9. Hypotheses¶
Rules say what was observed. Hypotheses ask what could explain it. AnaTele ranks the hypotheses only for episodes that end in a critical event (an autopilot crash report). In the five simulated flights that's only the motor-failure flight.
The four built-in hypotheses¶
| Hypothesis | Supported by (family, weight) | Counted against it |
|---|---|---|
| Mechanical / thrust or frame configuration | tracking-loss (control, 1) · motor floor saturation (propulsion, 1) · motor imbalance (propulsion, 0.75) · motor ceiling saturation (propulsion, 0.5) | – |
| Estimator / magnetic inconsistency | compass event before crash check (estimator, 1) · EKF compass innovation (magnetic, 1) · compasses inconsistent reported (magnetic, 0.25) | compass event flagged as coming after the upset (1) |
| Vibration-driven estimation degradation | persistent vibration (inertial, 1) · clipping (inertial, 1) | vibration checked and fine (0.5) · clipping checked and fine (0.5) |
| Electrical / power degradation | no support rules yet in this pack version | blocked if power-monitoring-gap fires |
The score¶
The family cap of 1.5 matters. Three rules that all measure motor commands are not three independent pieces of evidence. Capping each family stops one kind of evidence from dominating. Evidence from different families (say, control and propulsion) does add up.
| Label | When |
|---|---|
| strong support | score ≥ 2 |
| some support | score ≥ 1 |
| weak support | score > 0 |
| mixed support | there is both support and contradiction |
| limited or contradicting evidence | score ≤ 0 |
| unassessable | none of its rules could be checked (or a blocking rule fired) |
Worked example: the motor-failure flight¶
- Thrust/configuration: control family = 1 (tracking-loss). Propulsion family = 0.75 + 0.5 = 1.25 (imbalance + ceiling), under the cap. Score = 1 + 1.25 = 2.25, so strong support.
- Vibration: nothing triggered, and both vibration rules were checked and fine, giving −0.5 − 0.5 = −1.0: limited or contradicting evidence. Checking and finding nothing counts against an explanation.
- Magnetic: score 0. The post-crash "Compasses inconsistent" message doesn't count, because it isn't from the flight's own time window.
- Power: unassessable. The current pack has no rule that can support it, so it's listed as a reminder of what the log can't tell you. Confirming it needs battery voltage/current logs.
Toggle the evidence yourself:
This is a screening order, not a probability
A score of 2.25 doesn't mean "90 % likely". The weights are written-down expert judgements, and the list tells you what to inspect first. Each hypothesis also lists the discriminating evidence that would confirm or rule it out. For thrust problems that's ESC RPM/current telemetry, a props-off motor test and a physical inspection.