---
title: Planner decisions and the pricing catalog
description: Read the A to F graded capacity planner decision log and the live GPU pricing and availability catalog, in the Operations Console and with argusctl.
---

# Planner decisions and the pricing catalog



ARGUS records every launch decision the capacity planner makes, grades it A to F, and keeps a live catalog of GPU pricing and availability underneath it. Together they answer two questions: "why did the planner pick that provider?" and "what does GPU capacity cost right now?"

## The planner decision log

Every planner launch decision is persisted with a grade and rationale. Each row shows:

* time and decision key
* chosen provider
* grade (A to F)
* which gates passed or failed (compliance, capacity, budget)
* cost delta versus the cheapest alternative
* reason text

A decision graded D or F tells you exactly which gate dragged it down. Typical uses: post-incident audit ("why did the planner pick RunPod over AWS at 2:14 UTC?"), finding systematic bias in the grading rubric, and reviewing compliance-class routing (HIPAA workloads must not land on RunPod).

**Console:** the planner decisions page (route `#/decisions`) lists the decision rows. Backing endpoints: `GET /compute/planner/decisions` and `GET /compute/planner/summary`.

**CLI:**

```bash
argusctl compute planner          # recent A-F allocation decisions + rationale
```

## The GPU pricing catalog

The catalog is the live market reference: GPU pricing and availability pulled from multiple provider feeds. Three providers feed it:

* `aws_pricing_api` (on-demand pricing)
* `aws_spot_history` (spot rates)
* `runpod_gql` (RunPod)

For each provider the console shows row counts, freshness, a fallback flag, and the per-SKU cheapest rate.

**Console:** the PRICING REFERENCE section at the bottom of the COMPUTE page (route `#/compute`). It is reference data, not running pods. A stale provider here usually means the catalog refresher worker is stuck. Backing endpoints: `GET /compute/catalog/freshness`, `/compute/catalog/providers`, `/compute/catalog/skus`.

**CLI:**

```bash
argusctl compute catalog --provider aws --rate-kind spot --limit 10
```

`--rate-kind` selects the rate class (for example `spot`); `--provider` filters by provider.

## Strain: utilization and headroom

`argusctl compute strain` is the derived diagnostic that joins workers with their load: per-worker utilization percent (current cameras over capacity), headroom, over- or under-provisioned flags, idle burn, and recommendations ("needs a pod", "idle pod, stop it"). It supports `--watch` for a live-refreshing view.

```bash
argusctl compute strain
argusctl compute strain --watch 10
```

## Typical workflow

1. Compare provider prices in the catalog before launching a SKU.
2. [Launch the pod](/docs/compute/launch).
3. Later, check the planner decision log to confirm allocation decisions were graded well, and `compute strain` to confirm capacity matches demand.

Command reference: [argusctl compute](/docs/cli/compute).


---

For a semantic overview of all documentation, see [/sitemap.md](/sitemap.md)

For an index of all available documentation, see [/llms.txt](/llms.txt)

For agent-facing discovery, including API and MCP surfaces, see [/agents.md](/agents.md)