Reading a decision card: recommendation, impact, and evidence

What a decision card is

The Feed is a stream of cards, one full screen at a time, that you move through the way you would a stories app: read, act or dismiss, move to the next. Most of what appears there is content (articles, vendor news, video), but the cards that matter most are decision cards. A decision card is Verinode IQ surfacing something it found in your data, a cost that is drifting, a receivable that is stalling, a certification about to lapse, together with a dollar estimate where one is knowable, a plain-language reason, and a recommended next step. This article is about how to read one: what each number means, where it comes from, and what happens when you tap through.

Verinode does not decide anything on this card. It surfaces the pattern, states its confidence in the number, and recommends an action. You act, park it, or ignore it.

Where to find it

Open Feed from the sidebar (iq.verinode.ai/feed). The deck opens on a welcome or briefing slide, then scrolls one card per screen, either with your mouse wheel, the arrow/j/k keys, or by tapping. A filter bar above the deck narrows it to specific card types, with a count on each filter chip so you know how many of each are waiting.

Every card carries the same top bar regardless of type: a source icon and name on the left (Verinode IQ for decisions, with the IQ mark, since a decision always comes from your own agent, not an external publisher), a relative timestamp ("2h", "3d", "1w"), and on the right a Resurfaced badge if the card previously left your feed and has come back, plus a type badge (Decision, or Decisions when several have been bundled into one card).

The two decision card layouts

Verinode renders a decision one of two ways, depending on whether the underlying signal was produced by the newer decision-engine pipeline or an older detector. Both carry the same information, evidence and recommendation, confidence, dollar figure, just arranged differently.

The flip card (decision-engine cards)

This is the flip-to-consequence layout. The front face is the pitch:

• A colored pill at the top names the decision's direction: Revenue (with an up arrow) if it grows income, Cost (down arrow) if it cuts a cost, Cash Flow (up arrow) if it accelerates money in, or Risk (down arrow) if it lowers exposure. Next to it, the entity the decision is about (a vendor, client, or job) if one is attached.
• Below that, the dollar figure. What sits here depends on the register the detector assigned the number (see "Understanding the dollar figure" below): a bare recoverable figure, a figure paired with the condition it depends on, or, for risks that cannot be priced (an expiring certification, a missing insurance document), a plain consequence line instead of a dollar.
• If Verinode has enough peer data, a one-line peer comparison sits just under the figure, framing where you stand relative to others in your position.
• A consequence line in slightly muted text: the plain-English "if nothing changes" sentence.
• A copper action box: "Tap Act. Your assistant will review the data and prepare a plan." This is a preview of what pressing Act does, it does not do anything itself.
• A row of small pills underneath: a confidence label, an effort label, and a realization label, short phrases Verinode's decision framer writes to tell you at a glance how sure it is, how much work the fix takes, and how soon the benefit lands. Not every card carries all three; only the ones the framer populated appear.
• At the very bottom, "Tap for details."

Tap anywhere on the card and it flips (a genuine 3D rotation) to the back face, the evidence:

• Proof points, short factual statements backing the claim (for example, a specific rate, a specific date range, a specific comparison).
• A Causal chain box: the reasoning that connects what Verinode observed to why it means what it says it means, in a sentence or two.
• "Based on N data points," when the signal comes from a recurring pattern rather than a single observation, so you can gauge how much history sits behind the call.
• The same decision-type pill repeated for context.
• "Tap to flip back" to return to the front.

The flip is the whole point of this layout: the front is the decision you'd make in five seconds, the back is what you'd want to see if a partner asked "how do you know that."

The standard decision card (non-engine signals)

Older detectors render a flatter card, no flip:

• A severity dot and word (colored and capitalized: Info, Warning, or Critical) with the entity name beside it, and, if the signal carries one, a confidence pill (High, Medium, or Low, colored green, yellow, or neutral).
• The dollar figure, in the same large numeral treatment, when one exists. When it does not (an uninsurable or unpriceable risk), a risk-kind label ("Compliance risk," "Cash flow risk," "Revenue risk," or "Business risk," chosen from the signal's domain) paired with an urgency word ("Urgent," "Act soon," or "When you can," derived straight from severity) takes its place, followed by the plain consequence sentence.
• The card's title.
• An IQ recommends box holding the recommendation sentence, or, if there is no formal recommendation yet, the signal's underlying body text instead.
• A benchmark strip, when the signal has a computed peer delta: "Benchmark: +N.N std dev from median" (or negative), colored by how far off the median you sit.

Bundled cards

When Verinode has grouped several related signals (say, three vendor cost issues) into one card to keep your feed from becoming a wall of near-duplicates, you get a Decisions badge instead of Decision, a combined dollar total across every signal in the bundle, the shared recommendation, and up to four of the individual signals listed below it (with "+N more" if the bundle is larger). Acting on a bundle does not commit to any single item, it opens the Decisions workspace filtered to exactly the signals the bundle represents, so you can look at each one and decide separately. See the decision workspace for what that filtered view looks like.

