Snapshots, not deltas

This is the rule that catches most integrations.

• Every story.context carries the whole story, not just what changed. Leave out a field you didn't change, and every consumer erases it.
• Every snapshot carries a sequence_number that strictly increases for that story, and an updated_at that never goes backwards.
• An asset's asset_type never changes. An asset doesn't silently disappear from a later snapshot.
• A story that reached a terminal story_type (KILLED, SPIKED, ARCHIVED) doesn't come back.

seq 1  PLANNED   headline
seq 2  ACTIVE    headline, lifecycle, assets[feed]
seq 3  ACTIVE    headline, lifecycle, assets[feed, transcript]        ✓ whole story, seq up
seq 3  ACTIVE    …                                                    ✗ 409 sequence_number.not_increasing
seq 4  ACTIVE    headline, lifecycle, assets[transcript]              ✗ 409 snapshot.assets_dropped

Snapshot order

A single message can't show that it breaks a story; only the previous snapshot can. So the bus remembers what it needs from the last accepted snapshot of every story (its sequence number, updated_at, story_type, top-level members, asset ids and types, and which system wrote it and owns the story, never the payload), with no expiry, and checks each new one against it (SOM conformance, section 6). These checks report source: sequence:

Rule	Outcome
sequence_number.not_increasing	Refused, 409
updated_at.backwards	Refused, 409
snapshot.members_missing	Refused, 409
snapshot.assets_dropped	Refused, 409
asset.type_changed	Refused, 409
story_type.left_terminal	Refused, 409
sequence_number.gap	Accepted, with a warning under tolerated
correlation_id.changed	Accepted, with a warning under tolerated

Tests and reruns

Because the bus keeps each story's sequence state with no expiry, a test run needs a new story_id. A rerun under an old one starts again at sequence_number 1 and is refused as stale. Story state is kept per workspace; prefix your story ids with your app name so your own apps' runs don't collide.

After a refusal

Resynchronise from your own state: send a fresh, complete snapshot with the next sequence_number and a new message_id. The Playground in the portal, and your workspace's validate route, check a snapshot against the story's current state without publishing it.