# Shadow Interface 15-minute test

Version: 0.1

Status: Working protocol, not a validated benchmark

Framework: Fetch -> Render -> Resolve -> Retrieve -> Verify -> Refer

This protocol helps a team identify one reproducible gap between a human-facing page and a representation available to a specified machine consumer. It does not produce a universal quality score or establish the Shadow Funnel as a scientific model.

## Define the test

- Target page:
- Target machine consumer:
- Supported question the page should answer:
- Collection date and time:
- Rendering conditions or user agent:
- Reviewer:

Name the consumer before testing. A crawler, answer engine, browser agent, accessibility API, and internal retrieval system may receive different artifacts and should not be treated as interchangeable.

## Minute 1 to 3: Fetch

**Question:** Can the consumer access a stable response?

**Test:** Inspect the unauthenticated HTTP response, status, redirect chain, canonical path, and crawler directives.

- Artifact inspected:
- Observation:
- Evidence:
- Result: Pass / Gap / Not observable

## Minute 4 to 6: Render

**Question:** Does the meaningful content exist in the representation this consumer receives?

**Test:** Compare initial HTML, hydrated DOM, accessibility semantics, and structured-data blocks that are available under the defined rendering conditions.

- Artifact inspected:
- Observation:
- Evidence:
- Result: Pass / Gap / Not observable

## Minute 7 to 9: Resolve

**Question:** Can the consumer identify entities and connect their relationships?

**Test:** Trace internal identifiers and verify that each Person, Organization, Product, or Article resolves consistently within the collected representations.

- Artifact inspected:
- Observation:
- Evidence:
- Result: Pass / Gap / Not observable

## Minute 10 to 11: Retrieve

**Question:** Can the answer-bearing evidence survive outside the visual layout?

**Test:** Extract the smallest passage that answers the supported question. Read it without the surrounding layout and check whether its subject, context, and support remain recoverable.

- Artifact inspected:
- Observation:
- Evidence:
- Result: Pass / Gap / Not observable

## Minute 12 to 13: Verify

**Question:** Do visible claims and machine-readable claims support each other?

**Test:** Compare each load-bearing structured claim with evidence a person can inspect. Record missing, changed, and unsupported facts separately.

- Artifact inspected:
- Observation:
- Evidence:
- Result: Pass / Gap / Not observable

## Minute 14 to 15: Refer

**Question:** Can the consumer cite, attribute, route, or hand off correctly?

**Test:** Verify the citation target, entity attribution, destination state, canonical URL, and human handoff path.

- Artifact inspected:
- Observation:
- Evidence:
- Result: Pass / Gap / Not observable

## Record the output

Do not collapse the observations into one universal score. Record each gap independently.

| Field | Value |
|---|---|
| Stage | |
| Affected representation | |
| Expected supported fact, relationship, or action | |
| Observed output | |
| Reproduction steps | |
| Evidence URL, fixture, or hash | |
| Owner | |
| Proposed gate | |

A useful finding is reproducible, tied to a collected artifact, scoped to a named consumer, and assignable to an owner.

## Baseline comparison for research use

A future benchmark should run the same checks in two conditions:

1. As an unordered checklist.
2. As the proposed six-stage sequence.

The ordered Shadow Funnel adds diagnostic value only if it explains or predicts failures beyond the unordered baseline. If it does not, the sequence and its new terminology should be rejected.

Essay and methodology: https://smlee.dev/en/writing/shadow-interface