🤖 fix: budget accounting ignores OpenAI service-tier (priority) pricing

[ThomasK33]

Summary

Cost accounting and budget enforcement price every OpenAI request at the model's base rates, regardless of the effective service tier. createDisplayUsage(usage, model, providerMetadata, metadataModelOverride) has no tier input and models-extra.ts carries base-tier rates only, so a goal run with --service-tier priority (a higher-cost tier that the app labels as such) and --budget is admitted and metered as if it ran on the default tier. The represented dollar cap therefore under-counts real spend on priority runs.

Scope

• Applies to every model that supports OpenAI service tiers (GPT-5.x, GPT-6 Astra/Sol/Luna), on direct OpenAI and the Coder gateway OpenAI path. It is not specific to any single model's pricing entry.
• Surfaced by Codex security review on 🤖 fix: clamp GPT-6 Sol/Luna headless tool loops to reasoning none and correct their Codex context cap #4259 (🤖 fix: clamp GPT-6 Sol/Luna headless tool loops to reasoning none and correct their Codex context cap #4259 (comment)); deferred there because the fix needs a new accounting input (effective tier) rather than repairing a defect that PR introduced.

Proposed fix (either)

1. Carry the effective tier into usage accounting: extend createDisplayUsage (or its caller) with the pinned serviceTier and apply per-tier rate multipliers/rates from the catalog.
2. Or fail closed: reject budgeted runs when the effective tier is non-default until tier pricing exists.

Follow-up trigger

Next change to usage accounting or the OpenAI pricing catalog; owner: whoever next touches displayUsage.ts / models-extra.ts.

---

Generated with xum • Model: coder:anthropic/claude-fable-5-1 • Thinking: xhigh • Cost: $174.60

[linear-code]

MUX-167