> Historical report. The [September 9 review](progress-2026-09-09.md) and [current evidence](progress-2026-09-09.json) supersede its status and next-step statements. The first-candidate repeat arms are now complete; the active reference-repair study is unfinished. # KimoNav: completed braking and reserved confirmation Updated 7 September 2026. Ongoing research, not a published controller result. KimoNav asks whether timed metric whole-body programs can be executed reliably by a frozen motion generator and tracker. Kimodo provides controllable offline motion generation; the current execution loop uses ARDY for autoregressive G1 motion and SONIC for physical tracking. Their integration is inherited capability. Our study concerns the timed request contract, causal reference updates and measured failures. ## Completed experiments The braking sweep compares a native 6/24 success baseline with 0.2, 0.4 and 0.6 s terminal translation leads: 7/24, 7/24 and 6/24. None adds the registered two successes. No lead is promoted. Only 20 changed references require new physics; 52 full-denominator slots reuse exact native references. Reference-dependent initialization is a disclosed protocol deviation: reset procedure and seed match, but regenerated first poses differ, limiting same-state causal attribution. Reserved confirmation finishes 88 captures: 84 scored and four exact null controls. Native/K1/minimum-jerk succeeds on 2/12, 3/12 and 4/12 at simulator seed 1701. Native/minimum-jerk succeeds on 2/12 and 2/12 at seed 1702, and on 1/12 and 3/12 at a second repeat seed 1703 registered before its outcomes (26 captures, no falls or early endings; two paired gains, no losses, 3 sliding regressions above 1.1× native). Repeat candidates are regenerated from that seed's own measured histories. All selected arc outcomes fail; successes occur in reverse-pair programs. These are twelve related program designs, not 36 independent samples from repeating three seeds. Mean sliding improves with minimum jerk in both seeds. Individual sliding regressions against native remain on four primary and three repeat programs. Primary K1 has two early endings, retained as failures; its STOP diagnostics have ten available programs. No operational fall is recorded. The hindsight oracle's 8/12 and uniform-four's expected 3.5/12 are diagnostic recorded-bank comparisons. The result does not establish a consistent task-success improvement. All source locks and process exits verify, together with 300 independent prefix array checks and 132 full null-control array checks. No outcome-based rerun or threshold change was used. The first supplementary reporter failed on unavailable STOP speeds; a preserved, versioned reporting fix adds availability counts without altering registered scores or physical outcomes. Across R-A/R-B, 108 new physical captures were collected, including four null controls. ## What this changes The supported paper is a diagnostic requested/generated/executed study. A reference that passes task checks need not be dynamically feasible, and lower generated jerk does not guarantee original-clock success or per-program quality preservation. The twelve reserved programs have now been opened; any later method informed by them needs a new protected confirmation set. The native SONIC planner is the next practical comparison gap. Its pinned model passes six CPU interface probes, but the 30 Hz planner clock, planned-context update and explicit idle STOP still need a qualified benchmark bridge. A zero walking-speed override is not a stop instruction. No planner physics is claimed. Tracker adaptation and retention remain untested: G1-only loss/export, fixed sampling and vectorized parity are not fully qualified. No new ranker, predictor search or gradient training is launched from these confirmation outcomes. Live deadlines, repeated feedback, hardware, torque/ZMP and broad generalization remain outside the measured claims. The local manuscript is a draft, not a conference submission. ## Evidence - [Complete confirmation and paired results](confirmation-evidence.json) - [Success and sliding figure](confirmation-success-and-sliding.svg) - [Braking decision and caveat](braking-evidence.json) - [Recorded physical-state video](timing-failure.mp4) - [Video provenance](timing-failure-provenance.json) Earlier development and reconstruction results remain available from the project page. All public measurements are curated from source-bound local records.