One bus for story context.
Every tool on it.

No. A skill declares what is true about a story; the tool that owns the executor decides what to do about it; a person approves anything that reaches an audience. A clearance is granted to a specific version of the words, so a rewrite reopens the gate rather than inheriting the approval. If a rule cannot be evaluated, the output it guards stays held and a person decides.

It is about capturing their judgement. The decisions that shape a story already exist — in a chat thread, a rebrief, somebody’s head — and they are gone by the next shift. Writing them into the story puts each call on the record, attributed to whoever made it. The person moves from carrying state between systems to making the call.

One writer. The publisher’s story management system mints the story and owns its sequence_number — never the wire, not even on a flash. Everything else contributes through its own message family. Where skills disagree, gates combine by conjunction (any one hold means held), compliance fails closed, and a transform never lifts a hold.

Each newsroom runs its own, like any other message infrastructure in the building. When it is down, tools keep working standalone as they do today and reconcile on reconnect — nothing depends on the bus to stay on air. Because the story republishes in full, a tool that comes back reads one object and is current.

No. SOM defines an envelope and seven payload families carried on an ordinary publish/subscribe bus. The only topic rule is that a topic begins with som. — and a consumer never infers the payload type from the topic anyway; message_type is the only discriminator.

A retrieval query answers a single question for a single tool, and nobody else sees the answer. A production needs the opposite: one state every tool reads identically, gates that still hold when work crosses from one vendor to another, and an attributed record of each decision. The standard’s position is that models are consumers of SOM, not a substitute for it.

Never. Assets are references. At the TAMS junction a story points at a Source URI and a time range; a MAM or any other store is addressed the same way through a locator. Binding media to a story happens at the editorial act, never at ingest.

They stay. SOM is not a MAM, a rundown or a product and replaces nothing. MOS keeps working, with a bridge that makes it smarter, and what you have built with agents stays yours. One system is enough to start.

Put it under extensions["com.{vendor}.{field}"]. Consumers MUST ignore extension keys they don’t recognise and MUST NOT reject the message for them. Vendor enum values use the x- prefix instead — x-graphics_pack.

Because consumers join late, restart and miss messages. A story.context message is the complete state of the story, so omission means absent, not unchanged. A writer that sends only its own fields silently erases everyone else’s work — the conformance statement calls this the single most damaging error an implementation can make.

In the open. Every change arrives as a pull request to storyobjectmodel/som and anyone may open one. Within 1.x only additive changes ship; removing or renaming anything waits for 2.0, and a deprecated field keeps working until then. What is not yet settled is listed in the open register.

Nobody. The standard is open and unowned. Schemas, examples, tools and skills are Apache 2.0; the specification prose is CC BY 4.0. There is no fee and nothing to sign.

The method the model came from. Take one real story after transmission and trace how its context actually moved, hop by hop, between people and systems — where knowledge was created, where it was lost, and who carried it by hand. Story, Asset and Telling were transcribed from those sessions rather than invented.