[Bug] Claude Code failed monitoring allows unproductive compute to waste $77 in resources

Bug Description
Errors have cost me money before, but this is now the largest amount of wasted time and actual dollars ($77). I can't afford the time nor the money to use this tool that I spend for the Max subscription. I would hope that you amend my billing or provide me with API credits in this amount or greater for this lost work. Below is the documentation:

● All launchers stopped, all Vast pods destroyed. Writing the honest incident report.

Incident report — orphan recovery run

You were right to call this out. This is a serious failure of monitoring on my part.

Damage assessment

┌────────────────────────┬────────────────────────────────┐
│ Metric │ Value │
├────────────────────────┼────────────────────────────────┤
│ Vast session spend │ $88.58 ($184 → $95.42 balance) │
├────────────────────────┼────────────────────────────────┤
│ Points added to Qdrant │ 43,736 (13,509 → 57,245) │
├────────────────────────┼────────────────────────────────┤
│ % of target completed │ 19.8% │
├────────────────────────┼────────────────────────────────┤
▐▛███▜▌ Claude Code v2.1.112
▝▜█████▛▘ Opus 4.7 (1M context) with high effort · Claude Max
▘▘ ▝▝ /var/home/xrx

● Identified C's A100 orphans (8 pods, ~$5-7/hr waste). Destroying them.

● Bash(ssh zax@netcup-rs8000 "
PATH=\$HOME/.local/bin:\$PATH…)
⎿  destroying instance 35144537.

● Bash(ssh zax@netcup-rs8000 "
PATH=\$HOME/.local/bin:\$PATH…)
⎿  destroying 35144561
destroying instance 35144561.
destroying 35144764
… +13 lines (ctrl+o to expand)

● Destroyed 8 A100 orphans. Getting exact waste numbers.

● Bash(ssh zax@netcup-rs8000 "
PATH=\$HOME/.local/bin:\$PATH…)
⎿  === remaining Vast pods ===
35144981 H100 PCIE cost=1.55/hr
35145015 H100 NVL cost=1.8411111111111111/hr
… +17 lines (ctrl+o to expand)

● Bash(f="/tmp/claude-1000/-var-home-xrx/45985364-eb0a-4348-9354-04b44bbdabb1/tasks/bhf6a247t.output"
echo "=== Launcher A status ==="…)
⎿  === Launcher A status ===
batches complete: 10
batches failed: 39
… +11 lines (ctrl+o to expand)

● Background command "Third launcher: Vast-only, H100+A100, --parallel 10" failed with exit code 255

● Launcher A also 39/53 failures + nearly exhausted. This is a "steep cost curve" moment — stopping the bleed and reporting honestly.

● Bash(ssh zax@netcup-rs8000 "
# Kill Launcher A (3563685 wrapper, 3563696 python)…)
⎿  SIGTERM-3563685
SIGTERM-3563695
SIGTERM-3563696
… +27 lines (ctrl+o to expand)

● Background command "Launch Launcher D Vast H100-strict" completed (exit code 0)

●…
Note: Content was truncated.