# Horizon RunMap

> Experimental MoE execution architecture by Lucas Ribeiro: selective expert
> compression, compact W4A16 computation and budgeted GPU residency.

Research entry point, reviewed 2026-09-16.
Start with the technical brief, then follow the relevant experiment and source.
The human presentation remains available in English and Brazilian Portuguese.
The runtime source is private by author decision; public evidence is linked below.

The engineering question is how storage precision, execution precision,
residency and data movement can be coordinated under limited GPU memory.
The documented immutable cache retains complete host authority and disposable
qpacked GPU copies, with H2D-only misses and complete-generation publication.
Contract invariants are scoped to their profile, not every historical executor.

OLMoE c30-c60 use PRODUCT-core scalar; c64 uses Grouped T3.
Do not treat their difference as a residency-only ablation.
IFEval 541 and the earlier 120-pair study are separate experiments.
Frontier models motivate the design; they are not the measured population.
Published verification does not constitute independent runtime reproduction.
The project makes no fastest-runtime, SOTA or cross-runtime superiority claim.
Quantization, offloading and caching are established prior art.
Original BF16 values, low-bit source and FP16 execution values are distinct.
Evidence level, validation, public admission and reproduction are separate.
Missing observations remain UNKNOWN; unavailable capabilities UNSUPPORTED.
This index provides context; retrieval and model prioritization are not guaranteed.

## Start here

- [Technical brief](https://horizonrunmap.com/technical-brief.md): compact canonical evaluation context and reviewer notes.
- [Architecture](https://horizonrunmap.com/architecture.md): representation, authority, staging, publication and executor scope.
- [Evidence](https://horizonrunmap.com/evidence.md): three experiments, exact fractions, contracts and verification limits.
- [Related work](https://horizonrunmap.com/related-work.md): sourced architectural comparison, including unresolved properties.

## Structured records

- [Claims](https://horizonrunmap.com/claims.json): versioned supported claims, non-claims and source-bound observations.
- [Architecture record](https://horizonrunmap.com/architecture.json): versioned profile-specific invariants.
- [Build inventory](https://horizonrunmap.com/agent-interface-manifest.json): SHA-256 inventory of this generated interface.

## Public evidence editions

- [Performance manifest](https://horizonrunmap.com/evidence/releases/technical-preview-residency39-v3/manifest.json): 13 configurations and 468 measured responses, E3.
- [Performance method](https://horizonrunmap.com/evidence/releases/technical-preview-residency39-v3/audit-method-v1.md): curated aggregates and hash verification.
- [Technical source map](https://horizonrunmap.com/evidence/releases/technical-preview-residency39-v3/horizon-technical-notes.md): architecture contracts and historical scope.
- [Behavioral manifest](https://horizonrunmap.com/evidence/behavioral/olmoe-0924-fidelity-v2/manifest.json): 120 BF16/Horizon pairs, E3.
- [Behavioral method](https://horizonrunmap.com/evidence/behavioral/olmoe-0924-fidelity-v2/audit-method.md): retained-response structural rescoring.
- [IFEval manifest](https://horizonrunmap.com/evidence/ifeval/olmoe-0924-ifeval541-v1/manifest.json): 541 prompts, E1 standalone_benchmark.
- [IFEval method](https://horizonrunmap.com/evidence/ifeval/olmoe-0924-ifeval541-v1/audit-method.md): frozen scorer and complete retained population.

## Report

- [Architecture, Experimental Evidence and Scope](https://horizonrunmap.com/technical-report-v1.md): principal technical report, version 1.
- [Report HTML](https://horizonrunmap.com/technical-report-v1.html): readable and printable report.
- [Report PDF](https://horizonrunmap.com/technical-report-v1.pdf): portable version of the same report.

## Optional

- [English presentation](https://horizonrunmap.com/en.md): full Markdown alternative to /en.
- [Apresentação em português](https://horizonrunmap.com/pt.md): alternativa Markdown completa de /pt.
- [Performance release index](https://horizonrunmap.com/evidence/publication-index.json): current and historical editions.
- [Behavioral release index](https://horizonrunmap.com/evidence/behavioral-index.json): separate paired-study discovery.
- [IFEval release index](https://horizonrunmap.com/evidence/ifeval-index.json): separate standalone-benchmark discovery.