Monitors: say it in a sentence, review the whole monitor, keep the receipts
A plain-English sentence becomes a complete, editable monitor: what it watches, the identifiers it found, the platforms, the matching rule and exclusions, the frequency, where alerts go and what it will cost — with the reason for each choice. Nothing is created until you switch it on; drafts wait in review.
Every finding now carries its receipts: the raw evidence row it was made from and the reasons it matched. Alerts and the monitor page lead with the findings that matter — severity, then how many platforms carry them, then engagement — and count people as well as posts.
Each run compares this window with the one before (posts, people and platforms, with acceleration and newly active platforms called out) and groups the same discussion across platforms; the counts are computed, never estimated.
Verdicts are remembered: mark a finding irrelevant and later runs hold back the same conversation, link, wording, topic and author until you change it; mark it known or resolved and it stays quiet unless the discussion changes materially — twice the posts, a new platform or a higher severity — and then it comes back saying why. Mark one useful and it guides the relevance filter. New endpoints monitors/propose, monitors/drafts, monitors/feedback and monitors/export; export gives the evidence as CSV or JSON with the raw rows.
Each monitor now says what a check gives you. Intelligence monitors — complaints, competitors, switching, buying, questions, mentions — lead with a report: what changed and why it matters, each line with its receipts, the discussions that span platforms, what came back after a verdict, and suggestions labelled as suggestions; the full evidence and the export stay a click away. Feeds, reviews and follow-alongs list every new item. Both are one switch on the review page. A check that finds nothing new says so and sends nothing.
Counts are honest about their limits: distinct accounts are counted per platform and labelled so; a trend is claimed only when both periods were collected alike, otherwise the counts are shown with the limitation stated. Webhook endpoints can be created as Slack or Discord webhooks and receive the report in each platform’s own format; a monitor switched on without a test is marked as such until its first check.
A report may say only what its records establish. A model checks each sentence against the records it cites for its meaning — who said it, negation, doubt and time — and keeps it, rewrites it to what they establish, or sets it aside; names, numbers, dates and quantifiers are then checked deterministically, and the engine’s counted lines are recounted from this monitor’s records. Every record is counted as a firsthand claim, publisher coverage, a relayed item, the subject’s own side or unplaced, so circulation is never confirmation. The full set of records behind a count is one click away, a comment links to its thread, a missing time says so, discussions are named from their own evidence, posts found late are told apart from posts made since the last check, and what was set aside sits under Diagnostics on the monitor page rather than in the alert. With no model, the report shows counts and records and says the sentences were not checked.
Where alerts go is chosen per monitor, in plain words: the monitor page, your email, a Slack channel, a Discord channel, your own webhook, or your devices — each right away (from a severity you choose) or as a daily or weekly summary, so a monitor can check hourly, send notable findings to Discord at once and email a summary every morning. Slack and Discord are added by pasting the link their own settings give; a sample alert can be sent to any place before the monitor is on. Push to a phone or browser is offered when the deployment has push keys (on an iPhone, add Monocrawl to the Home Screen first). A line whose records were all recorded on earlier checks is background, not a development; a story repeated on two platforms is circulation, not corroboration.
Before a monitor goes live, the form says whether its monthly cap will sustain its schedule — hourly checks at ten credits run through a 300-credit cap in about a day — with how long the cap lasts and two one-click answers: raise it to what a month needs, or check less often. A post that only passes on a link and its title (a Hacker News story, a Reddit link post) now counts as relayed, not firsthand, and so does news shared from an outlet, whatever is said around the link; a story several accounts submitted is circulation, not several confirmations. The period comparison names both of its counts — posts made in the window and what this check added — and claims a multiple only against a base of five or more posts. An alert email’s link now returns you to that report after you sign in.
Two new ready-made monitors for games and creators: UGC bugs & exploit reports (your map, mode, game or mod, its other names and codes, and the five kinds of thing players report — worded as reports, never as confirmed faults) and Audience feedback & requests (a show, channel or creator, and what its audience asks for and says). Every path into a monitor — a preset, a sentence, the form — now ends on the same review: what is collected, what qualifies, how often, where it goes, what it costs, and what happens to what is not kept. Evidence links lead to the post or comment itself, with the page it linked to kept apart; a missing or future time is flagged and never counted as growth. The export offers two files: the retained findings, and everything one check collected with the decision on each item, the reason, and the check’s coverage.
Follow an account now reads the account’s own timelines — an X handle, a Bluesky handle, a YouTube channel, each optional — and says so; it never searches for the name. New Google Maps reviews takes a pasted Maps link and turns it into the listing (one places search, charged once, said before it happens).