Robotics knowledge already exists — as prose, not as structure. Entropy Box compiles it once into a typed graph that programs can consume.
This page is the static mirror. Complex-query retrieval, one-click plan assembly and the live dashboard run at xiangshang.ngrok.app — no registration, no install.
Regenerated from the artifact on disk by measure.py — no GPU, no network, no service. Snapshot 2026-08-30. The paper's public snapshot (28 August 2026) reports 2,511 topics, 37,673 capabilities and 11,397 assets; the artifact has since grown slightly.
It writes the detector you ask for. It cannot see that one task hides a dozen steps and the dependencies between them.
A craftsman who has mastered every part but never held the drawing. We build the drawing.
Writes a detector. Cannot deliver “pick up the red cup” end to end.
Perception → localization → planning → control. Which feeds which is what gets lost.
A new task means reading dozens of papers to compare routes by trial and error.
Most hours go to wiring modules together, not to the algorithm work.
RAG · GraphRAG · agentic search
Entropy Box
Each compiled unit is consumed 1.57× on average — 21,380 re-derivations avoided — and the curve has not flattened.
Amortization factor α against topics compiled.
Four stages under a persistent control plane. Every transition is mediated by a durable artifact and a mechanical gate.
On 2,362 adjudicated pairs, once the gate has flagged a pair the similarity score carries too little signal to decide the merge (embedding AUC 0.509, CI [0.453, 0.564]). Raising the threshold does not help: precision stays flat near 5–6% while recall of true duplicates collapses from 100% to 4%. Merging automatically above a threshold would be wrong more than nine times in ten — which is why the merge decision is deferred to an LLM adjudicator that reads both records and writes the rationale into the ledger. The similarity score flags; the model judges.
| Path | pairs | dup. | prec. | AUC |
|---|---|---|---|---|
| Embedding | 1,867 | 103 | 0.055 | 0.509 |
| Lexical | 495 | 8 | 0.016 | 0.689 |
Conditioned on having been flagged, so gate recall is unmeasurable here; the lexical row rests on eight positives. Full note →
Tightening the gate throws duplicates away without buying precision.
Every topic cites the documents it was assembled from, and because those are files the citation is mechanically checkable: retrieval documents resolve at 99.56%, chain documents at 82.0%. Cheap to compute, hard to game — and a system that cannot report it does not have one.
2,391 fields across 20.5% of topics are marked unverified rather than filled in. Zero markers would not mean a better artifact — it would mean a pipeline that fills gaps instead of declaring them.
This is the result we care about. On the EntropyBench Track-P plan-synthesis benchmark (24 tasks across 8 robotics domains), every ungrounded baseline — LLM direct, BM25 RAG, hybrid RAG — emits plans whose asset and interface claims are 100% unsupported. Entropy Box cuts that to 5.2% (−94.8pp vs hybrid RAG, CI [−97.4, −92.1]). The same grounding transfers to downstream simulation code generation: pass-1 executable plans rise from 0.58 (Vanilla RAG) to 0.92.
| System | unsup.↓ | viol.↓ | cov.↑ |
|---|---|---|---|
| LLM direct | 100% | 100% | 0% |
| BM25 RAG | 100% | 100% | 0% |
| Hybrid RAG | 100% | 100% | 0% |
| Entropy Box | 5.2% | 66.7% | 35.4% |
Unsupported claims · constraint violations · constraint coverage. Paired diff vs hybrid RAG, 95% CI in the paper §9. Reproduce from evaluation/.
| System | import | guards | pass-1 |
|---|---|---|---|
| Zero-shot | 0.55 | 0.16 | 0.33 |
| Web search | 0.69 | 0.42 | 0.50 |
| Vanilla RAG | 0.75 | 0.50 | 0.58 |
| Entropy Box | 0.98 | 0.88 | 0.92 |
Import validity · constraint guards · pass-1 executable plans. Grounding in the compiled artifact, not raw retrieval.
51 seconds, unedited. A goal is decomposed, matched across topics, and assembled into a plan whose nodes carry attributed open-source assets. The paper also reports two independent development cases that query the same public interface to build simulation applications.
Husky + Panda compose a farm-harvest simulation using the compiled substrate.
A humanoid G1 executes orchard-packaging tasks in MuJoCo, grounding steps in the same public knowledge base.
Cropped only, to remove OS chrome. Not sped up.
Search the ordering graph, filter by hub or bridge role, follow edges.
The curated taxonomy that drives compilation.
Nine compiled topics reproduced completely — every chain, description, repository and ledger.
Four stages, the admission plane, and the original research documents the compiler read.
Things we measured and things we got wrong.
What this is not, what we withhold, and what we have not established.
Data CC BY 4.0 · code MIT.
Three assertion strengths, plus negative assertions declaring what must not be retrieved.
Near-duplicate pairs with score, metric, adjudicator and written rationale.
Canonical English and Chinese names, with abstraction level.
Repositories, models, datasets, simulators and platforms, with upstream sources.
The three-level compilation backbone, bilingual.
Every number on this page, as structured output from
measure.py.
Four read-only endpoints, bilingual (中文 / English). Any client that speaks OpenAPI, MCP or REST can integrate — Claude, ChatGPT, Cursor, Trae, WorkBuddy.
| Endpoint | Purpose |
|---|---|
/api/consult | Solution consult |
/api/lookup | Entity lookup (CAP_ / AST_) |
/api/evidence/search | Evidence RAG |
/api/search | Hybrid search |
curl -X POST "https://xiangshang.ngrok.app/api/evidence/search" \
-H "Content-Type: application/json" \
-d '{"query": "robot obstacle avoidance algorithms",
"top_k": 5, "mode": "hybrid", "rerank": true}'
# OpenAPI schema: /openapi.json
# MCP server: /mcp/ontology_mcp_server.py
The benchmarks exist now; the honest gaps are the ones below. We would rather say so than let you find them.
/integrate
planner drops graph validity to 62.5% while eliminating unsupported claims.