Multi-producer x multi-consumer x multi-sink stress matrix #3

New Issue

2026-06-03T16:21:46+02:00

Brummel commented

2026-06-03 16:21:46 +02:00

Goal

Prove the pure compute substrate carries arbitrary M-producer x N-consumer x K-sink
DAGs deterministically and correctly. This is the milestone's central demonstration
("the infrastructure works") and it replaces the trading-half fieldtest that was
removed from the milestone scope. No trading domain.

Depends on

The sink issue — real multi-sink recording must exist first. Until then a run can
record only one stream.

Coverage

A systematic test / fieldtest matrix over the topology shapes:

source fan-out (one source into several nodes) — has a test today
node fan-out (one producing node read by several consumers) — supported but UNTESTED
fan-in (several producers into one consumer) — has a test today
diamond rejoin (unequal-latency paths rejoining) — has a test today
deep chains and wide layers — untested

crossed with:

multiple sinks recording distinct interior streams in one run
mixed scalar kinds across sinks (i64 / f64 / bool / timestamp)
a sink recording all K fields of a multi-field producer record (ties cycle 0005 in)
the firing modes (A as-of, B barrier, mixed) under fan-out / fan-in

Each combination asserts:

correctness of every recorded stream, and
determinism (C1): a second identical run is bit-identical.

Why now

Node fan-out and multi-sink are the two coverage holes the current engine tests leave
open. This issue closes them and becomes the green end-to-end gate for the milestone.

## Goal Prove the pure compute substrate carries arbitrary M-producer x N-consumer x K-sink DAGs deterministically and correctly. This is the milestone's central demonstration ("the infrastructure works") and it replaces the trading-half fieldtest that was removed from the milestone scope. No trading domain. ## Depends on The sink issue — real multi-sink recording must exist first. Until then a run can record only one stream. ## Coverage A systematic test / fieldtest matrix over the topology shapes: - source fan-out (one source into several nodes) — has a test today - node fan-out (one producing node read by several consumers) — supported but UNTESTED - fan-in (several producers into one consumer) — has a test today - diamond rejoin (unequal-latency paths rejoining) — has a test today - deep chains and wide layers — untested crossed with: - multiple sinks recording distinct interior streams in one run - mixed scalar kinds across sinks (i64 / f64 / bool / timestamp) - a sink recording all K fields of a multi-field producer record (ties cycle 0005 in) - the firing modes (A as-of, B barrier, mixed) under fan-out / fan-in Each combination asserts: 1. correctness of every recorded stream, and 2. determinism (C1): a second identical run is bit-identical. ## Why now Node fan-out and multi-sink are the two coverage holes the current engine tests leave open. This issue closes them and becomes the green end-to-end gate for the milestone.

Brummel added this to the Producer/consumer/sink substrate milestone 2026-06-03 16:21:46 +02:00

Brummel added the feature label 2026-06-03 16:21:46 +02:00

Brummel referenced this issue from a commit

2026-06-04 15:05:02 +02:00

test: M×N×K stress matrix — node fan-out, deep chains, wide layers

Brummel closed this issue

2026-06-04 15:05:02 +02:00

Brummel referenced this issue from a commit

2026-06-10 16:27:46 +02:00

spec: 0025 render root like any composite (slot stubs + role-name entries)

Brummel referenced this issue

2026-06-25 15:42:32 +02:00