Benchmarks

Web Context Bench (Python)

Local-page benchmark comparing six browser representations on token use, factual reading and target-element selection, with deterministic answer scoring.

AI-assisted editorial review against a primary source. This is a catalog annotation, not an independent reproduction or endorsement.

Original source ↗Read as Markdown Record JSON

What is this resource?

Resource type: Benchmarks
Environment: web
Topics: evaluation, grounding, efficiency, screen-understanding
Record publication date: Not recorded
Evidence retrieved: 2026-10-20T11:51:30.874397Z
Resource ID: github:charlypoly/web-context-bench-python

What licenses are recorded?

Code, training data, weights, and documents can have different terms. A blank declaration means unknown, not unrestricted use.

Code: MIT (Complete MIT license text in the repository root at the inspected commit; repository code only.)
Data: Not recorded
Document: Not recorded
Weights: Not recorded

License review notes

Code: The inspected root LICENSE contains complete MIT terms for repository software. The methodology explicitly leaves redistribution permission for captured third-party practice-site pages unresolved; code MIT is not a blanket snapshot/data license. Source 1

Evaluation protocol

Focus: Browser-representation token use, Factual READ accuracy, ACT target-element selection
Setting: offline
Observations: Top-document HTML, Markdown converted from HTML, Playwright accessibility snapshot, Browser Use indexed DOM, Stagehand hybrid snapshot, Viewport screenshot
Actions / outputs: READ text answer, ACT selector, text/label, accessible role/name, DOM index, Stagehand reference or screenshot point, according to the representation
Evaluation: READ uses normalized exact match against an answer or allowed alias. ACT resolves predicted identifiers to ground-truth elements, or checks screenshot points inside the target box. Representation-token accounting subtracts fixed-prompt tokens; screenshots use billed input usage minus the fixed prompt. These are answer/target correctness and token metrics, not completed-task success.

Scale and counting units

Not recorded

Fixed local-page observations; the reported run uses claude-sonnet-5 and a shared prompt, without scrolling or an agent loop. Browser Use disables its viewport cutoff and sends untruncated text; Stagehand uses a separate matched browser load. Short-string representability labels are approximate. Offline does not mean API-free. Third-party snapshot redistribution remains unresolved; results were not reproduced.

Compare source-reviewed profiles

Explore related topics

Evaluation GUI grounding Efficiency Screen understanding

Scope and limitations

This is a single-snapshot representation evaluation on local fixtures and captured pages, not a live multi-step agent benchmark. Its shared-prompt, one-model setup differs from complete Browser Use or Stagehand agents. The methodology notes unresolved redistribution licensing for third-party practice-site snapshots; MIT is recorded for repository code only. Selected scoring code was read, but no models, browser tasks or claimed results were run or reproduced.

Primary evidence and verification

Source retrieval is separate from experiment reproduction. Use the original publication or publisher page when citing claims about the resource.

editorial primary review
Retrieved 2026-10-20T11:51:30.874397Z. Supports: name, kind, summary, topics, scope, notes, environments, details.benchmark_protocol.focus, details.benchmark_protocol.setting, details.benchmark_protocol.observations, details.benchmark_protocol.actions, details.benchmark_protocol.notes.
Content fingerprint
14775f6e3daba69d4684380fd33bf717bdc3093b65be796eb863bbd27d0c0bab
github repository identity
Retrieved 2026-10-20T11:05:35.063004+00:00. Supports: id, url.
Content fingerprint
3e22e1cbfa47bcf98f6f70a5f6f162a31a509a9f2091a84679bb5231977e29d9
github repository identity
Retrieved 2026-10-20T11:50:54.545491Z. Supports: id, url, notes.
Content fingerprint
fd30e94e78e7167a49579922259fde1c7d7f8259e9f00afba15faf3a71966a94
editorial primary review
Retrieved 2026-10-20T11:52:30.131541Z. Supports: summary, environments, topics, scope, notes, licenses.data, details.benchmark_protocol.focus, details.benchmark_protocol.setting, details.benchmark_protocol.observations, details.benchmark_protocol.actions, details.benchmark_protocol.evaluation, details.benchmark_protocol.notes.
Content fingerprint
1c3c0cbf9f9d9f7cd9e7a2f027ecb217fd5c16fa157916281a15bb7f403f6f1c
editorial primary review
Retrieved 2026-10-20T11:53:28.567861Z. Supports: summary, topics, scope, notes, details.benchmark_protocol.evaluation, details.benchmark_protocol.notes.
Content fingerprint
8cada5d71b5a61e7aff2e1ac2b951c266eb2db7c082c06c5a13aa127d42b3fc5
editorial primary review
Retrieved 2026-10-20T11:51:41.499735Z. Supports: licenses.code.
Content fingerprint
9b64ef672f5091b6074d02502f6d3c161d94308a3d9d142d7437dcddf611bbc4

Suggest a correction

Open a source-backed correction. Include this record ID and the original source. Read our editorial policy and coverage boundaries.