User:Shirik/CollabRC/Commands

The distributed network relies on commands on an IRC channel to enable a many-to-many relationship between humans and bots. One bot, User:CollabRCBot, acts as a client that automatically makes decisions on its own. Other bots make decisions which flag edits for review with a given probability. If CollabRCBot decides that the probability is high enough to warrant an automatic revert, then it is done. Otherwise, human clients will prioritize flagged edits to give them an extra review.

The following is a list of IRC commands that enable this functionality.

Bots

FLAG <oldid> <probability>

Marks an edit as appearing to be vandalism. The probability indicated is the likelihood that the edit is vandalism. This is used for prioritizing and automated reverting. Probability may also be ??? indicating no assertion is made towards the confidence of the report.

Secondary commands

!die

Kills the bot, no questions asked. Please PM the bots so they dont all die.

!rm "<username>"

Removes <username> from the blacklist, recognized by DASHBot-2

!add "<username>"

Adds <username> to the blacklist, automatically removed after 6 hours.

!empty

Removes all users from the blacklist.

!purge

Forcibly remove all users in the blacklist that are older then 6 hours. There is no real need to do this, but it is there for debugging.

!assess

Returns the bot's verdict on a particular edit, no learning action taken. Good for debugging.

Clients (including CollabRCBot)

WILLREVIEW <oldid> <random unsigned 32-bit integer>

Indicates that the client will take this edit into its queue. Only two human clients at a time should review an edit. This allows for a more even distribution of edits. The top two random integers after 5 seconds will be given the edit.

GOOD <oldid> [nofeedback]

Indicates the edit was determined to be in good faith. If the nofeedback flag is supplied, then bots should not feed back this "good" result into its filtering scheme.

BAD <oldid> [nofeedback]

Indicates the edit was determined to be vandalism and was reverted. All clients should discard this edit if they are tracking it. If the nofeedback flag is supplied, then bots should not feed back this "good" result into its filtering scheme. This flag is used by CollabRCBot because the determination is made based on the AI, so this would result in a feedback loop.

Additional flags

• Any command can be followed by the command REM on the same line. Any text following REM will be treated as a remark which does not impact automated functionality. Bots may supply this to provide additional information about the decision that was made (such as REM contains repeating characters) Clients may use this to provide additional information to users.