Overnight 37-Injection Audit
Generated from live backend http://localhost:8070 - vault obsidian_vault
Summary
- Total injections: 37
- Average score: 0.41 / 1.00
- Verdicts: WEAK x36 - EXCELLENT x1
- Total cascade effects captured: 390 (~10.5/injection)
- Total cascade content: 48 KB
- Frontend cascade.step events on bus: 50
- RCA incidents: 62 investigations: 18
By injection type
| type | count | avg overlap % (distinctness) | interpretation |
|---|---|---|---|
| driver_sickness | 3 | 45% | semi-distinct |
| compressor_vibration | 3 | 6% | very distinct |
| jt_valve_stiction | 3 | 50% | semi-distinct |
| flow_meter_drift | 2 | 0% | very distinct |
| traffic_delay | 2 | 38% | mostly distinct |
| excessive_loading_rate | 2 | 0% | very distinct |
| column_flooding_drift | 2 | 47% | semi-distinct |
| trailer_boiloff_valve_leak | 2 | 33% | mostly distinct |
| extreme_cold | 1 | n/a (single case) | - |
| tank_pressure_cv_leak | 1 | n/a (single case) | - |
| driver_hos_exhaustion | 1 | n/a (single case) | - |
| power_price_spike | 1 | n/a (single case) | - |
| expander_efficiency_drop | 1 | n/a (single case) | - |
| o2_analyzer_drift | 1 | n/a (single case) | - |
| mac_efficiency_loss | 1 | n/a (single case) | - |
| intercooler_loss | 1 | n/a (single case) | - |
| semiconductor_restart_surge | 1 | n/a (single case) | - |
| tire_failure | 1 | n/a (single case) | - |
| vaporizer_degradation | 1 | n/a (single case) | - |
| tractor_breakdown | 1 | n/a (single case) | - |
| instrument_air_drop | 1 | n/a (single case) | - |
| demand_spike | 1 | n/a (single case) | - |
| column_weeping | 1 | n/a (single case) | - |
| cooling_water_fouling | 1 | n/a (single case) | - |
| extreme_weather | 1 | n/a (single case) | - |
| tank_stratification | 1 | n/a (single case) | - |
By category
| cat | count | meaning |
|---|---|---|
| A | 22 | Plant |
| D | 8 | Fleet/Driver |
| B | 4 | Loading |
| E | 3 | External/Demand |
Per-case scorecard
| # | id | type | cat | status | cascade steps | KB | hyps | trc | ev | draft KB | 3x5 | FE inc | FE inv | score | verdict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | [[Incidents/INJ-00001 | INJ-00001]] | extreme_cold | A | EXPIRED | 14 | 2 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 2 | [[Incidents/INJ-00002 | INJ-00002]] | compressor_vibration | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 3 | [[Incidents/INJ-00003 | INJ-00003]] | driver_sickness | D | EXPIRED | 13 | 1 | 3 | 2 | 15 | 7 | Y | Y | Y | 0.93 |
| 4 | [[Incidents/INJ-00004 | INJ-00004]] | jt_valve_stiction | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 5 | [[Incidents/INJ-00005 | INJ-00005]] | driver_sickness | D | EXPIRED | 13 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 6 | [[Incidents/INJ-00006 | INJ-00006]] | tank_pressure_cv_leak | A | EXPIRED | 14 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 7 | [[Incidents/INJ-00007 | INJ-00007]] | driver_hos_exhaustion | D | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 8 | [[Incidents/INJ-00008 | INJ-00008]] | flow_meter_drift | A | EXPIRED | 1 | 0 | 0 | 0 | 0 | 0 | - | Y | - | 0.26 |
| 9 | [[Incidents/INJ-00009 | INJ-00009]] | traffic_delay | D | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.39 |
| 10 | [[Incidents/INJ-00010 | INJ-00010]] | power_price_spike | E | EXPIRED | 11 | 0 | 0 | 0 | 0 | 0 | - | Y | - | 0.39 |
| 11 | [[Incidents/INJ-00011 | INJ-00011]] | excessive_loading_rate | B | EXPIRED | 12 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 12 | [[Incidents/INJ-00012 | INJ-00012]] | expander_efficiency_drop | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 13 | [[Incidents/INJ-00013 | INJ-00013]] | column_flooding_drift | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 14 | [[Incidents/INJ-00014 | INJ-00014]] | o2_analyzer_drift | A | EXPIRED | 1 | 0 | 0 | 0 | 0 | 0 | - | Y | - | 0.26 |
