mule.bot Book the teardown
FM-01 · PUBLIC EDITION
THE OWNER'S SIDE OF AUTONOMY

There are 0 things an autonomous machine can be asked to do on a mine, quarry, or construction site.We wrote them all down.

mule.bot works for the site, not the machine. We measure what autonomous fleets actually produce, find where the lost points go — vendor, site, or process — and sit on your side of the table.

Book the 20-minute teardown   Read the field manual →
6,397
mapped use-case nodes, nine divisions, six levels deep
G0–G7
deployment gates across 8 phases — published in full
$150k
what one burdened operator actually costs per year
60–85%
of manned baseline — where underperforming fleets typically run
§ 01 The instruments
Reference · Free

The Library

The use-case taxonomy, browsable. The full G0–G7 gate stack, printable. Field notes from live deployments. The map, in public.

Open the library →
Instrument · 12 minutes

The Scorecard

Score your fleet against the taxonomy and the gate stack. Get your leak profile — where operations like yours typically bleed.

Run the scorecard →
Study · Annual

The Benchmark

The Annual Autonomous Fleet Performance Benchmark. Ramp curves, availability, cost per ton — vendor-anonymized. Know where you stand.

Join the study →
§ 02 What attribution looks like
SAMPLE
Production-Loss Ledger — page 4 of 21
SITE: ████████ · FLEET: 14 TRUCKS · PERIOD: 28 SHIFTS
NodeLoss mechanismAttributionHrs / periodTons$ / yr run-rate
HAU.02.03.04Spot-at-crusher retry loop, approachVENDOR41.2−9,870−$412,000
HAU.04.01.02Comms-loss fleet stops, ridge segment SITE28.6−6,850−$286,000
LOD.03.02.05Queue imbalance at face, dispatch rulePROCESS22.1−5,300−$221,000
HAU.07.02.01Waterline skipped: trucks in service pre-G4SITE19.8−4,740−$198,000
MNT.03.01.03Road grade out of spec, segment , speed deratePROCESS14.3−3,420−$143,000
Page subtotal — 5 of 31 attributed mechanisms126.0−30,180−$1,260,000

Redacted page from a real readout. Every lost production point gets a taxonomy node ID and an owner. Fog becomes a bill.

We don't work for autonomy vendors.

Not because they're bad at what they do — because nobody can sit on both sides of the same table. We take three engagements at a time, all of them owner-side. Independence isn't a value on this website. It's the product.

The 20-minute teardown.

Bring your fleet; we'll bring the map. Three specific findings about your operation, live on the call — no deck, no pitch. If we can't find anything, that's good news about your fleet.

NO PREPARATION REQUIRED · NOTHING TO INSTALL · NOTHING TO SIGN

Book the teardown →
The Library · FM-01 §L

Give away the map.
Keep the mule.

Our working references, in public. Buyers of judgment aren't blocked by lack of information — they're blocked by capacity, cover, and risk. So the information is free.

§ L1 The Autonomous Site Operations Taxonomy

Six levels: Domain → Division → Use Case → Operation → Maneuver → Action. 6,397 nodes. Below: the nine divisions and sample use cases, free to browse two levels deep.

LEVELS L3–L5 — OPERATIONS, MANEUVERS, VARIANTS, ACCEPTANCE TESTS — ARE APPLIED INSIDE ENGAGEMENTS.
§ L2 The Gate Stack — published in full

Eight phases, in execution order, each with an exit gate. Print it. Put it in your program review. If your deployment can't say which gate it's at, that is the finding.

G0
Go / No-Go
The decision is made on your economics, not the vendor's. An honest baseline exists before anyone signs anything.
G1
Signed Scope
Every use case the site is buying, enumerated, with acceptance criteria written by the buyer.
G2
The Floor
The site is physically and digitally ready: roads, comms, survey, people. Autonomy amplifies site discipline — in both directions.
G3
The Backlog
Site procedures converted into a specified, testable backlog. What "done" means is written down before commissioning begins.
⚓ The Waterline
Per-machine entry into commissioning. Never waived. Every truck that crosses below the waterline unproven surfaces later as chronic underperformance — with interest.
G4
Site Acceptance
The system demonstrates the signed scope on your site, against your acceptance tests — not a reference site's.
G5
Production Rate
The fleet meets the rate the business case was built on. Measured by you, attributed honestly.
G6
Steady State
Exceptions are cataloged, owned, and shrinking. The fleet survives crew rotation, weather, and a bad week.
G7
Self-Sufficiency
The 4 a.m. test: something stops the fleet at 4 a.m. and your people restart it without calling anyone. That's the finish line.
§ L3 Field Notes
FN-011 · AUG 2026

Why fleets stall at 75% — and why it's almost never one thing

The gap is a stack of small, attributable losses hiding under one big excuse. How we decompose it, and the three mechanisms we find at nearly every site.

