Skip to main content

​
Runs: what happens on a schedule

Each run does the same four things:
  1. Searches every source the listener uses, with the queries in its plan.
  2. Pulls the post text wherever the platform allows it, rather than judging a search result.
  3. Filters, cheaply first: page shape, off-topic, your exclusions, blocked domains, anything outside the data window, and duplicates. All of this happens before any model is called, so filler costs nothing.
  4. Judges what survives: a reranker scores every candidate against the listener’s intent, and only the best reach a classifier that labels intent, scores it, and writes one line on how you might reply.
A run that dies is marked as failed rather than left looking like it is still going. Every run records its step logs — sources searched, hits kept, results skipped as too old or duplicated, signals saved — and those are visible in the listener’s Activity tab and on the Run history screen. Rescanning further back is possible when you first set a listener up, so a new listener is not staring at an empty screen until its window fills.

​
Signals: what a run found

A signal is one public post, judged. Signals live with the listener that found them, because the vocabulary and the useful columns differ per objective.

​
The classes

Each objective has its own labels. The first group is what you see by default; the rest are stored and filed behind their own tabs rather than thrown away. Local prospect signals also carry the review’s star rating and a list of themes. The star rating and the score are deliberately drawn differently: they are two different numbers, and a one-star review about parking is a one-star review and a ten-score signal.

​
What you can do with one

  • Rate it. Thumbs up or down. This is not cosmetic — rated signals are fed back into that listener’s next classification as worked examples.
  • Save it. Saving adds it to your workspace’s Saved list. Saved is visible to the whole workspace on purpose: a star is a statement to your team, not a private bookmark.
  • Add it to a list. Named collections, so “Q1 outreach” works the same way Saved does.
  • Write a note. Notes are keyed to the signal and follow it everywhere, including onto its pipeline card.
  • Assign it to a teammate.
  • Add it to the pipeline, one at a time or in bulk — see The pipeline.
  • Draft outreach from it — see Outreach drafts.
  • Open the source, which is always one click away. Nothing here is a summary you have to take on trust.
Signals are marked read as you open them, so a busy morning leaves a clear line between what you have triaged and what you have not. The filters above the table are applied in the database, not in your browser, so the counts above the list always describe the list below it:
  • Classification — the tabs
  • Source — one platform, or all
  • Theme — for review-based signals
  • Search — matches words, with stemming, so report finds reporting
  • Saved — your workspace’s starred signals
Lists page as you scroll, newest post first, with undated posts after dated ones.

​
Roll-ups

Where one row per post is the wrong shape, there is a roll-up: one row per rival, per account, or per place, with the signals underneath. A competitor losing eleven users is one row, not eleven mentions. Programmatic access to signals and roll-ups is described under search_signals and get_rollup in MCP & agents.