🧠 Beyond Text: Cache Compression and Knowledge Transfer in Multi-Agent Systems
UQO Working Paper No. 30 — A Reproducible Study of Fidelity, Communication Cost, and Repeated Handoffs
TL;DR — Under which task, model, cache, network and reuse conditions does each communication representation (raw text, summaries, structured evidence, lossless codecs, learned prompt compression, native/quantized/evicted KV caches, learned cache mappings) preserve useful information at the lowest measured end-to-end cost? Real experiments on Apple Silicon (MLX) with Qwen2.5-3B / Llama-3.2-3B, genuine two-node cache transfer over a 1 GbE LAN, repeated-handoff chains with an evaluator-only fact ledger, fan-in with conflicting evidence, a same-family learned KV mapper, hosted-API receivers, and a development-fitted router.
📌 Status
See STATUS.md (phase gates), DECISIONS.md (dated decisions), FINAL_REPORT.md (completion status and limitations).
🗂 Repository layout
protocol/ hypotheses.md (registered), analysis_plan.md, frozen.json (protocol hash), run matrices
literature/ sources.csv (34 verified sources), audit_group{1,2,3}.md, method_notes.md
configs/ models, data, resources (budgets/timeouts), pilot/core experiment configs
src/wp30/ backends (MLX), cache adapters (sparse/quantized/serialize/selection), methods (A0–A8), runner, chains, fanin,
transport (two-node worker), hosted (API track), mapper (Track C), router (E8), analysis, CLI
tests/ cache correctness (12), leakage (3), scoring/data (5)
data/ manifests (splits, licences), generated/docs (agent-visible), generated/keys (evaluator-only), raw/longbench
runs/raw per-attempt JSONL records (schema wp30-attempt-1), chains/ hop messages ; runs/logs ; runs/checkpoints (mapper, router)
results/ tables/ figures/ provenance.json claim_ledger.csv
paper/ LaTeX source + compiled PDF
reproduction/ requirements-lock.txt, minimal.yaml, report.json🚀 Quick start (analysis-only reproduction, no inference)
uv venv --python 3.12 .venv && source .venv/bin/activate && uv pip install -r reproduction/requirements-lock.txt && uv pip install -e .
make reproduce-analysis # rebuilds results/tables and results/figures from runs/raw
make paper # compiles paper/main.pdf 🔁 Inference reproduction (Apple Silicon, ~15 GB of checkpoints from Hugging Face mlx-community)
python -m wp30 doctor --output evidence/environment.json
python -m wp30 data prepare --config configs/data.yaml && python -m wp30 data validate --manifest data/manifests/splits.json
python -m wp30 models probe --config configs/models.yaml
python -m pytest -q # cache correctness suite runs a real 1.5B model
python -m wp30 run --config configs/core.yaml --resume
python -m wp30 reproduce --profile reproduction/minimal.yamlTwo-node experiments need a second Mac with the same checkpoint revision: python -m wp30 worker --port 9430 --model <repo> on the
receiver node, then python scripts/e6_network.py --host <ip> ... on the sender node. Hosted-API experiments need ANTHROPIC_API_KEY
/ OPENAI_API_KEY and are capped by configs/resources.yaml.
📄 Citation
Boucher, S.-P. (2026). Beyond Text: Cache Compression and Knowledge Transfer in Multi-Agent Systems. UQO Working Paper No. 30.