RIPPLE REFERENCE
touch of grey
Scorecard · Category 1 of 5

Everything You Need to Know About Safety

Six numbers: the five Netradyne driving rates, and the FICO score that sits with them. They are the hardest exclusions on the sheet — no research step, no judgement call, no appeal path built into the rule. Over the line is over the line.

1 Damage 2 Terminated 3 Attendance 4 Safety 5 FICO 6 POD 7 CED 8 DC DPMO 9 CDF 10 DSB

The five safety rates

All five come off the weekly Amazon Scorecard, one column each, one row per driver. They are rates, not counts — events per hour of driving — so a short week and a long week are compared on the same footing. Amazon sources them from Netradyne, the in-cab camera system.

RateWhat trips itThresholdResult
Seatbelt Off RateDriving with the belt unbuckled. Includes the short hop between stops — the camera does not care that it was twenty feet.> 0.4Ineligible
Speeding Event RateOver the posted limit by Netradyne's margin, on its read of the road's speed limit.> 0.4Ineligible
Distractions RateEyes or hands off the task — phone in hand, looking away from the windshield, reaching across the cab.> 0.4Ineligible
Following Distance RateToo close to the vehicle ahead for the speed being driven.> 0.4Ineligible
Sign/Signal Violations RateRolled stop signs, red lights, missed signals.> 0.4Ineligible
Any one of the five, not the average A driver at 0.0 on four rates and 0.45 on the fifth is Ineligible. There is no averaging, no trade-off, and no "but everything else was clean." The rule reads each of the five independently and the first one over the line ends it.

The threshold is 0.4, and it is strictly greater-than: a rate of exactly 0.4 passes. This matters more often than it sounds like it should, because Netradyne's rates land on round numbers constantly.

FICO

Not the credit score. Amazon's FICO here is a composite safe-driving score on a 1–850 scale, built from the same Netradyne telematics that feed the five rates above — so it is not independent evidence, it is the same evidence scored a second way. It became a real exclusion on 2026-09-20, on Dan's call, and sits immediately next to the safety rates because it is a safety number: "just like the .4 safety rule."

ScoreMeansResult
800–850Passing band.Passes this rung
1–799A real score, below the band.Ineligible
blank / "No Data" / 0No score yet — not a bad score.Rule does not fire
Blank is not zero Treating an empty FICO cell as 0 would cut an entire roster the first week the rule ran. The column is still arriving unevenly from Amazon — 0 of 108 drivers populated on W35 DIN6, 4 of 105 on W36, 15 of 17 on DII5 W36, then back to 0 of 110 and 0 of 17 on W37. The rule is live and correct; Amazon simply is not filling the column in consistently yet.

Adding FICO changed no past verdict — nobody in the entire archive scores in the 1–799 band. The regression test asserts exactly that, so the rung is proven inert on history rather than assumed to be.

Why these rungs sit where they do

Safety is rung 4 and FICO rung 5 of ten. Everything above them — Damage, Terminated, Attendance — comes out of Jake's tracker rather than the Scorecard, so it is checked first. Everything below is either a delivery-quality number or something that needs research.

Because the ladder stops at the first match, the order among rungs 1–8 only decides which reason gets printed for a driver who breaches several. It never decides who gets paid. Safety was moved to print above POD on 2026-09-20 for exactly that reason: when a driver breaks both, "Safety" is the more useful thing to see on the sheet than "POD".

Where the numbers come from, and where they go

StageWhat happens
SourceAmazon's weekly DSP Scorecard PDF, one row per driver. Read straight off the PDF with pdftotext -layout — no hand-typing since 2026-09-20.
UpstreamNetradyne (idms.netradyne.com) is where individual events can be reviewed and, where wrong, disputed.
Rulerules.py — constants SAFETY_MAX and FICO_MIN_ELIGIBLE. One copy, both stations.
OutputThe Final_Cut tab of the biweekly workbook, which feeds payroll directly.

Dan confirmed on 2026-08-25 that the Scorecard itself — not GeoTab, not a separate telematics export — is the source of truth for the safety thresholds. That is why no second system needs to agree before a safety exclusion stands.

One ladder, both stations

DIN6 and DII5 run the identical rules off one file. They did not always: until 2026-09-20 each station had its own copy, 0.4 was written down twice, and the copies had already drifted — DII5's was written before Attendance became a real exclusion and never gained that rung, so a DII5 driver with an attendance occurrence was paid a bonus a DIN6 driver with the same occurrence lost. Unifying them was proven behaviour-neutral across 447 driver-weeks with 0 verdict changes.

Per-metric tips

The other four categories

Source of truth Thresholds from rules.py (SAFETY_MAX, FICO_MIN_ELIGIBLE). Narrative from references/ripple-bonus-rules-at-a-glance.md and references/ripple-bonus-correction-vocabulary.md. Last checked 2026-09-20.
Powered by dkdcreations