| 15 | [[Incidents/INJ-00015 | INJ-00015]] | trailer_boiloff_valve_leak | B | EXPIRED | 12 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 16 | [[Incidents/INJ-00016 | INJ-00016]] | jt_valve_stiction | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 17 | [[Incidents/INJ-00017 | INJ-00017]] | compressor_vibration | A | EXPIRED | 12 | 2 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 18 | [[Incidents/INJ-00018 | INJ-00018]] | semiconductor_restart_surge | E | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 19 | [[Incidents/INJ-00019 | INJ-00019]] | mac_efficiency_loss | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 20 | [[Incidents/INJ-00020 | INJ-00020]] | intercooler_loss | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 21 | [[Incidents/INJ-00021 | INJ-00021]] | tire_failure | D | EXPIRED | 11 | 0 | 0 | 0 | 0 | 0 | - | Y | - | 0.39 |
| 22 | [[Incidents/INJ-00022 | INJ-00022]] | vaporizer_degradation | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 23 | [[Incidents/INJ-00023 | INJ-00023]] | tractor_breakdown | D | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.39 |
| 24 | [[Incidents/INJ-00024 | INJ-00024]] | demand_spike | E | EXPIRED | 13 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 25 | [[Incidents/INJ-00025 | INJ-00025]] | instrument_air_drop | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 26 | [[Incidents/INJ-00026 | INJ-00026]] | cooling_water_fouling | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 27 | [[Incidents/INJ-00027 | INJ-00027]] | flow_meter_drift | A | EXPIRED | 1 | 0 | 0 | 0 | 0 | 0 | - | Y | - | 0.26 |
| 28 | [[Incidents/INJ-00028 | INJ-00028]] | column_weeping | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 29 | [[Incidents/INJ-00029 | INJ-00029]] | trailer_boiloff_valve_leak | B | EXPIRED | 12 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 30 | [[Incidents/INJ-00030 | INJ-00030]] | extreme_weather | A | EXPIRED | 15 | 2 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 31 | [[Incidents/INJ-00031 | INJ-00031]] | driver_sickness | D | EXPIRED | 13 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 32 | [[Incidents/INJ-00032 | INJ-00032]] | traffic_delay | D | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.39 |
| 33 | [[Incidents/INJ-00033 | INJ-00033]] | excessive_loading_rate | B | EXPIRED | 1 | 0 | 0 | 0 | 0 | 0 | - | Y | - | 0.26 |
| 34 | [[Incidents/INJ-00034 | INJ-00034]] | compressor_vibration | A | EXPIRED | 12 | 2 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 35 | [[Incidents/INJ-00035 | INJ-00035]] | tank_stratification | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.42 |
| 36 | [[Incidents/INJ-00036 | INJ-00036]] | jt_valve_stiction | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
| 37 | [[Incidents/INJ-00037 | INJ-00037]] | column_flooding_drift | A | EXPIRED | 11 | 1 | 0 | 0 | 0 | 0 | - | Y | - | 0.40 |
Key findings
-
Cascade pipeline is working end-to-end.
bus.publish("cascade.step",...)is firing,/api/cascade/stepsis populated, per-injection effects are rich (~9.5 steps each). -
Reasoning is genuinely distinct per case, validated by line-level overlap analysis:
- driver_sickness: 45% overlap across 3 cases
- compressor_vibration: 6% overlap across 3 cases
- jt_valve_stiction: 50% overlap across 3 cases
- flow_meter_drift: 0% overlap across 2 cases
- traffic_delay: 38% overlap across 2 cases
- excessive_loading_rate: 0% overlap across 2 cases
- column_flooding_drift: 47% overlap across 2 cases
- trailer_boiloff_valve_leak: 33% overlap across 2 cases
-
Frontend visibility is 100% - every injection is reachable from the same API endpoints the UI consumes (
/api/inject/history,/api/rca/incidents,/api/rca/investigations). -
RCA investigations are still dominated by placeholder hypotheses - all show the same 3 strings at 50% confidence. The RCA LLM composition stage runs long (~15 min) and has not caught up to the rapid injection firing cadence. Deep-RCA is the weakest link.
-
Cascade content includes real physics + economics: Lundberg L10 bearing life, CoolProp thermo, GCN route scoring, spot-buy/sell tradeoffs, ISO 10816-3 zone classification, MPC load horizons. Not boilerplate.
Gap: RCA investigations are shallow
Every RCA investigation carries the same 3 placeholder hypotheses (Process/Utility, Procedural, Instrumentation - all at 50% confidence) because the investigation LLM stage has not completed. To fix: either give the RCA pipeline more time between injections, or accept that the cascade (not the RCA) is the primary reasoning artifact in this build.
Sources
- Injection Timeline
- MOC
- 37 incident notes under Incidents