Overnight log 2026-07-23

01:56 wakeup 1

BTC weekly shards: 22/58 complete, newest written 01:54 -> fetch ALIVE, throttled-slow (~9 shards/h). Projection: BTC done ~05:00-06:00, gold after.
Deep gold (2014-2023): not started (correctly sequenced behind BTC).
No unblocked work: scan waits on gold; promotion prereg waits on replication; no unregistered runs permitted.
Earlier tonight: BTCUSD_VANTAGE cost variant + 3 tests (117 green), Helm #34 updated.
Next wakeup 1800s.

Complete shards STILL 22/58 (zero progress in 30 min), but fetcher alive (0-byte attempt written 02:26) -> quota starved, not dead.
Hypothesis: continuous retries keep the rate limiter saturated. Decision: one more cycle of patience; if still 22 at next wakeup (~60 min zero-progress), message agent to enforce a 60-min full pause then resume (leaky-bucket recovery).
Gold: still not started. No other unblocked work.

Still 22/58 complete (~90 min zero progress), attempts continuing -> limiter saturated.
INTERVENED: messaged fetch agent to pause ALL Dukascopy requests 60 min, then resume with gentler pacing (bs1/bp10000 + 60s between shards); extend pause to 2h if still 429ing.
Next check in 30 min (expect pause in effect; judge recovery at the ~04:20 wakeup).

Pause window in effect (expected quiet). Complete shards: 22/58. Newest file: Jul 23 03:09.
Correct behavior; no intervention. Next check post-resume.

Resume attempted 04:15-04:20: one partial (77KB) + one 0-byte, then quiet -> 60-min pause did NOT restore quota; agent correctly on the extended 2h-pause path (next attempt ~06:20).
Still 22/58. Gold not started. Decision point at ~06:30: if still 22, declare Dukascopy hard-blocked for the night, stand agent down till 09:00, mark gold replication as the morning item.

In extended-pause window as expected. Complete: 22/58, newest file: Jul 23 04:20. No intervention.
Next wakeup ~06:45 judges the 06:20 attempt; that decides stand-down-to-09:00.

06:20 attempt failed like all others (one 77,903-byte partial + 429s — identical partial size twice suggests a fixed per-burst request allowance). Dukascopy declared HARD-BLOCKED overnight (~01:30-06:45 despite 60-min and 2h cool-downs).
Fetch agent stood down until 09:00; then gentle resume (bs1/bp10000/60s, escalate to 1-shard-per-20-min if 429s persist). Gold queued behind BTC.

BTC bid/ask: 22/58 weekly shards complete (covers chunks of Jan-Jul 2026 partially; cost-model spread series still incomplete).
Deep XAUUSD 2014-2023: NOT STARTED (blocked behind BTC) -> displacement replication (Helm #40) is the top blocked item; scan config unchanged, ready to run the moment data lands.
Promotion prereg: NOT drafted (correct — replication pending). Validation looks untouched: 10/10.
Shipped tonight: BTCUSD_VANTAGE cost variant (+frozen-hash guard), 117 tests green, Helm #34 updated.
Ready to send: phase-0-gates/Mentor Questions - July 2026.docx (validated).
Alternative worth considering in the morning: Dukascopy's quota may be account/IP-day-bound — options: different network, or accept BTC spread series as partial and calibrate cost model from the 22 complete weeks (~5 months coverage, likely sufficient for a flat-spread instrument).