Cost Hotspots

Overview

Cost Risk findings are, by default, a flat list — one entry per pattern detected, wherever it happens to sit. Cost Hotspots regroups that same list around a different question: not what patterns exist, but which specific nodes are actually where the money goes. A workflow can have eight Cost Risk findings scattered across eight different nodes, or eight findings stacked on one node — those are very different remediation jobs, and a flat findings list doesn't make the difference obvious.

What question it answers

"If spend is a problem, where exactly do I start looking?" Not a dollar figure — FlowScan doesn't see pricing or execution volume, so it never states one — but a ranked list of nodes, ordered by how much structural cost risk is concentrated there.

How to read it

Each hotspot is one node, with:

• nodeId / nodeName — the node the cost risk is concentrated on. Only Cost Risk findings are counted here; a node's overall finding count may be higher if it also carries, say, a Reliability finding.
• detectedMultipliers — a short, human-readable label for each distinct cost pattern found on that node, for example "LLM inside loop — cost scales with item count" or "No output cap — unbounded output tokens." Deduplicated, so a node with two related findings shows each multiplier once.
• costExposureScore — a per-node total, and not the same number as the workflow-level Cost Exposure Score described in the Cost Risk article. This one is a simple sum: every finding on that node contributes points by severity (critical 30, high 20, medium 10, low 5, info 1), added together. It exists to rank nodes against each other within this one list — read it as "worse than the node below it," not against any absolute scale.
• productionImpactTier — carried over from that node's highest-priority finding, so it can be cross-referenced against Break Points.
• fix — the recommendation from the single highest-priority finding on that node, the one worth doing first if the node has more than one.

Hotspots are sorted highest score first, so the top of the list is where cost risk is most concentrated.

Example: a node running an LLM call inside a loop, with no output cap, might show detectedMultipliers: ["LLM inside loop — cost scales with item count", "No output cap — unbounded output tokens"] and a costExposureScore well above a node with a single medium-severity finding, because two stacked high-severity patterns on one node add up fast — and that stacking is exactly what this block is built to surface.

How to decide what to do

Work from the top of the list down, and treat multiple multipliers on one node as a single combined fix rather than separate tasks.

1. Start with the highest-scored node. It isn't necessarily the single worst pattern in the workflow — that's what Break Points and rank are for — but it's where the most cost risk has accumulated in one place, which usually means one fix (moving an LLM call out of a loop, say) clears several detectedMultipliers at once.
2. If detectedMultipliers lists more than one pattern on the same node, look for a single structural fix rather than fixing each independently. A call moved outside a loop and given a token cap addresses "inside loop" and "no output cap" in one change.
3. Don't compare costExposureScore across different workflow reports — it's a within-report ranking, not a normalized score. Compare a workflow's Cost Exposure Score band (surfaced in the Executive Summary and detailed in the Cost Risk article) across reports instead.
4. A workflow with no Cost Risk findings produces an empty hotspot list. That's the expected, clean result, not a sign the analysis skipped anything.

Related reading

• Cost Risk — the findings this block groups, and the separate 12-factor Cost Exposure Score
• Executive Summary — where the costExposure band surfaces at the top of the report
• Break Points — the cross-category ranked list these same findings may also appear in