Blackridge answers the questions provider invoices can't — which workflow created the spend, who owns it, what changed, where money was wasted — and backs every answer with graded, reproducible evidence.
This workflow billed the same work repeatedly across failed attempts — $126.97 realized waste, already incurred.
9 evidence rowsretry groups linked, not inferred from timingreplay availablewatch the evidence →
01
The problem
AI spend moves for reasons invoices can't explain. A retry loop bills the same work four times. A fallback quietly routes traffic to a model that costs 6× more. An agent explores twelve branches and keeps one. The invoice shows one number, went up. Finance asks why. Engineering has no evidence — only dashboards.
Why did the bill increase?Named cause, not a trend line.
Which workflow created the spend?Across providers — not per API key.
Which team owns it?Attribution with provenance, not guesses.
What changed, and when?Windows compared, deltas evidenced.
Where was money wasted?Realized waste, never mixed with modeled savings.
Can Finance trust the number?Can Engineering reproduce the investigation? Same evidence, both answers.
02
The evidence
Every claim carries an evidence grade — OBSERVEDDERIVEDINFERREDSIMULATED — plus coverage rates, versioned pricing, and a reproducibility hash. Unknown is not zero. If a claim can't be proven, it says so.
Findings are claims: impact, basis, evidence grade, and a four-layer drill-down that ends at raw records.Investigate a spend spike: zoom the window, follow the workflow, replay the execution.
03
The product
Workflow reconstruction — which billable calls belonged to the run, and which contributed to the output.The request ledger: every captured call with cost, route, cache, attribution, and confidence.
Under the hood: a token economics control point for live model traffic and a GraphQL investigation API over captured evidence. Point clients at the control point for canonical economic evidence; the investigation API powers the assessment UI, workflow replay, request ledger, and exports.
The gateway component is the sensor, not the product. The product is what happens after: deterministic economics derivation, workflow reconstruction, graded evidence, and findings you can defend in a budget meeting. Northwatch Systems builds AI infrastructure products; Blackridge is the flagship — AI runtime economics and economic forensics.
