# Semantic Fractal Topology — Technical Evaluation Brief

Version: 2.0

## Current disposition

Semantic Fractal Topology is an implemented, browser-local deterministic world-state and solver runtime. It validates typed spatial state, constructs evidence contracts, routes supported operations to named deterministic solvers, records explicit missing evidence, and preserves immutable revision history.

The first retrieval-efficiency greenlight did not pass.

## B18 R1 result

The frozen seven-case comparison covered six operation classes and one missing-evidence control under full-scene, lexical top-k, one-hop graph, and SFT operational context construction.

SFT achieved:

- required entity, relation, and metric-fact recall: 1.0;
- false authoritative executions: 0;
- deterministic solver parity with full scene: 1.0;
- exact source/revision traceability: 1.0.

The predeclared efficiency target required median serialized context at no more than 60% of full-scene bytes. SFT measured 78.5827%. The result is `FAILED`, and the frozen model holdout was not run.

The lexical and graph conditions were transparent local baselines, not production Vector RAG or GraphRAG systems. The result does not establish model-answer, context-efficiency, latency, scaling, interoperability, robotics, or production-digital-twin advantage.

## Implemented surfaces

- typed semantic scene graph and metric frames;
- content-addressed scenes and immutable revisions;
- deterministic typed-command compilation;
- bounded fit, constraints, reachability, assembly, service-access, clearance, collision-candidate, and move-proposal solvers;
- bounded glTF, OpenUSD, CAD, BIM/IFC, and robot-scene projections;
- read-only JSON and MCP resolver interfaces;
- authored architecture, assembly, robot, game, and consumer fixtures.

Passing authored fixtures establish deterministic mechanism behavior only.

## Suitable evaluation scopes

1. Reproduce B18 R1 from the frozen packet.
2. Evaluate one declared deterministic solver class on a new public-safe corpus.
3. Normalize a bounded source-format subset against a golden projection and explicit omission list.
4. Inspect branch isolation, merge-candidate construction, and acceptance boundaries.
5. Define a new training/evaluation split for a future context experiment without tuning the frozen holdout.

## Public evidence

- Runtime overview: https://www.glyphd.com/sft
- D25 runtime: https://www.glyphd.com/demos/d25-semantic-fractal-topology
- N20 architecture: https://www.glyphd.com/notes/n20-semantic-fractal-topology
- B18 benchmark: https://www.glyphd.com/benchmarks/b18-semantic-fractal-topology
- B18 R1 receipt: https://www.glyphd.com/evidence/sft-b18-r1/receipt.v1.json
- B18 R1 raw rows: https://www.glyphd.com/evidence/sft-b18-r1/raw-rows.v1.csv

## Contact and legal boundary

The evaluation route can generate a browser-local request for a technical or licensing discussion. No public artifact creates confidentiality, grants implementation rights, offers a warranty, or supplies a patentability, freedom-to-operate, professional-engineering, building-code, or safety conclusion.