# KimoNav remaining work — 9 September 2026 continuation

Completed: the 24-program live binary implementation check (10/24 versus 8/24
current, two gains/no losses), with declared recovery accounting; and the 18-case
reference-envelope grid, which admits 16 geometric repairs but no full-task
nominal forecast. Both saved-output audits pass. These are development studies.

1. **Repair stopping and timed progress jointly.** Wider bounds advance the
   hold but can lose position. Retain all proposals and test all original task
   constraints; do not promote a candidate solely for low predicted STOP speed.
   A nominally admitted new proposal needs its own registered matched execution.
2. **Validate prediction of repair effects.** Use separate development variation
   to measure false admissions, intervention residuals, gains/losses and quality.
   Separate margin calibration from evaluation. Keep the wrong-sign pelvis result
   and rejected offset model visible; do not tune on final36.
3. **Let simulation continue during computation.** Current binary preparation
   takes 11–18 s paused despite an event one second before STOP. Measure readiness,
   activation age, stale rejection and missed deadlines during repeated feedback.
4. **Freeze before final36.** Fix method, comparator resources, compute budgets
   and analysis before opening the reserved parameter-variation programs. Preserve
   program pairing, dependencies, failures, coverage and quality regressions.
5. **Complete human manuscript review.** Align claims with the actual evidence,
   complete the assistance inventory and verify the submission candidate against
   current official venue rules. No paper submission has occurred here.

Hardware, onboard sensing, obstacles, payloads and broad environmental transfer
are later scopes. They are not implied by the present flat-ground G1 simulation.

[Completed evidence and accounting](/assets/kimonav/continuation-2026-09-09.md).