The nightly pull runs at 12:00 UTC. This is what it added and what dropped off — the same added/fell-off figures the manual upload used to show you, kept per run so nothing happens silently.

Anything the log cannot show — who was really at the wheel and when, a yard move, a later start. A name is enough; the ID is not needed.

Diagnosis Doctor answers why a violation has not fallen off the report. It leaves the Omnitracs report for three reasons only: the record was corrected and the driver accepted it, a recalculation decided it was never real, or it aged out. Anything else and it stays. Fixable is not fixed.

A missing Off Duty entry is a risk, not a violation — most tidy themselves up when the driver goes off duty the next morning. The count that matters is how many actually produced one, which is matched against your open violations, not assumed. Logging out is not going off duty. When the entry is missed the duty period never closes and the next day’s limits are still measured from the last one — which is how a driver who rested all night picks up a 14-hour violation on his first mile. Same driver every week is coaching; same truck across different drivers is a power harness.

Three generations of one-pager are in circulation, so two rows can give you two different documents. This rewrites them all in the current format. It is free — the rules engine writes these sheets and no model is called. It runs in slices to stay inside Cloudflare’s free browser allowance, so run it again until it says nothing is left.

Answers both at once: the last 10-hour reset and what’s left (drive / 14-hr window / weekly), and whether the 16-hour Big Day was used in the past 168 hours — with the date it’s available again.

Measures every duty tour to the hundredth of an hour (13.78h, not “about 14”), grouped by Route ID from dispatch. Routes averaging over 13 hours are manufacturing 14-hour violations — this is where you fix the schedule instead of coaching the driver.

One row per incident, not per re-trigger. Consecutive days of the same driver + type collapse into a single violation (a 70-hour that re-fires all trip counts once). Flags multi-day incidents and likely stuck-in-drive.

Days with a recorded sensor/ELD failure, grouped by vehicle and site — find the trucks that keep breaking. Heavy pull; runs in batches.

Every log edit in the last 7 days across all fleets, split driver vs. manager/office. Heavy pull — runs in batches. After it loads, ask the Assistant “who made the manager edits at [site]?” for the per-person breakdown.

Every edit for that site and period — who edited, the line, when, and whether it was re-certified. Longer periods pull in windows, so 6 months takes a bit.

Reads each driver’s profile from Omnitracs: HOS rule set, Personal Conveyance, Yard Move and its termination settings, Big Day, Adverse Weather, split OFF/SB pairings, status and ELD exemption. Flags drivers missing a permission that would otherwise clear a violation. Read-only — nothing is changed.

The tracker only sees a UVA once it has already become a violation on a named driver — 11 of them. The real queue lives in the Omnitracs host. This asks Omnitracs directly: it tries 14 candidate endpoints and sweeps the driver-log feed that already works, in case unassigned activity is arriving there as a pseudo-driver. Read-only — every call is a GET and nothing can be assigned or rejected.

He has logged out and gone home before the block appears. Kris Moore’s 10 August block landed on a day he had already certified at 07:54 the next morning. Omnitracs’ instructions — annotate the false entry, then hand-calculate available hours to the next reset — are impossible unless someone catches it first. That is this page.

Every driver, every day in the window — not just rows that already carry a violation. A block eats the driver’s available hours whether or not it tipped him over, and the ones that have not yet are the drivers about to be caught by it.

Two rules do the work here. 49 CFR 395.30(c)(2) requires an annotation on every change made to a driver’s record, and 395.32(b) on every unidentified segment a driver assumes. The ELD specification sets a 4-character minimum — “ok” and “x” are not annotations. Text the ELD writes by itself (rule-set changes, inspections, “adjust time”) lands in the same column and is not an annotation; it is shown separately below.

Used as the From on Outlook drafts the Fix Checker builds. No login, no account — just so the draft opens as you. Nothing is emailed from here.

Tick the sites you are responsible for. Reports and lists default to these — the rest stay available but out of the way.

How long a site gets before a violation is overdue, and who to copy when it goes past due.

Which model answers in Assistant and Deep Dive. Fix Check does not use a model at all — its verdicts are deterministic code, so this setting does not affect them.

Claude needs the ANTHROPIC_API_KEY secret. The model comes from the CHAT_MODEL variable — claude-opus-5 for the sharpest answers, claude-sonnet-5 for a cheaper one. Deep Dive can be set separately with DEEP_MODEL.

A snapshot of every violation and all your work. Take one before any big change.

Load the UnassignedVehicleActivity CSV export. The Fix Checker uses it to tell a driver's own driving from motion a truck booked to him because its association never released — the Kris Moore case. Re-loading the same export is safe; rows update in place.

One-time migration from a previous Excel tracker.

Brings your historical open items in, so fall-off detection has something to compare against.