Smoke runs, not benchmark results. These trials check the pipeline. They carry no ranking and support no comparison between configs or harnesses.

Smoke runs

Each run checked the path from a trial to the analysis summary to this page: Harbor ran each trial in a container, the injector logged its decisions, and the analysis read the job directories and wrote the summary shown here. Newest first.

track-n-codex

10 trials, 8 faults applied, between 5 and 26 eligible tool calls per trial. Source: results/smoke/2026-10-30-track-n-codex/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean580.0%$1.5661$0.31320
moderate580.0%$1.2472$0.24948

Per trial

TaskProfileRewardCostEligible callsCompletion status
n-csv-exportn-csv-export__bu9hjdmclean0.00$0.16585success
n-ledger-fixn-ledger-fix__3BDKBe2clean1.00$0.475826success
n-make-buildn-make-build__Ybu7YaJclean1.00$0.353121success
n-port-conflictn-port-conflict__gAcy3NUclean1.00$0.164910success
n-timesheet-payrolln-timesheet-payroll__HHt4Phwclean1.00$0.406418success
n-csv-exportn-csv-export__HdJ5dq8moderate1.00$0.14965success
n-ledger-fixn-ledger-fix__weAPYw7moderate1.00$0.416422success
n-make-buildn-make-build__xJdNmaumoderate0.00$0.351116success
n-port-conflictn-port-conflict__eVQ9HKmmoderate1.00$0.148110success
n-timesheet-payrolln-timesheet-payroll__BvgTCRNmoderate1.00$0.18215success

track-s-codex

10 trials, 25 faults applied, between 10 and 52 eligible tool calls per trial. Source: results/smoke/2026-10-30-track-s-codex/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean5100.0%$0.4994$0.09990
moderate580.0%$0.5663$0.113325

Per trial

TaskProfileRewardCostEligible callsCompletion status
s-cycle-counts-cycle-count__BPTfHhAclean1.00$0.154638success
s-error-catalogs-error-catalog__oRs7sLBclean1.00$0.065610success
s-followupss-followups__tKXfc55clean1.00$0.066712success
s-p1-reports-p1-report__om9Ln4Jclean1.00$0.076622success
s-release-trains-release-train__SWkqNw9clean1.00$0.135945success
s-cycle-counts-cycle-count__AK9JidEmoderate1.00$0.161438success
s-error-catalogs-error-catalog__rjSNh6Gmoderate0.00$0.069111success
s-followupss-followups__fKtDd28moderate1.00$0.082317success
s-p1-reports-p1-report__4CZwPQQmoderate1.00$0.094829success
s-release-trains-release-train__dUeTvqpmoderate1.00$0.158752success

track-n-fxh

10 trials, 22 faults applied, between 9 and 71 eligible tool calls per trial. Source: results/smoke/2026-10-21-track-n-fxh/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean5100.0%$2.2090$0.44180
moderate560.0%$2.6543$0.530922

Per trial

TaskProfileRewardCostEligible callsCompletion status
n-csv-exportn-csv-export__zhCUBgkclean1.00$0.339814success
n-ledger-fixn-ledger-fix__LaXE99iclean1.00$0.782358success
n-make-buildn-make-build__8yQCvdaclean1.00$0.660426success
n-port-conflictn-port-conflict__XXE8vVZclean1.00$0.169811success
n-timesheet-payrolln-timesheet-payroll__a6FChJyclean1.00$0.25669success
n-csv-exportn-csv-export__Hy66mQWmoderate0.00$0.393922success
n-ledger-fixn-ledger-fix__AE7Sbtsmoderate0.00$1.183171success
n-make-buildn-make-build__Wg2z7WGmoderate1.00$0.639125success
n-port-conflictn-port-conflict__KQaHb2xmoderate1.00$0.167216success
n-timesheet-payrolln-timesheet-payroll__m6BDUEwmoderate1.00$0.271014success

track-n-pi-oc

20 trials, 35 faults applied, between 4 and 37 eligible tool calls per trial. Source: results/smoke/2026-10-19-track-n-pi-oc/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean10100.0%$2.4158$0.24160
moderate1060.0%$2.7866$0.278735

