Opportunities and right-sizing

Opportunities is the work queue: what to act on, in what order, with the evidence to justify it.

Three lanes, not one backlog #

Findings are separated by whether they are actually ready:

Lane	Meaning
Actionable now	Resource-level, evidence complete, safe to action
Portfolio review	Real but needs a decision above resource level — commitments, subscription-wide changes
Evidence needed	Insufficient telemetry coverage or observation window

A single undifferentiated list is how backlogs die. Separating them means the actionable queue stays short enough to actually work through.

Where findings come from #

Vendor recommendations are collected and normalized, so they sit alongside everything else rather than in a separate portal.

Flux Signals are Flux’s own deterministic rules — versioned, tested, and explainable — covering idle and oversized compute, unattached disks, orphaned resources, stopped-but-billing machines, aged resources, hybrid-benefit eligibility, and retiring services.

Where both flag the same resource, the finding is marked corroborated, which raises its confidence.

How right-sizing evidence is built #

Right-sizing uses reconciled utilization from multiple telemetry sources, with each metric keeping its source, aggregation method, observation window, and lineage.

Three rules govern it:

Missing telemetry is never idle. A resource with no data is reported as not covered, never as unused. This is the single most important rule in the product — the failure mode it prevents is deleting something that was busy but unmonitored.

Disagreement goes to review. When two sources materially disagree about the same resource, the finding is routed for review rather than resolved by picking the more convenient number.

Coverage is stated. Every finding carries what percentage of the window was actually observed, so thin evidence is visible rather than implied.

Valuation #

Findings are valued two ways:

• Gross monthly savings — the full modelled saving
• Risk-adjusted savings — discounted by the confidence in the finding

Risk-adjusted is the number to plan with. It reflects that a low-confidence finding on thin evidence is not worth the same as a corroborated one on a full observation window.

Valuation records its basis: which cost figures were used, which price source, which method version, and when it was computed.

Confidence #

Each finding carries a confidence score built from how long it has persisted, how many consecutive observations support it, whether it recurred after a previous remediation, whether another source corroborates it, and how fresh the underlying evidence is.

A finding that appears once and vanishes is noise. One that persists across weeks of observations is real.

Evidence packs #

Any finding exports with its inventory context, cost, telemetry, confidence, freshness, uncertainty, and draft implementation, validation, and rollback notes — enough to attach to a change request without rewriting it.

Tracking outcomes #

Set a finding to accepted or implemented and Flux measures what it actually saved. See Measuring realized savings.