kairn

Evidence details

The receipts behind the public numbers.

This page keeps the benchmark split visible: clean pass/pass savings, fresh checks, endurance and MCP runs, official evaluator rows, and results that are useful but not counted as headline savings.

How to read this evidence

Clean savings

Counted only when both baseline and Kairn pass the quality gate.

Quality/source focus

Reported separately when Kairn passes and baseline misses quality.

MCP evidence

Requires actual MCP calls plus useful source guidance, not just a configured server.

Fresh repeat-1 checks

Useful new signal, but repeat-3 is needed before it becomes a headline benchmark.

Kairn is not measured as a basic code RAG layer. RAG finds relevant candidates. Kairn is measured on what should actually reach the agent during the task, and how much context is worth spending.

Clean pass/pass savings

Rows where both baseline and Kairn completed the task successfully, so token reduction can be counted directly.

WorkflowTypeBaselineKairnReductionQualityStatusNotes
Click enduranceLong session
repeat-3
90,73361,52842.7%pass/passclean savings10-turn debugging session with zero scope violations.
Node SemVer enduranceLong session
repeat-3
69,51358,14216.4%pass/passclean savingsFresh JavaScript package endurance row outside the earlier Python-heavy set.
OpenLibrary SWE-ProOfficial evaluator
official row
21,93210,04854.2%pass/passclean official savingsOfficial evaluator pass/pass row; counted separately from quality-rescue rows.

Latest fresh checks

Newer post-launch checks that are useful signal, but not promoted to headline claims until repeated.

WorkflowTypeBaselineKairnReductionQualityStatusNotes
Fresh source-routing checkSource routing
repeat-1 fresh check
122,98075,33938.7%3/3 pass/passfresh repeat-1Active-valid assistance with scope discipline passing. Needs repeat-3 before it becomes a headline benchmark.
Fresh SWE-Pro governance sliceOfficial evaluator
fresh official slice
4/8 known local passes9/9 valid official passes9.3% clean median10/10 local passfresh officialPost-governor branch check. Strongest signal is source/scope discipline; one row excluded for evaluator image availability.

Endurance and MCP evidence

Long-session and MCP rows where Kairn had to stay useful beyond the first turn or through actual MCP calls.

WorkflowTypeBaselineKairnReductionQualityStatusNotes
httpcore MCP enduranceMCP endurance
repeat-3
recordedrecorded18.5% median100%clean MCPActual MCP calls, useful returned files, and zero scope violations.
Requests MCP post-fixMCP endurance
repeat-3
recordedrecorded28.1% median3/3 passclean MCPUseful MCP file returns in all three runs; zero scope violations.
Click MCP-firstMCP endurance
repeat-3
recordedrecorded52.6% medianpass/passclean MCPMCP-first variant of the Click endurance suite.

Official evaluator evidence

Rows run through the official evaluator path. Clean savings are counted only when both sides pass.

WorkflowTypeBaselineKairnReductionQualityStatusNotes
SWE-Pro official campaignsOfficial evaluator
25 official rows
15/25 passed20/25 passed6 clean rowsofficial evaluatorquality/source-focus + clean savingsClean rows saved 63,393 observed tokens with a 22.1% median reduction; rescue rows are reported separately.
SWE-Pro new expansionOfficial evaluator
20 new rows
13/20 passed15/20 passed5 clean rowsofficial evaluatorfresh official expansionNew expansion rows saved 51,509 observed tokens across clean pass/pass savings rows.
SWE-Pro governance slice clean rowsOfficial evaluator
3 clean pass/pass rows
174,620154,62711.4%official pass/passclean official savingsLatest slice clean rows had positive savings; source/scope discipline was the stronger result.

What does not count as clean savings

These rows are useful for product learning, but they are not used as headline savings claims.

WorkflowTypeBaselineKairnReductionQualityStatusNotes
Fresh live canaryLive canary
repeat-1
missed qualitypassedsupporting onlyquality rescuenot clean savingsUseful because Kairn improved the outcome, but not a pass/pass savings claim.
SWE-Pro token regressionsOfficial evaluator
tracked regressions
official passofficial pass4 regressionspass/passnot clean savingsSome official pass/pass rows used more tokens with Kairn; these are tracked as optimization targets.
Suppressed or passive rowsGovernor behavior
varies
variessilent or tinysupporting onlyvariesnot active savingsCorrect silence is product evidence, but it is reported separately from active token wins.

Claims to keep straight

SafeKairn can reduce token use on source-rescue, debug, MCP, and endurance workflows.
SafeKairn may stay silent when confidence is low; that is intentional suppression.
SafeCodex CLI/session is the most-tested path; MCP is the portable editor path.
CarefulSavings vary by task and model; official SWE-Pro rows separate clean savings from quality/source-focus evidence.
AvoidDo not claim universal 20-70% savings or public works-anywhere reliability yet.
Back to overviewInstall Kairn