Per trial

TaskProfileRewardCostEligible callsCompletion status
n-csv-exportn-csv-export__pMHrSY4clean1.00$0.13765success
n-csv-exportn-csv-export__z5wbZGRclean1.00$0.14135success
n-ledger-fixn-ledger-fix__6gebiCnclean1.00$0.438823success
n-ledger-fixn-ledger-fix__6uvzMhMclean1.00$0.381122success
n-make-buildn-make-build__N9Srnjoclean1.00$0.383721success
n-make-buildn-make-build__UN7B9y3clean1.00$0.334515success
n-port-conflictn-port-conflict__Adwvcoyclean1.00$0.163811success
n-port-conflictn-port-conflict__he3Hnzyclean1.00$0.124012success
n-timesheet-payrolln-timesheet-payroll__Kv9DYWjclean1.00$0.17275success
n-timesheet-payrolln-timesheet-payroll__eMZ4vCvclean1.00$0.13834success
n-csv-exportn-csv-export__Aim58SSmoderate1.00$0.12155success
n-csv-exportn-csv-export__uCKk7YKmoderate1.00$0.19898success
n-ledger-fixn-ledger-fix__5SwRJVEmoderate0.00$0.461631success
n-ledger-fixn-ledger-fix__E2Phx2emoderate0.00$0.732737partial
n-make-buildn-make-build__UCdo8DPmoderate0.00$0.355121success
n-make-buildn-make-build__zDyr6Pumoderate0.00$0.350016success
n-port-conflictn-port-conflict__FZNrLPjmoderate1.00$0.135012success
n-port-conflictn-port-conflict__FpTfUzVmoderate1.00$0.121818success
n-timesheet-payrolln-timesheet-payroll__6z4okBtmoderate1.00$0.12064success
n-timesheet-payrolln-timesheet-payroll__P3xZYjcmoderate1.00$0.18935success

track-s-fx

10 trials, 22 faults applied, between 14 and 61 eligible tool calls per trial. Source: results/smoke/2026-10-19-track-s-fx/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean5100.0%$0.8972$0.17940
moderate560.0%$0.6270$0.125422

Per trial

TaskProfileRewardCostEligible callsCompletion status
s-cycle-counts-cycle-count__LWBJxqBclean1.00$0.223338success
s-error-catalogs-error-catalog__REyXweaclean1.00$0.263861success
s-followupss-followups__nnRmGNNclean1.00$0.118015success
s-p1-reports-p1-report__DbuXPpdclean1.00$0.096422success
s-release-trains-release-train__Husknftclean1.00$0.195645success
s-cycle-counts-cycle-count__HNRBpn4moderate1.00$0.194046success
s-error-catalogs-error-catalog__sHpNZXtmoderate1.00$0.104414success
s-followupss-followups__3bGRK8Dmoderate0.00$0.068417partial
s-p1-reports-p1-report__dNXbZsZmoderate0.00$0.076526success
s-release-trains-release-train__dpvszP9moderate1.00$0.183749success

track-s-pi-oc

20 trials, 41 faults applied, between 9 and 69 eligible tool calls per trial. Source: results/smoke/2026-10-18-track-s-pi-oc/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean10100.0%$0.8282$0.08280
moderate1070.0%$0.8180$0.081841

