Every result, with its method and its limits.
Two evaluations are published. Both concern LatentGraph, the repository-context component. Both were scored by the party named in the method line.
No result on this page measures whole-factory productivity. Each result measures one component, on the task set and repositories named in its method line.
Better context. A measurable difference.
LatentGraph · repository-context evaluation · June 3, 2026Claude Code tasks resolved+2 tasks
10 / 14from 8 / 14
Without LatentGraph8 / 14
With LatentGraph10 / 14
API cost per resolved task66% lower
$1.33from $3.93
Without LatentGraph$3.93
With LatentGraph$1.33
What was measured
Task resolution and API cost per resolved task, with and without repository context supplied to the same coding agent.
On what
14 selected hard tasks across two repositories. API cost includes failed attempts.
By whom, when
Company-reported evaluation, run by LatentForce, June 3, 2026.
What it does not prove
It does not measure whole-factory productivity, and it does not measure any component other than LatentGraph.
Built around real codebases.
Healthtech customer · POC record · July 2026PR tasks completed+6 tasks
8 / 10from 2 / 10
Without LatentGraph2 / 10
With LatentGraph8 / 10
What was measured
PR tasks completed by the customer’s own engineers, scored by the customer.
On what
Four repositories totaling approximately 17 million lines of code. Code volume describes evaluation scope, not a result.
By whom, when
Customer-scored POC record, July 2026. Original score sheets have not been independently verified.
What it does not prove
It does not measure Workspace or the complete stack, and the organisation is not named.
Components with no published result
LatentGateway, SkillPatch, Workspace, managed agents and trace analytics have no published evaluation. Where there is no measured result, this site publishes no number.
LatentGatewaySkillPatchWorkspaceManaged agentsTrace analytics