release: 0.7.4 — correct the 0.7.2 n-bound claim (docs only)

CHANGELOG.md

@@ -1,5 +1,30 @@
 # Changelog
 
+## 0.7.4 — 2026-07-16
+
+Documentation only — runtime code is byte-identical to 0.7.3. Released so the
+corrected record ships with the package rather than sitting unreleased on main.
+
+### Fixed
+
+- **Corrected a false claim in the 0.7.2 changelog.** It said an out-of-range
+  `n` "took payment, then 400'd at the provider and lost the prepaid USDC".
+  BlockRun's gateways settle **on success** — `settlePaymentWithRetry` runs
+  after the upstream call — so a request that dies at the provider settles
+  nothing and costs the caller nothing. Verified on every path, including the
+  async one: the poll endpoint answers `payment_status: "not_charged"` with
+  "Upstream generation failed. No payment was taken." for failed jobs.
+
+  The `n` bound stays — it saves a pointless signed round-trip, a facilitator
+  verify, and an upstream call, and answers 400 locally — it was just never
+  about losing money. A changelog that cries wolf about funds is worse than one
+  that says nothing.
+
+  The mistake came from the SDK's `"API error after payment"` wording, which
+  reads as *funds gone* on failures that took nothing; it has now misled two
+  reviewers and this changelog. Fixed upstream in blockrun-llm#25, which reports
+  whether settlement actually happened.
+
 ## 0.7.3 — 2026-07-16
 
 ### Fixed

blockrun_litellm/__init__.py

@@ -26,4 +26,4 @@
 from blockrun_litellm.provider import BlockRunLLM, register
 
 __all__ = ["BlockRunLLM", "register", "enable_local_logging"]
-__version__ = "0.7.3"
+__version__ = "0.7.4"

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "blockrun-litellm"
-version = "0.7.3"
+version = "0.7.4"
 description = "LiteLLM adapter for BlockRun — call x402-paid AI models via LiteLLM (custom provider or local OpenAI-compatible proxy)"
 readme = "README.md"
 license = "MIT"