Patterns to confirm: turning observations into SOP steps

What an observed pattern is

While you talk with IQ, an agent listens for a second thing underneath your question: how you actually run your business. If you mention how long a step takes, which chemical your crew uses, who handles a particular task, or a workaround you have adopted, the agent quietly captures that as a process observation. You never see an "I noticed that..." acknowledgment in the chat, and most conversations produce nothing at all.

An observed pattern is one of three kinds of record inside Processes, alongside My SOPs (the SOPs you have written or adopted) and Standards (IICRC, LEAN, OSHA, EPA, and state reference SOPs). Every observed pattern starts life pending, waiting on you. This article covers the observed-pattern profile: the inferred step, the artifacts the detector saw, the similar pending patterns sitting alongside it, the target SOP it would extend, and the confirm, edit, and dismiss flow that promotes it into a real, documented step.

Verinode does not decide your process is real. It surfaces what it heard you describe, in your own words, and you are the one who says "yes, that is how we do it" or "no, that is not right."

Where to find it

Open Processes from the Operations group in the sidebar (/processes). Pending patterns show up in two places:

• The Explore row, where a To Confirm tile shows how many patterns are waiting, with a sub-label reading "Nothing Waiting" at zero or "Tap To Confirm" once there is a backlog. A small dot grid under the number is sized to the backlog. Clicking it opens the All Processes view pre-filtered to observed patterns. Beneath the four KPI tiles in that same Explore row, any cluster that has already gathered enough matching observations renders as its own compact card labeled Pattern to confirm, with Confirm, Edit, and Dismiss buttons right on it.
• The All Processes view directly, or the Most recent row on the Processes home. Each individual pending pattern shows up as a record under the Pending Confirmations kind. Clicking one opens the full detail card this article walks through.

The hero at the top of Processes also carries a To Confirm secondary stat reading "Patterns Waiting On You" whenever the count is above zero.

Cluster tiles vs. the detail card

There are two related but distinct surfaces, worth telling apart:

• A cluster tile, on the Explore row, represents a group of observations that the nightly clusterer job already decided belong together on the same inferred step. Its badge reads "N× captured," the number of observations feeding that group. Confirm, Edit, or Dismiss here acts on the whole group at once.
• The detail card below opens from a single observed-pattern record. It shows that one observation's own evidence, plus other pending observations that share its category, so you can see how common this type of report is across your queue before you decide anything.

Both surfaces write to the same underlying record and both close out with confirm, edit, or dismiss. Use whichever one you land on.

What the detail card shows

Hero stats

Four stats sit at the top of the card:

• Status: Pending, Confirmed, Edited, or Dismissed. Green once confirmed or edited, neutral while pending.
• Confidence: the agent's own estimate of how sure it is that this is a real, repeatable process rather than a one-off remark, shown as a percentage. A confidence at or above 80% shows in green. This is the model's own confidence in the capture, not a vote from other operators, and it reads as a dash when the specialist did not attach one.
• Cluster: the number of other pending observations that share this one's category, plus this observation itself (so a record with no matches shows "1"; three category-siblings show "4"). A "similar pending" label appears underneath once there is at least one. This count is a simple category match to give you context, not the same grouping the nightly clusterer used to build a confirmable cluster tile.
• Captured: the date this observation was first recorded.

The eyebrow above the title also carries a stance pill: HEALTHY, DRIFT, or EXPOSED. For a pending pattern, this reflects how urgent your confirmation is, not the quality of the underlying process. It reads EXPOSED when a separate open finding is specifically tied to this record and is waiting on your decision to move, DRIFT once the pattern has sat unconfirmed for more than two weeks (or, for a fresher one, a calmer prompt naming the confidence and the specialist that captured it), and HEALTHY once you have confirmed, edited and confirmed, or dismissed it, because from the platform's point of view nothing is left outstanding.

Just under the stats, a short synthesis paragraph turns the stance, the gain from acting, and the cost of leaving it pending into one plain sentence written for your record specifically, not generic copy.

Inferred Step

This section restates what the agent captured:

• The inferred step text in full (it can run longer than the tile title truncates to). Reads as a dash if nothing was captured.
• Pattern Type, when the agent tagged one: a free-form label like "Artifact Extraction," "Peer Gap," or "Specialist Capture," shown in Title Case. It describes how the pattern was noticed, not whether it is correct.
• Captured by, naming the specialist agent that wrote the observation.
• Status, repeating the hero stat as plain text.

Sources

This section title reads Source for a single piece of evidence or Sources (N) once there is more than one. A line under the heading explains what you are looking at: each row is an artifact the AI detector saw on its way to this inferred step, and confirming the observation rolls these up as evidence on the resulting SOP step.

