← Intelligence

Weekly small experiment · 2026-W45

Bridge
or bus?

A shared language removes many translations. It also creates one place where every conversation can stop.

Send the same multi-stage jobs through changing service interfaces. Direct bridges repair pair by pair. A shared layer repairs service by service. Static and random baselines show what each architecture is actually buying.

Loading the seeded world…

01 / integration policy

Direct bridges

Repair the first stale pair needed by the oldest job.

READY

02 / integration policy

Shared layer

Repair one service adapter and depend on a recoverable hub.

READY

03 / integration policy

Static bridges

Never repair after a schema changes.

READY

04 / integration policy

Random repair

Patch a seeded random stale pair from current demand.

READY

Rust nodes changed schema on this tick. The hub and version board are an evaluator view; all policies receive the same current state and no future events. A repair is one abstract action—not a measured engineering cost.

Try this first

Freeze the interfaces. Keep the outage.

Press Show outcome. Then set schema drift to Stable. Direct, static and random bridges now deliver every job in the saved batch; the shared layer still loses a fraction to hub downtime. Restore drift, then choose two repeated routes. A small working set changes the result again.

A shared language changes the repair surface.

Each job names an ordered route through distinct services. Direct delivery needs a current adapter for every adjacent pair. Shared delivery needs one current adapter for every service on the route—and a live hub. Schema drift invalidates adapters that mention an old integer version.

Every policy sees the same queue, versions, hub state and past outcomes. Each receives the same repair opportunities and can attempt the same first three queued jobs per tick. Direct and shared systems deliberately have different adapter counts and failure domains. That difference is the mechanism, not a hidden advantage.

The fan-out is useful. The dependency is real.

Mean delivered jobs out of 320 · seeds 1–64
SettingDirectSharedStaticRandom
Default127.234319.34415.71970.922
Stable schemas320319.344320320
Rapid drift53.828317.4387.04720.359
Twelve services39.562319.28111.15621.250
Fragile hub127.234284.85915.71970.922
Two repeated routes319.688319.32820.719319.547

In the default batch, the shared layer averages 37.156 repairs and 0.326 ticks of delivery latency. Direct bridges average 152.625 repairs and 15.357 ticks because one version change can stale many demanded pairs. With two repeated routes, direct repair reuses a small working set and narrowly leads.

This is not a universal ranking. The shared layer compresses adapter fan-out; the direct system avoids a common outage boundary. Stable interfaces, route concentration, repair capacity and hub reliability change the answer.

What the controls change

Seed
Replays arrivals, routes, version changes, hub events and random repairs.
Services
Changes the number of distinct interfaces in the world.
Route length
Changes how many services each job must cross.
Schema drift
Per-service chance of incrementing its version each tick.
Hub outage
Chance the shared layer goes down while available; recovery stays at 35% per tick.
Repairs
Adapter-update opportunities available to every policy.
Route pool
Broad demand creates new combinations; a small pool repeats known routes.

Where the toy stops.

This is a deterministic teaching world, not an agent-framework benchmark, API-gateway test, service-mesh recommendation, production reliability forecast, biological model or trading system. Services and schemas are integers. Payload transformation, security boundaries, network latency, partial failure, compute and engineering cost are absent.

All 640 matched seeded worlds and 2,560 policy outcomes are retained across ten predeclared settings. No seed was removed. Delivery, expiry, blocked attempts, repair actions and latency remain separate. The private source scan supplied the question—not instructions, facts or public prose—and is excluded from the release.

Method & parameter guide ↓Source package ↓All batch results ↓Read the model

Previously: The Useful Core · Explore the eight small worlds.