14 min · READ →
FN-010 · JUL 2026

Envelopes, not bubbles

Range-based make/break safety connections beat static exclusion zones — for throughput and for the safety argument. What we watched it change on a live site.

11 min · READ →
FN-009 · JUL 2026

Commissioning at scale is a different sport

Commissioning one truck is a procedure. Commissioning the fourteenth while ten produce is an operating model. Where the waterline discipline earns its name.

12 min · READ →

The map is free.
Reading your site with it is the work.

Twenty minutes, three findings, live. If we can't find anything, that's good news about your fleet.

Book the teardown →
The Scorecard · Instrument 02

Where does your fleet leak?

Eight questions, scored against the taxonomy and the gate stack. You'll get a leak profile — the areas where fleets like yours typically bleed, and where you stand. This is the short public version; the full 24-question instrument ships with your report.

QUESTION 01 / 08

Loading…

Leak profile

Builds as you answer. Longer bar = more leak.
Self-assessment has a ceiling: it can tell you where you leak, but not whose fault it is. Attribution — vendor, site, or process — takes instruments. That's the teardown.
The Benchmark · Annual Study · Vol. I recruiting now

The Annual Autonomous Fleet Performance Benchmark

Every vendor publishes their best site. Nobody publishes the distribution. We're building the reference dataset for what autonomous fleets actually do — vendor-anonymized, site-anonymized, owner-verified.

§ B1 What we measure
B-01
Ramp curves
Months from first autonomous load to plan rate — the number every business case guesses at.
B-02
Productivity vs. manned baseline
The honest percentage, and how it moves by fleet size and site type.
B-03
Fleet stops & exceptions
Stops per shift, minutes per stop, and who has to touch the machine to clear it.
B-04
Cost per ton, loaded
Licensing, staffing deltas, rework — the full stack, not the brochure math.
B-05
Staffing ratios
Builders, tenders, controllers per truck — before, during, and after ramp.
§ B2 Participate

One hour. Full report, free, first.

Participation is a structured one-hour interview plus a data sheet you approve. Participants get the full report before anyone else, plus their own fleet plotted against the (anonymized) distribution.

✓ Noted. We'll be in touch within two working days.

ANONYMITY IS STRUCTURAL: RESULTS PUBLISH IN BANDS, NEVER SITE-IDENTIFIABLE. YOU APPROVE YOUR DATA SHEET BEFORE ANYTHING LEAVES THE ROOM.

The Firm · FM-01 §F

Two practitioners.
One side of the table.

We've spent years inside live deployments — commissioning weeks, ramp curves, safety cases, the 4 a.m. phone calls. mule.bot exists because everything we learned sat on the wrong side of the table.

§ F1 The practitioners
[ Photo: actual dirt ]

Maghee

Deployment & Performance

Fleet deployments from first truck to steady state: commissioning, ramp recovery, safety-case architecture, and the economics underneath them. Author of the Framework and the Deployment Playbook.

[ Photo: actual dirt ]

Ben Miller

Requirements & Systems

The machine that turns site procedures into testable specifications: use case → actor → epic → story → acceptance test. Author of the AHS Agile Framework and co-author of the Taxonomy.

§ F2 House rules
R-01

Nobody waives the waterline.

Every truck proves itself before it produces. Sites that skip this pay it back later, with interest, as chronic underperformance nobody can explain.

R-02

Attribution before argument.

"Vendor issue" and "site issue" are opinions until the loss has a node ID, a number, and an owner. We don't attend meetings about opinions.

R-03

The buyer defines done.

Acceptance criteria written by the seller are not acceptance criteria. They're marketing with a signature line.

R-04

Our deliverable is our absence.

Every engagement ends with your people passing the 4 a.m. test without us. Consultants who architect their own permanence are a leak too.

1
2
3

We run three engagements at a time. That's the whole firm, and it's on purpose — the deployments that go wrong are the ones where the people who wrote the plan aren't in the room.

We don't work for autonomy vendors.

Not because they're bad at what they do — because nobody can sit on both sides of the same table. Independence isn't a value on this website. It's the product.

Start with twenty minutes.

Bring your fleet; we'll bring the map. Three findings, live. No deck, no pitch.

Book the teardown →