DEBT BONIXER · ALL SILO DEBT UNIFIED · γ₁=14.134725141734693

DEBT BONIXER · TWO DEBT TYPES UNIFIED

Two debt types in the fleet: Tech Debt (sorries, open TRBs, undeployed LABRs, broken KCFs) and Cloud Cost Debt (daily Azure spend, GPU pools left running, over-provisioned nodes). The Debt Bonixer surfaces both in one view. CFO lens + CTO lens together. GravDebt engine drives the cloud cost gravity model. γ₁ floor = cost floor.

APRIL COST · CA$7,332 AT DAY 19

Tracking ~CA$11,600 vs CA$10,552 March baseline. Day 95 = May 9. Check: cloud-cost-report.sh daily 8pm EDT. Scale-down script fixed (while IFS read -r bug). GPU pools: all 0 as of last check.

H100: ~CA$15/hr/node · T4: ~CA$1.50/hr/node · NEVER auto-scale — alert only

GPU POOL WATCH · COST CRITICAL

Any NC* node pool count > 0 with no active GPU workload = alert Kay immediately. Do NOT auto-scale down. krs-portal + local-llm Pending (node mem 100%, scaled down nightly — expected).

Rule: GPU pool >0 + no workload = alert · Kay decides

SILO	OPEN TRBs	BROKEN KCFs	UNDEPLOYED LABRs	DEBT LEVEL	DEBT HELIX
msi01	TRB-OSS-JOHN-001	None	None	LOW	/debt-helix-msi01
msclo	TRB-MSCLO-DNS-001	DNS (FIX)	None	MED	/debt-helix-msclo
yone	None	None	None	LOW	/debt-helix-yone
forge	None	None	V3 portal (queued)	LOW	/debt-helix-forge
pcdev	TRB-PCDEV-LCM-001	LCM not deployed	LCM-RUST-001	MED	/debt-helix-pcdev
lilo	—	NAS mount pending	V3 portal (queued)	MED (new)	—

GRAVDEBT: Cloud cost has gravity — the longer a node runs at high cost, the harder it is to stop (team reliance, cached state, contract locks). GravDebt models this as a gravitational field. Nodes with high gravity = high switching cost. V3 portal surfaces GravDebt per AKS cluster as a heat map. Current: aks-eose-aaas-dev (Canada East) is the primary cost driver.