Grex experiments
Grex runs AI assignments across computers and keeps their results and review evidence. Here we document experiments on how that work gets done: the hypothesis, the setup, the result and the decision it changed.
Each record includes a method diagram, a short companion video and a downloadable result summary. Unfavorable results belong here too.
Showing 13 records across all themes. A record may cover more than one theme.
Read the experiment · Watch the explanationGrex completed 8 of 30 tasks strictly correctly, on time and without assistance, compared with 13 in the previous day's pilot. Duplicate simulated effects fell from 34 to 15 and token-bound breaches fell to zero. Correctness did not improve, and no correction rescued a failed task.
Grex 2.0
Completed · study closed
Read the experiment · Watch the explanationBoth arms finished 13 of 30 tasks strictly correctly, on time and without assistance. Grex took about 2.1 times the total elapsed time. The later audit found that corrections never received reviewer feedback, so this pilot cannot establish whether a functioning review-and-correction loop helps.
Grex 2.0
Completed · diagnostic evidence
Read the experiment · Watch the explanationThe controlled site correctly reported four new faults, retained their identities and later reported their resolution. But the overflow stage produced no accepted report in three attempts, so the controlled proof failed. All three production windows were eventually accepted; only two arrived within the planned two-hour window.
Grex 2.0 · October–November 2026
Closed · controlled proof failed
Read the experiment · Watch the explanationChanging the resale platform left item and batch profit based on the old platform’s fee. Grex produced a two-line product repair and six focused regressions across three executions. Independent checks confirmed the released behavior. This is one assisted, verified repair, not proof of autonomous discovery or an entirely green application test suite.
Grex 2.0 · October–November 2026
Completed · one repair verified
Read the experiment · Watch the explanationAcross six filing families and three slots per arm, full history produced 17 accepted slots out of 18; trimmed history produced 13 out of 18. The trimmed arm accumulated six deadline-exhausted attempts and two outputs with critical regressions. The policy was not adopted as the default.
Grex 2.0 · October–November 2026
Completed · policy not adopted
Read the experiment · Watch the explanationGrex passed six of ten document families; the direct workflow passed five. Both matched the identity, date, number and name fields, with remaining acceptance failures concentrated in amendment interpretation. Grex used about 121 aggregate worker minutes versus about 16 for the baseline. This small, model-confounded comparison supports a near-tie, not superiority.
Grex 2.0 · October–November 2026
Completed · bounded comparison
Read the experiment · Watch the explanationThe content change shipped on October 30. As of November 4, the experiment has a recorded baseline but no outcome decision. Deployment is verified; search growth has not been established.
Grex 2.0 · October–November 2026
Observation open · decision scheduled for November 27, 2026
Read the experiment · Watch the explanationThe unchanged instructions produced an accepted brief. Ordinary retries produced an accepted brief on attempt two. The frozen refined candidate passed mechanical checks but failed independent review with two unsupported claims. This was one source-window group, not a success-rate study.
Grex 2.0
Completed · adoption on hold
Read the experiment · Watch the explanationAcross three rounds, nine worker executions produced no reviewable brief. Two further launches failed before worker execution, and four planned slots were not launched. No independent research review or correction ran.
Grex 2.0 · October 2026
Closed · quality target missed; no briefs reviewed
Read the experiment · Watch the explanationWorker-only and planner-plus-worker submissions passed the same selected cases on all three tasks. Planning added about nine seconds per pair. All six model reviews accepted, but a separate source-level probe found an untested fallback bug in both diagnostic submissions. That task’s broader correctness remained uncertain.
Grex 2.0 · October 2026
Completed · diagnostic case uncertain
Read the experiment · Watch the explanationAcross five pairs, direct execution and the earlier Grex chain each passed 17 of 41 judged rubric items. Individual pairs differed, and the judging had known limitations. The decision was to simplify single-task execution while keeping mission-level controls outside the experiment’s scope.
Historical Grex 1 · October 2026
Completed · historical architecture
Read the experiment · Watch the explanationThe archived runs produced task results and several actionable failures. GAIA and retail-task samples completed more fully than some tool and terminal tasks. Different model mixes, harness revisions and samples prevent a platform-lift claim. These are historical attempts from the architecture that preceded Grex 2.
Historical Grex 1 · August 2026
Historical · mixed completion
Read the experiment · Watch the explanationThe retained evidence covers Grex-side calibration, not the planned seven-pair comparison. Repeated searches rediscovered unsuitable posts, and one published internal brief contained refusal text as its suggested reply. A separate reviewer then judged that brief not worth reading. No Grex-versus-Hermes precision result was established.
Historical Grex 1 · August 2026
Calibration only · comparison held
