refactor(pow): drop the redundant pre-mining created_at stamp

The queue re-wrapped the template with a fresh created_at before handing it
to the miner, but the miner now stamps the top of its first pass from the
same clock -- so a job that waited in the queue or was restored from disk
picks up "now" either way. The send-without-pow fallback keeps its own
stamp; nothing mines on that path.

Also records the shipped created_at work in the plan doc, including why the
new parameter goes after isActive (so trailing-lambda call sites keep
binding to it).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017pEArg58zWxDJ7EzzzPdXT

commons/src/commonMain/kotlin/com/vitorpamplona/amethyst/commons/service/pow/PoWPublishQueue.kt

@@ -214,15 +214,10 @@ class PoWPublishQueue(
             difficulty = difficulty,
             persistAs = persistAs,
             owner = pubKey,
-            mine = { isActive ->
-                val toMine =
-                    if (clock != null) {
-                        EventTemplate<T>(clock(), template.kind, template.tags, template.content)
-                    } else {
-                        template
-                    }
-                PoWMiner.mine(toMine, pubKey, difficulty, minerThreads, isActive, clock)
-            },
+            // the miner stamps the first pass from the same clock, so the job
+            // picks up "now" whether it waited in the queue or was restored
+            // from disk — no need to re-wrap the template here.
+            mine = { isActive -> PoWMiner.mine(template, pubKey, difficulty, minerThreads, isActive, clock) },
             publish = onMined,
             // send-now fallback: the same template, minus the nonce the miner would
             // have added. created_at follows the mined path — refreshed to "now" for

quartz/plans/2026-08-13-gpu-pow-mining.md

@@ -197,10 +197,12 @@ is no longer blocking any UI could use more.
 NIP-90 kinds 5970/6970 — hand the template to a DVM with real hardware. That beats
 every on-device option and is already specced.
 
-## Advancing `created_at` while mining
+## Advancing `created_at` while mining — SHIPPED
 
 Not a speedup — a correctness fix that this analysis is a prerequisite for, because
-it interacts with the midstate.
+it interacts with the midstate. Implemented as described below; `PoWMiner.run/mine`
+take an optional `refreshCreatedAt: (() -> Long)?`, covered by
+`quartz/…/PoWMinerCreatedAtTest.kt`.
 
 NIP-13: *"It is recommended to update the `created_at` as well during this process."*
 We do half of it. `PoWPublishQueue.enqueue(refreshCreatedAtOnStart = …)` and the
@@ -224,6 +226,11 @@ frozen behaviour. The consumers are already correct — `PoWNostrSigner` forward
 `mined.createdAt` to `signer.sign(…)`, and the queue publishes the mined template — so
 this is contained inside `PoWMiner` plus one flag at the call sites.
 
+The new parameter goes **last**, after `isActive`, so the trailing-lambda call sites
+keep binding their lambda to `isActive`. Kotlin would otherwise silently retarget them
+at `refreshCreatedAt`; the `Boolean`/`Long` mismatch makes that a compile error rather
+than a bug, but those sites were rewritten to a named `isActive = { … }` anyway.
+
 **"When we can" is already modelled.** Reuse `refreshCreatedAtOnStart`'s predicate
 rather than inventing a second one; its comment already states the exclusion —
 *"Must stay false for scheduled posts, whose future created_at is intentional."*