[ARC-05] [TV-11] Pub/Sub Phase 2 — empirically-driven architecture

Phase 2 — **starts**: 6 July **ends**: 21 September
**Milestone: design proposal by mid-September**; the remaining weeks validate and harden it.

## Goal

Arrive at an empirically proven design for the pub/sub dissemination layer.

The formal models (#76) define the candidate topologies (the **M models**) and the adversary assumption: a fraction **μ** of silent Byzantine nodes. A topology is viable when it reliably produces **good graphs**:

> **Good graph** (the models' full-coverage criterion, `formal_spec/hybrid_dissemination/models/README.md`): every message of every honest publisher reaches all other honest nodes — in system terms, publishing from any honest node on a topic reaches *all* other honest subscribers of that topic. For directed models this is strong connectivity of the honest graph, judged on the standing per-epoch structure alone.

We test this Monte-Carlo style — generate many per-epoch graphs from the verifiable randomness, remove the adversarial nodes, check the remaining honest graph — and measure the trade-off axes: **good-graph probability** (security), **latency** (hops), and **bandwidth** (messages). We want to show that our configurations produce good graphs with overwhelming probability — or, where they fail, under exactly which parameters — and pick the configuration with the best security/efficiency trade-off.

## Outcome

1. **Complete formal analysis** of the M models (#76): validated laws and a cross-model recommendation.
2. **Complete prototype & experiments** (#79): the same models run empirically, complementing the analytical findings — the data-driven proof behind the design.
3. **Design proposal** (#91, mid-September): the measured architecture, its adversary model and the remaining gaps only an incentive/fee layer can close, positioning against existing art (GossipSub), and a runnable reference configuration — published as a **CIP** if time permits.
4. *(If time permits)* a **resource-constraint model** from the experiment measurements, to discriminate further should several M models prove compliant with our requirements.

## Out of scope (phase 3+)

Full Byzantine resistance (residual gaps go to the incentive layer) · incentive/fee mechanism design · on-chain integration · real network transport · provable-misbehaviour cryptography.