resolving that ambiguity. Promoting
it into a CRM contact collapses uncertainty that the workflow has not removed.

A usable result depends on the team's intended action. Researching an account
can tolerate candidates that a customer-facing outreach process should send
for review. The evaluation must preserve that distinction instead of treating
all nonempty fields as equally useful.

## Complementary coverage matters more than duplicated coverage

A second source adds value when it resolves records the first source misses
under the same acceptance rule. If both return the same accepted rows, the
second may provide redundancy but little incremental coverage.

The distinction affects provider order. Running a broad source first can be
reasonable, but a specialized source may belong earlier for a segment it
handles particularly well. One global ordering need not be optimal for every
input population.

The same reasoning applies to validation. More checks are useful only when
they resolve a meaningful uncertainty. Repeating equivalent evidence does not
necessarily make an identity decision stronger.

## Outcome categories preserve the explanation

An accepted result, an ordinary miss, a rejected candidate, a conflict, and a
provider error should remain distinguishable in the evaluation. They imply
different limits and different possible improvements.

A miss may justify another source. A conflict may require a precedence policy.
A failed request says little about coverage until the access or execution
problem is understood. Combining them into a single failure percentage loses
the reason the route did not produce an accepted result.

Deepline's maintained Plays expose their own result and attempt contracts.
They do not automatically know the evaluator's ground truth or define the
team's CRM acceptance policy. Those criteria must be supplied by the workflow
and review process.

## The evaluation itself has limits

A dated, reviewed sample estimates performance on that population. It does not
freeze future data quality or guarantee delivery to an email address. Changes
in sources, input mix, and acceptance policy can change the useful result rate.

[Enrichment cost](https://deepline.com/docs/guides/enrichment-cost-optimization.md) adds the economic view:
accepted coverage and the cost of reaching it need to be considered together.

## Apply this understanding

* [Run and evaluate a waterfall](https://deepline.com/docs/guides/waterfall-enrichment-claude-code.md).
* [Validate an incoming file](https://deepline.com/docs/guides/validate-incoming-csv-files.md).
* [Inspect the prebuilt contracts](https://deepline.com/docs/plays/catalog.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.