Vendor-neutral protocol · version 1.2
Test an HTML artifact handoff with evidence—not feature claims.
Ten repeatable tests cover publishing, review, recovery, automation, access, portability, setup time, and price transparency. This release publishes the protocol, frozen fixture, blank scorecard, and machine-validatable result format; it does not claim vendor results.
Methodology
One artifact, three roles, one evidence bar.
- 1. Use the frozen test artifact. Download the versioned fixture and verify its SHA-256 checksum before every run. It includes headings, a table, a link, an image alternative, a stable update region, sharing metadata, and one interaction.
- 2. Start clean. Record the product, plan, date, browser, region, and whether an account or payment method is required.
- 3. Test three roles. Repeat as publisher, signed-out reviewer, and API, CLI, or MCP client where supported.
- 4. Save primary evidence. Capture the tested URL, elapsed time, response or screenshot, documentation URL, and limitation for every score.
- 5. Recheck volatile facts. Verify prices and plan limits on the access date; never carry them forward without a fresh source.
Frozen fixture
Run every product against the exact same file.
Fixture HASB-2026-F1 is dependency-free and 5,047 bytes. It provides a stable paragraph for comments and agent updates, required semantic elements, explicit sharing metadata, an accessible image alternative, and a deterministic button interaction.
SHA-256 fe6f01956e9ac8e8424872bb40d862e5852d9ff3d28d359066045aed45dc3fb8The protocol
Ten tests. Twenty possible points.
Durable URL
Publish a revision and confirm the original HTTPS URL still resolves to the current artifact.
Rendered review
Open the artifact signed out and complete its primary interaction without downloading a file.
Contextual feedback
Attach feedback to a section or passage, reload, and confirm that the context is preserved.
Version recovery
Create two revisions, identify both, and restore or export the earlier state.
Agent read and update
Use a documented API, CLI, or MCP interface to read state and update one defined region.
Access control
Verify the declared public, unlisted, or restricted state in a separate signed-out session.
Sharing metadata
Record the canonical URL, robots directive, title, description, and preview image exposed by the page.
Portability
Export the artifact and verify that its meaningful content remains usable outside the product.
Setup time
Measure from a clean start until a second person can open the first reviewable URL.
Price transparency
Record the tested plan, price source URL, access date, and any limit encountered during the run.
Scoring
Conservative by default.
2
Reproducible pass
The test succeeds through a documented, generally available workflow and the evidence is saved.
1
Qualified pass
The outcome is possible, but requires a documented workaround, higher plan, or material side channel.
0
Not verified
The capability is absent, the test fails, or the available evidence is insufficient.
Do not rank products with missing evidence. Publish the individual test records, plan constraints, and access dates beside any aggregate.
Portable results
Publish evidence in a format others can check.
The draft 2020-12 JSON Schema requires all ten criterion IDs exactly once, binds each score to its status, requires evidence for a pass, preserves the tested plan and environment, and makes the tester relationship explicit. The downloadable example is intentionally blank and contains no vendor result.
Citation
Reuse the protocol with attribution.
Naidu, Raunaq. “Open HTML Artifact Sharing Benchmark Protocol 2026.” HTML Docs, 31 Aug. 2026, https://www.html-docs.com/guides/html-artifact-sharing-benchmark.Need the deeper collaboration checks?
Use the 15-point checklist when review safety and agent continuity need a fuller audit.