bug: TOOL_COSTS credit values don't match canonical spec

[stackbilt-admin]

Problem

src/cost-attribution.ts defines credit costs as:

// image_generate
baseCost: 5
// with quality multipliers:
//   draft   × 1 = 5 cr
//   premium × 3 = 15 cr
//   ultra   × 5 = 25 cr

The canonical credit spec defines:

Quality tier	Credits
draft	1 cr
standard	2 cr
premium	4 cr
ultra	8 cr
ultra_plus	10 cr

The canonical spec is the source of truth. TOOL_COSTS is approximately 5× over the spec for every tier.

Impact

• At current TOOL_COSTS values, a 500-credit pack ($9.99) buys ~100 draft calls instead of ~500
• The pricing math on the purchase flow and the actual deduction are misaligned
• Cost audit (scheduled 3-5 days post-deploy) will catch per-unit economics, but this needs to be corrected before that audit is meaningful

Fix

Align TOOL_COSTS to the canonical table. The quality multiplier approach can stay — just rebased from the correct baseCost:

// draft=1, standard=2, premium=4, ultra=8, ultra_plus=10
image_generate: {
  baseCost: 1,
  qualityMultipliers: { draft: 1, standard: 2, premium: 4, ultra: 8, ultra_plus: 10 },
}

Or enumerate each tier directly. Either way, the canonical spec wins.

Related

• Canonical credit cost spec
• edge-auth #150 — credit pack implementation
• bug: credit idempotency key uses crypto.randomUUID() — double-charge on retry #57 — must also fix idempotency key before credits go live

[stackbilt-admin]

Edited: removed internal tool references from issue body per disclosure policy scrub pass.