Each row shows:

• A type label: Email, Voice note, Photo, Upload, or Attachment, or a humanized version of the raw type when it does not match one of those.
• A timestamp, shown as relative time ("4h ago," "3d ago") for anything under 30 days old, or a plain date beyond that, and a confidence percentage for that specific piece of evidence when one was recorded.
• The snippet itself: the operator's own words, verbatim, in italics, not a paraphrase.
• A short caption underneath, when available: "from email [id]" when it came off a forwarded email, "from upload [filename]" when it came off an uploaded file, or a generic "from [id]" otherwise. IDs are shortened so the caption stays scannable. The caption line does not render when no identifying artifact field was captured.

This section is the evidence trail: if you are not sure why the agent thinks your team runs a step a certain way, this is where you check what it actually read. The section is omitted entirely when the observation carries no evidence.

Similar Pending

When other pending observations share this one's category, they show up here as Similar Pending (N), capped at eight, most recent first. A line under the heading explains the mechanic plainly: confirming a few of these together is what trips the cluster threshold and lets the SOP-drafter draft an update for you.

Each sibling row shows its own inferred step text, the specialist that captured it, the date, and its confidence percentage when available. This section is omitted entirely when no other pending observation shares the category.

Target SOP

When you already have an active (non-archived) SOP in this observation's category, this section names it and shows how many steps it currently has, plus the line "ready for an additional step." This is the SOP a confirmed pattern is destined to extend. If you have no SOP yet in this category, this section is omitted; you can still confirm the pattern, it just has nowhere to attach as a next step until an SOP exists in that category.

Open Tips and Findings tabs

Two tabs sit alongside the Details tab on every process record, observed patterns included, and both hide themselves when empty:

• Open Tips, badged with a count, lists open signals tied specifically to this record: a headline, a short rationale, and a recommended action where one exists.
• Findings, badged with a count, lists active decisions elsewhere on the platform that reference this record, each with its own headline and cost-of-inaction line.

Both are read surfaces. Acting on a tip or a finding happens from the decision workspace or acting on decisions, not from this card.

The confirm / edit / dismiss flow

The action buttons live on the cluster tile in the Explore row, not on the detail card itself, so the flow below describes what happens when you click them there. Use the detail card first to read the evidence, then act from the tile.

1. 1Read the pattern. The tile shows the inferred step text (or the raw captured snippet if no step description exists yet) and, underneath, the operator's own words in a quoted line. The corner badge reads "N× captured," the number of observations feeding this cluster.
2. 2Choose Confirm, Edit, or Dismiss.
3. 3- Confirm accepts the pattern exactly as captured. The button reads "Confirming…" while it saves, then the tile clears and a toast reads "Pattern confirmed."
4. 4- Edit opens a text box pre-filled with the inferred step, letting you rewrite it in your own words before accepting it. Click Save & confirm when ready; the toast reads "Confirmed with your edit." Cancel discards your changes and returns to the read-only view without confirming anything.
5. 5- Dismiss rejects the pattern. The button shows "Dismissing…" while it saves, then a toast reads "Pattern dismissed."
6. 6The cluster updates behind the scenes. Confirming, with or without an edit, cascades every observation feeding the cluster to a confirmed state, so they are available as evidence the next time an SOP-drafting pass runs. Dismissing unlinks the contributing observations from the cluster instead of confirming them: they are not thrown away, and if a stronger signal clusters them again later, they can still surface as a fresh pattern to confirm.

Why this matters

A pattern that sits in your head, or in one crew lead's head, disappears the moment that person is not the one running the job. Confirming a pattern is the first half of fixing that: it tells Verinode this is a real, repeatable part of how you work. Promoting it into a documented SOP step, via the target SOP shown on the card, or via the cluster and draft pipeline once enough related observations confirm, is what makes it stick for everyone else on your team.

Empty states

• Nothing pending. The To Confirm tile on Processes reads "Nothing Waiting" and shows 0. No cluster tiles appear on the Explore row.
• No sources captured. The Sources section on the detail card is omitted when the observation has no evidence array.
• No similar pending observations. The Similar Pending section is omitted when no other pending observation shares this one's category.
• No target SOP yet. The Target SOP section is omitted when you have no active SOP in the observation's category.
• No pattern type or captured-by specialist. Individual fields on the Inferred Step section simply omit themselves or show a dash when the underlying data was not captured.
• No open tips or findings. The Open Tips and Findings tabs do not appear at all when there is nothing in them.

Best-practice example

Say IQ's water-mitigation specialist notices you mention, three separate times across different conversations, that your crew waits a minimum of 48 hours before pulling drying equipment once a job reads under a certain moisture level. Those observations cluster overnight. The next morning, the Explore row on Processes shows a cluster tile: "Wait minimum 48 hours before pulling equipment," badged "3× captured." You open the underlying record first to read the actual quotes behind it, and all three sources line up with what you remember saying. Back on the tile, you click Edit, tighten the wording to match your crew's exact terminology, then click Save & confirm. The three contributing observations flip to confirmed, and because you already have an active Water Mitigation SOP, the pattern is now positioned to extend it as a documented step, so the next new hire who runs a water job follows the same standard your best crew already follows by habit.

Data sources

Related reading: The decision workspace, acting on decisions, forwarding documents, connecting your data.