Section 6 · Recorded runs
Oracle trajectory against self-reported position
Every figure on this page is read from run manifests written by the harness. Pose series are downsampled for display. The controller under test is simple_navigator. Runs come from the Breakbench Rust simulator, not Gazebo. Nav2 is not involved.
80 of 80 grid runs · 151 manifests · nominal n = 25, spread converged
6.1 · Selected runs
Requested odometry scale above 1. The navigator declares the goal reached in its own odometry frame while the true pose stops short of the goal region. No safety-invariant violation on this run.
20260902T190001Z-2ccef6c133 · odom_axis · seed 125
Oracle poses from the plant ground-truth stream; belief from the delivered odometry record.
- Self-report success
- true
- Oracle mission failure
- true
- Safety violation
- false
- Verdict uses self-report
- false
- True terminal x
- 6.544 m
- Believed terminal x
- 7.011 m
- Min clearance
- 0.1384 m
- Oracle samples
- 4501
- Requested latency
- 0.00 s
- Delivered latency mean
- 0.000 s (n=451)
- Requested odom bias
- 0.08
- Delivered scale
- 1.080 (n=4501)
self-report reason: odom_goal_reached · simple_navigator_odom_frame
6.2 · Self-report against the oracle
Two confusion matrices, kept separate
Self-report vs safety invariant
Oracle safe means no safety-invariant violation. Mission failure without a safety violation is not folded into this matrix.
| Cell | Count | Wilson |
|---|---|---|
| success+safe | 149 | 149/149 p̂=1.00 [0.97, 1.00] |
| success+violation | 0 | 0/149 p̂=0.00 [0.00, 0.03] |
| failure+safe | 0 | 0/149 p̂=0.00 [0.00, 0.03] |
| failure+violation | 0 | 0/149 p̂=0.00 [0.00, 0.03] |
Self-report vs mission completion
Oracle complete means the true pose entered the goal region before timeout. This matrix is not the safety-invariant confusion matrix.
| Cell | Count | Wilson |
|---|---|---|
| success+complete | 74 | 74/149 p̂=0.50 [0.42, 0.58] |
| success+mission_failure | 75 | 75/149 p̂=0.50 [0.42, 0.58] |
| failure+complete | 0 | 0/149 p̂=0.00 [0.00, 0.03] |
| failure+mission_failure | 0 | 0/149 p̂=0.00 [0.00, 0.03] |
6.3 · Single-axis estimates
Safety-invariant violation was 0 of 5 at every sampled severity; mission failure was not
Latency, safety violation
n = 5 per cell. Interval is Wilson, z = 1.96.
| delay s | k / n | p̂ [low, high] |
|---|---|---|
| 0 | 0/5 | 0.00 [0.00, 0.43] |
| 0.15 | 0/5 | 0.00 [0.00, 0.43] |
| 0.4 | 0/5 | 0.00 [0.00, 0.43] |
| 0.8 | 0/5 | 0.00 [0.00, 0.43] |
Odometry bias, safety violation
n = 5 per cell. Interval is Wilson, z = 1.96.
| bias | k / n | p̂ [low, high] |
|---|---|---|
| 0 | 0/5 | 0.00 [0.00, 0.43] |
| 0.08 | 0/5 | 0.00 [0.00, 0.43] |
| 0.2 | 0/5 | 0.00 [0.00, 0.43] |
| 0.4 | 0/5 | 0.00 [0.00, 0.43] |
Latency, mission failure
n = 5 per cell. Interval is Wilson, z = 1.96.
| delay s | k / n | p̂ [low, high] |
|---|---|---|
| 0 | 0/5 | 0.00 [0.00, 0.43] |
| 0.15 | 0/5 | 0.00 [0.00, 0.43] |
| 0.4 | 0/5 | 0.00 [0.00, 0.43] |
| 0.8 | 0/5 | 0.00 [0.00, 0.43] |
Odometry bias, mission failure
n = 5 per cell. Interval is Wilson, z = 1.96.
| bias | k / n | p̂ [low, high] |
|---|---|---|
| 0 | 0/5 | 0.00 [0.00, 0.43] |
| 0.08 | 5/5 | 1.00 [0.57, 1.00] |
| 0.2 | 5/5 | 1.00 [0.57, 1.00] |
| 0.4 | 5/5 | 1.00 [0.57, 1.00] |
6.4 · Two-axis grid, mission failure
Mission failures k of n per latency and bias cell
| lat \ bias | 0 | 0.08 | 0.2 | 0.4 |
|---|---|---|---|---|
| 0 s | 0/5 | 5/5 | 5/5 | 5/5 |
| 0.15 s | 0/5 | 5/5 | 5/5 | 5/5 |
| 0.4 s | 0/5 | 5/5 | 5/5 | 5/5 |
| 0.8 s | 0/5 | 5/5 | 5/5 | 5/5 |
Safety-invariant k was 0 in every recorded grid cell. q for a scalar margin is undeclared, so this is not a sub-margin joint-violation claim. Scalar m_i(q) is not computed.