Documentation
¶
Overview ¶
Package review records a person's verdict on the answers the bot composed in shadow mode, which is what stands between shadow mode and posting (spec 0010 D12, spec 0023).
It reads the answers pkg/respond records and owns only the verdicts. Deciding that the answers are worth posting is left to the person reading the tally: nothing here lifts shadow mode (0023 D7).
Index ¶
Constants ¶
const Target = 30
Target is how many reviewed answers to real questions 0010 D12 asks for before shadow mode can be lifted (R-ANS-18).
Variables ¶
var ( // ErrNotReviewable is returned for an answer that cannot take a verdict: // posted, over budget, expired, or not there at all (0023 D3). ErrNotReviewable = errors.NewSentinel("review.not_reviewable", "review: not a reviewable shadow answer") // ErrCitationsUnjudged is returned when an answer with citations is given // no verdict on them (0023 D5). ErrCitationsUnjudged = errors.NewSentinel("review.citations_unjudged", "review: the answer's citations need a verdict") // ErrNoCitations is returned for a citation verdict on an answer that has // none to judge (0023 D5). ErrNoCitations = errors.NewSentinel("review.no_citations", "review: the answer has no citations to judge") )
var ErrServerRequired = errors.NewSentinel("review.server_required",
"review: more than one server is configured; name one with --server")
ErrServerRequired is returned when several servers are configured and the reviewer did not name one (spec 0029 D4).
var ErrUnknownServer = errors.NewSentinel("review.unknown_server", "review: that server is not configured")
ErrUnknownServer is returned for a --server that is not configured.
Functions ¶
func Open ¶
Open opens the database at path, the one serve records into (0023 D2), and returns a recorder over it with the function that closes it.
func Server ¶ added in v0.11.0
func Server(configured []chatplatform.ID, named string) (chatplatform.ID, error)
Server picks the server a review is for: the one named, which must be configured, or the only one configured. With several configured, naming one is required, so production is never judged by someone who thinks it is stage.
Types ¶
type Item ¶
type Item struct {
AnswerID int64
// Question is the question as stored, redacted. Empty, with QuestionExpired
// set, once the question's retention has passed.
Question string
QuestionExpired bool
Posture respond.Posture
Title string
Body string
Cited []answer.Citation
RecordedAt time.Time
// Posted reports whether the answer reached the channel, rather than
// being composed in shadow mode (0025 D8).
Posted bool
}
Item is one answer as the reviewer sees it (0023 D4).
type Queue ¶
type Queue struct {
// contains filtered or unexported fields
}
Queue walks the reviewable answers, oldest first.
Each Next reads the database afresh rather than a list taken at the start, so an answer serve records mid-review joins the end (0023 D9). Skips live on the Queue and nowhere else: a new run offers them again.
func New ¶
func New(rec *respond.Recorder, space chatplatform.ID) *Queue
New starts a review run over the answers rec records for one server (spec 0029 D4).
func (*Queue) Exclude ¶
Exclude marks an answer as not counting, for a replayed benchmark or calibration question (0023 D5). It is never offered again.
func (*Queue) Next ¶
Next returns the oldest reviewable answer not skipped in this run, or false when none is waiting.
type Tally ¶
type Tally struct {
Target int
Reviewed int
Answers int
Declines int
Posted int
Correct int
CitationsJudged int
CitationsApt int
Excluded int
Waiting int
// Readers' 👍 and 👎 across every answer, and how many answers have any
// (spec 0027 D8). Totals only, so no vote sits beside a verdict.
Helpful int
NotHelpful int
Voted int
}
Tally is what `review status` reports (0023 D7). Reviewed excludes excluded answers, and Answers and Declines split it.