Understanding the dollar figure

Every impact dollar on a decision card carries a label above it and, depending on its register, a different rule for how the figure can be shown. Verinode classifies every decision's dollar into one of two registers you'll actually see on a card (a third, profile-gap, is used internally to route missing-data prompts away from the feed entirely, so you won't see a bare dollar figure attached to a gap):

• Hard (recoverable). The label reads "Recoverable" (or a more specific phrase your detector wrote, such as a particular clause it found) and the dollar stands on its own: "$X recoverable." These are figures Verinode is confident are actually collectible, an overcharge, a rebate you are owed, a rebillable line, so the number is never softened with "at risk" or "exposure" language.
• Soft (conditional). The label reads "Could recover" (or a reviewer-written variant) and the dollar always travels with the condition it depends on, for example "~$X/yr if you act on this" or a more specific clause the detector supplied. A soft dollar is never shown bare, the condition is part of the sentence, not a footnote, because the number only holds if that condition is true.

The unit attached to the figure matters as much as the number. Verinode tags every dollar with a time base, and the card renders the honest one: "/yr" for an annual run rate, "/mo" for monthly, "/wk" for weekly, "/job" for a per-job figure, and "one-time" (not "/yr") for a single pool of money like a stale-WIP catch-up or a one-off billing correction. A one-time recovery is never relabeled as if it repeats every year, that would overstate the number.

Where the figure comes from is disclosed separately from the register. When you open the full decision (tap Act, which takes you to the decision workspace), the header shows whether the number is Calculated or Estimated:

• Calculated means Verinode's confidence in the underlying math is high, the figure is built directly from your own records.
• Estimated means the confidence is lower, the figure is a reasonable projection rather than a hard tally.

Sometimes that tag carries a second word after it: Peer cohort means the estimate leans on how comparable operators in your position typically perform, and Industry baseline means it leans on a broader industry reference point rather than your own peer group specifically. Not every card shows the refinement, it only appears when the detector recorded which kind of baseline it used.

Confidence, effort, and realization

On the flip-card front, the small pill row underneath the action box is where Verinode packs three separate judgments into a glance: how sure it is the pattern is real (confidence), how much work acting on it takes (effort), and how soon you'd feel the benefit (realization). These are written per-signal by the decision framer, so exact wording varies, and a card only shows the pills it actually has values for.

On a standard decision card, confidence shows up differently, a colored pill next to the severity dot reading High, Medium, or Low, taken straight from the signal's own confidence field.

Why it matters

Below the headline number and title, Verinode writes a short "why this matters" sentence connecting the pattern to a consequence you'd care about, distinct from the recommendation itself. On the flip card this is the consequence line on the front face; on the back, the causal chain box does the same job in more depth. This is where Verinode explains its reasoning in plain language rather than showing you the underlying math.

Acting on a decision

The bottom of every decision card holds three buttons in a row:

• Act (the primary button, sometimes relabeled: "Do it" for an action-plan step, "Reply" for an agent message, "Review N" for a bundle). For a single decision, tapping Act takes you into the full decision workspace at /decisions/[id], this is where the recommendation "expands": the complete argument, the Calculated/Estimated basis, the plan Verinode drafted, and any track record from calls like this that have worked for you before. See the decision workspace for a full walkthrough of that view.
• Not Now, the middle button, opens a small picker asking "Why not now?" with quick reasons (Too busy, Need more info, Not convinced, or Other). Parking a decision this way is not the same as dismissing it, a parked card is expected to resurface later.
• Ignore, the third button, dismisses the card. A brief "Dismissed" confirmation appears with an Undo link before the card actually leaves your feed, so a stray tap is recoverable.

Bundled decision cards dismiss or park every signal in the bundle at once, not just the first one, so declining a "3 vendor issues" card clears all three.

Empty states

If you have reached the end of your feed, the deck ends on a card that reads:

All caught up Verinode IQ is continuously analyzing your data and scanning industry sources. New decisions, insights, and updates will appear here as they surface. Check back soon. Your next briefing is building.

If you have applied a filter that currently matches nothing, you'll see: "No items match this filter." Neither state means anything is broken, it means Verinode has not yet detected a new pattern worth a card, or the filter you picked has nothing in it right now.

Related reading

• The Feed: the deck itself, filters, briefing card, and how cards get into your feed in the first place.
• The decision workspace: the full argument, plan, and evidence you land on after tapping Act.
• Acting on decisions: what happens after you commit to a recommendation.
• How benchmarks work: how the peer-cohort and industry-baseline comparisons behind a decision's basis are built.