generation-budget-guard

Generation Budget Guard

Cost control is a rubric ★ row. Draft cheap, spend once on the final, cap the project, and know before you blow the budget — not after.

Zero baked-in prices. Every unit price is a user input — provider prices are date/plan-volatile and [unverified — confirm on provider pricing page]. Pull the live number from the provider's pricing page and pass it in.

Workflow

1. Set the per-project cap up front (a design input, from the brief or the engagement).
2. Route cheap draft → premium final — iterate on a low-cost draft model to lock composition + brand, then spend on one premium final render. Don't premium-render every iteration.
3. Log each spend line with ../../scripts/gen-budget.py add — model, tier (draft/final), your unit price, count. It computes the line cost; it never invents a price.
4. Check status — gen-budget.py status --budget <cap> (wired into /check-generation-budget) reports spent, remaining, and by-tier, and exits non-zero if over budget (LOUD-FAIL, never a silent pass).
5. Dedup/cache — reuse an approved asset (and prefer an editing round-trip) before generating a near-duplicate.

The cheap-draft-then-final loop

Phase	Model tier	Spend
Explore composition + brand	Draft (cheap)	Many cheap renders
Lock the chosen composition	—	Curation gate
Final render	Premium	One render
Ship resolution	Upscale round-trip	Cheaper than a premium re-roll

Anti-patterns

• Premium-rendering every iteration (draft first).
• No per-project cap (budget as a quarterly surprise).
• Baking a provider price into a tool or a client quote (prices are [unverified]).
• Re-generating a near-duplicate instead of reusing / round-tripping an approved asset.

See also

• Best-practice: ../../best-practices/route-cheap-draft-before-premium-final.md
• ../../scripts/gen-budget.py, ../../commands/check-generation-budget.md