Stage Bots

Stage Bots are automations that fire when a story moves into a stage. The automation runs a bot procedure, written in Tactus, which writes a new version of the story. Then the engine, not the procedure, moves the story to the next stage. The next stage's bots fire in turn, so a chain of stages becomes a chain of bots.

This is built and tested. Distinctive Mark, our open-code demonstration site about trademark law, runs one: a TTAB ruling is analysed into a draft, fact-checked against the opinion, copy-edited with the site's Editorial Checks, and then reaches typesetting, where no bot waits and an editor publishes. In our runs so far it costs about a cent and a half per ruling.

The rules that keep it safe

• One run per story, stage and version. Delivering the same stage twice starts nothing twice.
• A hop limit. After five bot runs on a story (or the limit on the automation) nothing more starts and a hard-gate Review Request says so.
• Bots never publish. A bot saves a version. It cannot move the story and cannot publish; the engine moves it, and an editor publishes.
• Bot actions are labelled. The version a bot wrote says so, and the story's trail records each move, who made it and why.
• A bot can stop the chain. A bot that cannot do its job, for example a fact-checker that could not read the opinion, returns a hold with a reason. The story stays where it is and a flag asks a person to look.
• Human review is not a stage. Wherever a person is wanted, the engine raises a Review Request.

Set it up

papyrus automations create --id distinctive.ruling.fact-check --title "Fact-check a ruling story" \
  --on-stage reporting --next-stage copywriting --monitor ttab --publication distinctive \
  --procedure distinctive.ruling.fact-check --max-hops 5

The procedures are in the open: fact-check and copy-edit. Not built yet: a per-story cost cap, and firing from a move on the console board (the command line fires them). See the cookbook recipe.