Repository navigation
Accuracy pass across every feature - #28
Merged
Merged
Conversation
Data: correct implementation languages (Vite/Vue -> TypeScript, Nmap -> C++, TensorFlow -> C++), drop hard "depends_on" edges for optional runtimes (CUDA, containerd), fix edges that read as wrong claims (Cloudflare "supports" DDoS attacks, Falco/Trivy/Snort as threat modeling, k6 as integration testing, protobuf as part of gRPC, Sigma.js on WebGPU), update stale facts (Coq -> Rocq, Remix -> React Router v7, OpenGL frozen at 4.6, Cuckoo deprecated, ESP32 is hardware) and two Wikipedia titles that pointed at other subjects. Guided path captions and famous-app notes corrected (Kafka open-sourced 2011, Figma on WebGPU, OpenAI's extra clouds, Wikipedia's OpenSearch move). Validator: new semantic rules (directional edges both ways, alternative_to across node types or alongside built_on/depends_on, non-offensive tools that "support" an attack). Scanner: Gradle version catalogs and plugins blocks are read (Android, Kotlin Multiplatform, Compose, Firebase now detected), setup.py is parsed with ast, huggingface-hub no longer maps to Transformers, @types/jest maps to Jest. Three new corpus fixtures; ground truth corrected where the scanner was right. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EMFEEMXAuR2wLnpNEoZQFz
Revert the TensorFlow, Sigma.js and k6 edge changes (each was defensible as written) and retype the PyTorch-CUDA edge rather than retarget it, so field affinities stay the same and no whole field moves on the ring. Falco now links to EDR inside the security field. layout.json regenerated: 238 nodes shift within their own clusters, no field changes place. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EMFEEMXAuR2wLnpNEoZQFz
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EMFEEMXAuR2wLnpNEoZQFz
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Requested by Rush · project thread
Summary
This PR only fixes accuracy. Nothing in the UI changes. I audited the data and output behind every feature, fixed what was wrong, and added checks so the same kinds of mistakes fail CI next time. All fixes live in
scripts/expand/accuracy_pass.py, which is idempotent and comments each fix with why it was wrong.Tech graph (seed data)
integrates_with. CUDA is optional, and Kubernetes accepts any CRI runtime.related_to.integrates_withprotobuf.related_to.mature(frozen at 4.6 since 2017).deprecated(unmaintained since 2019).platform(hardware), not a developer tool. Sass is a preprocessor, not a standard.Guided paths
Famous apps
not_in_atlas). Please double-check this one. It comes from memory and I could not verify it from here.Repo X-Ray scanner
plugins {}blocks are now read. Before, Android, Kotlin Multiplatform, Compose and Firebase were missed whenever they were applied through convention plugins (nowinandroid, kotlinconf-app). This adds a newgradle-pluginecosystem inpackage_map.json.setup.pyis now parsed withastand never executed. Before, it was selected but ignored, so huggingface/transformers missed torch and numpy.huggingface-hub(the Hub client) mapped to Transformers. It is now never-map.@types/jestnow maps to Jest.kotlin-android-catalog,py-setup-py-lib, and the negative caseneg-gradle-plugin-lookalikes.huggingface-hubadded to the PyPI name-collision fixture.Stack recommender and role paths
deprecatedorhistoricaltechnology, and roles never teach one.Validation added
scripts/validate_dataset.pyhas a newvalidate_semanticspass. It fails on:built_on,written_in,depends_on,part_of_ecosystem) that runs both waysalternative_tobetween different node types, or on the same pair asbuilt_on,depends_onorwritten_insupports,implements,enablesor isused_foran attack concept (this would have caught the Cloudflare edge)backend/tests/test_accuracy_pass.py(17 tests): regression guards for the facts above, the semantic rules, the Gradle and setup.py parsers, and a check thataccuracy_pass.pyhas nothing left to apply.Layout
The edge and node counts are unchanged (1,194 / 3,517), so edge ids stay stable. I kept cross-field affinities the same, so no field moves on the ring.
layout.jsonwas regenerated: 238 nodes move a little inside their own cluster.Not changed (worth a look later)
cross-platform-uiis Declarative UI andmessage-brokeris Publish–Subscribe. Renaming slugs would break share links and learning maps, so I left them.frontend/public/example-scans/were not regenerated. That needs repo access through the GitHub API or git, and the Gradle and setup.py improvements will show up the next time they are rebuilt.Test plan
pytest(backend): 272 passedvitest: 160 passed,tsc --noEmitcleanscripts/validate_dataset.py: OKhttps://claude.ai/code/session_01EMFEEMXAuR2wLnpNEoZQFz
Generated by Claude Code