Per trial

TaskProfileRewardCostEligible callsCompletion status
s-cycle-counts-cycle-count__WnfNWtsclean1.00$0.152337success
s-cycle-counts-cycle-count__hmcRWwGclean1.00$0.136138success
s-error-catalogs-error-catalog__7zdBkPcclean1.00$0.04129success
s-error-catalogs-error-catalog__8kbXaXLclean1.00$0.062410success
s-followupss-followups__8nLRDe3clean1.00$0.042812success
s-followupss-followups__CwgvvQUclean1.00$0.064112success
s-p1-reports-p1-report__9GgKphtclean1.00$0.045022success
s-p1-reports-p1-report__PFKT7v3clean1.00$0.044722success
s-release-trains-release-train__DXJAjTFclean1.00$0.112145success
s-release-trains-release-train__jB5ZJFFclean1.00$0.127645success
s-cycle-counts-cycle-count__HJCKEi9moderate1.00$0.144642success
s-cycle-counts-cycle-count__YCVDEwvmoderate0.00$0.146969success
s-error-catalogs-error-catalog__5BSjevcmoderate1.00$0.043713success
s-error-catalogs-error-catalog__VuHXiEdmoderate1.00$0.042313success
s-followupss-followups__Tt52ciSmoderate1.00$0.045116success
s-followupss-followups__Ux5PXvzmoderate1.00$0.049317success
s-p1-reports-p1-report__EeWH3vGmoderate0.00$0.058826success
s-p1-reports-p1-report__MPiKsMVmoderate0.00$0.048124success
s-release-trains-release-train__EPv9myQmoderate1.00$0.127250success
s-release-trains-release-train__PfXCWPBmoderate1.00$0.112049success

track-s

12 trials, 32 faults applied, between 12 and 57 eligible tool calls per trial. Track S, fx-claude and cc-claude on three service tasks, one attempt per task and profile (D-0030). The tables pool both configs. One attempt per cell supports no comparison between them. The cost columns undercount fx-claude: some of its trials have no cost in result.json because the gateway settled them later; the settled costs are in settled-usage.json next to the summary. Source: results/smoke/2026-10-15-track-s/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean6100.0%$0.4988$0.09980
moderate6100.0%$0.4545$0.113632

Per trial

TaskProfileRewardCostEligible callsCompletion status
s-duplicate-merges-duplicate-merge__2HXwTSkclean1.00$0.116351success
s-duplicate-merges-duplicate-merge__HPFQagHclean1.00-51success
s-followupss-followups__cPgaDXAclean1.00$0.117115success
s-followupss-followups__jcXcKHAclean1.00$0.070312success
s-p1-reports-p1-report__QeAsNYZclean1.00$0.125022success
s-p1-reports-p1-report__o6FV6Bzclean1.00$0.070022success
s-duplicate-merges-duplicate-merge__J2Nuj3emoderate1.00$0.130257success
s-duplicate-merges-duplicate-merge__yXP3Tammoderate1.00$0.167757success
s-followupss-followups__G7CFa9xmoderate1.00-21success
s-followupss-followups__hkc4tfBmoderate1.00$0.078317success
s-p1-reports-p1-report__R7UjSspmoderate1.00$0.078426partial
s-p1-reports-p1-report__sXP77Wnmoderate1.00-26partial

6 trials, no fault applied, between 2 and 4 eligible tool calls per trial. Track N, cc-claude only. The tasks of this run were later rebuilt under D-0023, D-0024 and D-0026, so its trials sit below the exposure floor of 8 in SPEC 10.3 and the moderate profile had few eligible calls to fault. Source: results/smoke/2026-10-11/summary.json.

Per profile

ProfileTrialspass^1Total costMean costFaults applied
clean3100.0%$0.1146$0.03820
moderate3100.0%$0.0778$0.02590

Per trial

TaskProfileRewardCostEligible callsCompletion status
n-ledger-fixn-ledger-fix__EGtrgcbclean1.00$0.04024success
n-port-conflictn-port-conflict__DuaYAQ4clean1.00$0.03453success
n-shipments-summaryn-shipments-summary__CdnkvnHclean1.00$0.03983success
n-ledger-fixn-ledger-fix__3rTZbnfmoderate1.00$0.03164success
n-port-conflictn-port-conflict__tyGHefFmoderate1.00$0.02033success
n-shipments-summaryn-shipments-summary__k2REVrcmoderate1.00$0.02592success