"Discuss: taking a decision to the HQ agent"

What Discuss is

Every row in your Decisions log, a decision plan, an at-risk franchisee flag, a consent request, a program audit, a program violation, is a real row in a real network table. Discuss is the mechanism that hands one of those rows to your HQ Co-COO so it can reason about that specific item instead of the network in general.

Open a decision from the gallery and two things happen at once: the decision opens in the workspace panel in the middle of your screen, and the same click hands that decision to your agent panel on the right. The panel's description switches from the generic "Network Decisions" prompt set to language specific to what you just opened, its suggested-prompt row refreshes to match, and if the panel was collapsed it opens automatically. From that point, anything you ask the agent, "what's driving this," "draft the outreach," "how does this compare to last quarter," is answered against that one decision, hydrated with the underlying network data behind it, not a generic summary of the whole log.

Verinode never takes the action for you. Discuss grounds the conversation; you still decide whether to act, park, or ignore.

Where to find it

Open Decisions from the HQ sidebar, or go directly to hq.verinode.ai/decisions. It's the franchisor inbox: decision plans you've pushed to the network, franchisees flagged for intervention, consent requests you've sent, and program audits and violations, all in one gallery, timeline, or calendar.

Click any tile to open it. That single click is what puts the decision in front of your agent. There's no separate button to press first, opening the decision is Discuss.

The hq_kind token

Behind the scenes, every HQ decision row carries a signal type that starts with hq_. The moment you open a decision, that signal type is translated into one of five plain tokens, and it's this token, not the raw signal type, that decides everything downstream: which prompts the agent panel offers, and which network tables get read to ground the conversation.

| What you see in the gallery | hq_kind token | Where it comes from | |---|---|---| | A decision plan you've drafted, activated, or completed | decision_plan | the network data | | A franchisee flagged for at-risk intervention | intervention | the network data | | A consent request you've sent a franchisee (including Item 19 cohort invites) | consent_request | the network data | | A program compliance audit | program_audit | the network data | | A program violation | program_violation | the network data |

A decision that doesn't match one of these five, or hasn't finished loading its kind, falls back to a generic "Network decision" framing rather than failing outright.

What changes in your agent panel

Once a decision is bound, the panel's description and its row of suggested prompts change to match the kind. These are the exact prompts you'll see:

Decision plan. Description reads as the plan's adoption status. Prompts: "What does franchisee adoption look like so far?", "Draft a reminder broadcast to slow movers", "Which franchisees haven't acknowledged this plan?", "What's the next milestone?"

Intervention. Description reads "[Franchisee], at-risk intervention." Prompts: "What's driving the flag for [franchisee]?", "Draft an outreach note to [the operator]", "What's the right next step for this intervention?", "Show similar interventions in our cohort."

Consent request. Description reads "[Franchisee], consent request." Prompts: "Why did we ask for this consent?", "What does the operator gain by approving?", "Draft a follow-up nudge", "Show me other consent requests this quarter."

Program audit. Description reads "[Program name] audit." Prompts: "Summarize the audit findings", "Compare to the rubric thresholds", "Draft a response to the auditor", "Show similar audits across the network."

Program violation. Description reads "[Program name] violation." Prompts: "What triggered this violation?", "What are the resolution options?", "Draft a remediation plan", "Has this franchisee had similar violations?"

You aren't limited to the suggested prompts, they're a starting point. Type any question and the agent answers it against the same grounded data.

What actually grounds the conversation

Selecting a decision only tells the agent panel what kind of thing you're looking at, its name and a short summary. The real substance loads the moment you send your first message: a server-side hydration step reads the source row plus a small set of related rows from the same network, and folds all of it into the specialist's context for that turn. This is what lets you ask "how does this compare" and get a real answer instead of a guess.

What gets pulled, by kind:

• Decision plan. The plan's title, kind, status, and its created / activated / completed dates, plus its full body text. Then adoption: how many franchisees the plan was pushed to, split into acknowledged, in progress, completed, and declined, alongside the plan's network completion rate and its median days to acknowledge and to move to in-progress, when those have been computed.
• Intervention. The flagged franchisee, the flag reason, the intervention type, status, and when it was flagged, plus any internal notes on file. Then up to three prior interventions on that same franchisee, each with its date, reason, type, and status, so you can see whether this is a first flag or a pattern.
• Consent request. The franchisee, the request kind, the rationale you gave, which metrics the request covers, its status, when it was created and when it expires, and a decline reason if the franchisee said no. Then up to three other consent requests sent to that same franchisee, for pattern context.
• Program audit. The program's name, type, and status, the audit type, score, status, and when it was conducted, its findings text, and its next due date. Then up to three other recent audits in the same program, and up to three open or recent violations in that program, so a single audit reads in the context of the program's overall track record.
• Program violation. The program's name, type, and status, the violation's reason, severity, escalation state, and when it was detected and resolved. Then up to three other violations in the same program, for the same track-record context.

If the hydration step can't load (a deleted row, a network hiccup), the conversation still proceeds on the name and summary alone, just without the extra grounding, so a bad lookup never blocks the chat outright.

The privacy boundary, held throughout

Whether a franchisee's real name appears in Discuss, on the tile, in the agent's suggested prompts, and in the grounding it reads, depends entirely on how your network's data model is configured. Under the default, privacy-first configuration, independent-operators networks, a flagged franchisee is shown as an anonymized label like "Franchisee #4B2A," a stable per-operator token, everywhere, in the panel description, in the prompts, and inside the grounding text the agent reads. Only networks explicitly configured as a single-entity, multi-location operation show the real location name in its place. This is the same anonymization your Compliance and Reputation views use; Discuss doesn't add a second, weaker path around it.

Program audits and violations never carry a franchisee name at all, they're grounded on the program, not the operator. Consent requests and interventions are the two kinds that name a franchisee, and both go through the same anonymization gate before they ever reach the agent.

Discuss versus Act, Park, and Ignore

Discuss never changes a decision's status. It only grounds a conversation. The status-changing buttons on a tile are separate, and what each one does depends on the kind:

• Decision plan. Act moves a draft to active, then an active plan to completed. Park pauses an active plan. Ignore archives it from any state.
• Intervention. Act moves a queued flag to in-progress, then an in-progress one to resolved. There's no Park for an intervention. Ignore marks it no action needed.
• Consent request. HQ has no Act or Park here, franchisees approve or decline their own requests. Ignore withdraws the request.
• Program audit. Act approves a submitted or disputed audit. Park withdraws it. Ignore disputes it.
• Program violation. Act resolves it. Park escalates it. Ignore waives it.

Because a consent request only gives HQ a withdraw button, Discuss is the practical way to actually reason about one, whether to press for a response, drop it, or wait, before you decide to withdraw it.

Empty states

If your network has no decisions on file at all, the gallery reads: "No decisions for you yet. They'll land here as Verinode spots cost savings, risk, and growth opportunities in your data." There's nothing to discuss until a decision plan, intervention, consent request, or program audit exists.

If you open the agent panel without any decision selected, on the Decisions page generally, it shows the route-level framing instead of a decision-specific one: "Network Decisions," described as "Decisions That Need The Network," with four starting prompts: "What's the highest-impact decision right now?", "Walk me through this decision," "Which franchisees does this affect most?", and "Draft the outreach this decision needs." Open any specific tile and the panel narrows to that decision's own prompts and grounding, as described above.