16 entries in 2026.
Nobody marks a Secure Job Envelope traveler "ready." The file earns each level by containing what that level needs, and both implementations compute the same answer. The rulebook — what each level requires, the four guards on quoting and awarding, what an amendment does, why an awarded job locks, and what a level is not.
A fingerprint is only useful if every program computes the same one, and JSON gives you a dozen ways to write the same data. The Secure Job Envelope's canonical form, rule by rule — key order, the number range, cents instead of decimals, strict dates — and the real bug one shared test file caught on its first run.
Earlier posts called it a fingerprint. This one opens the hood — exactly which fields of a job go into the Secure Job Envelope's hash, which are deliberately kept out, why a quote survives an award but not an amendment, and the three things a hash is not.
Our flagship rebuild is built, loaded, and in parallel testing — and still isn't in production. The shop runs on its old system, the new one runs alongside it, and go-live is targeted for November. Here's what parallel testing is for, what it has to prove, and the questions to ask before anyone switches your shop over sooner.
The Secure Job Envelope, walked through one job from the public examples — a CNC bracket composed by a buyer, quoted by one shop, amended, re-quoted by another, awarded, and locked. What each side sees at each step, and why a change as small as the shipping terms sends the first quote back to the start.
Why a job between two shops should be a file both of them can check — not a portal, not a network, not a PDF. Five design decisions behind the Secure Job Envelope, each one a problem we kept seeing on the order desk, and what each one costs.
Every rung of this ladder stops at your wall. The plater, the heat treater, the shop that takes your overflow — the job still leaves as an email, a PDF, and a phone call. The Secure Job Envelope is our answer — an open file format for a job, with the honest list of what it does and doesn't do yet.
Most shops should climb the ladder one rung at a time. Sometimes the floor underneath is failing. What a full rebuild actually looks like from the inside — the data archaeology, the replay of 268,861 historical builds, and a go-live that has to be earned — told from one in progress.
Every rung of this series ends with "and here's the number." This post is about the number — how a sprint's KPI gets picked in discovery, written on the quote, measured with the same instrument before and after, and what happens when it misses. Including what we do with an RFP.
Most shops already have a system of record on the floor — the paper traveler, with its stamps and initials and the job number in Sharpie. Making it digital is not a big floor-software project. It's three questions, one scan, and a screen the floor actually looks at.
Lights-out gets sold as a machine problem. Often the reason a shop can't run an unattended shift isn't the machine — it's that nothing can tell the machine what to do next without a person in the office. Here is the list of handoffs that have to happen without a human, and where that list stops at your wall.
One customer PO, keyed into the quote, keyed again onto the traveler, keyed a third time into the books. A small tool that reads it once and drafts the work order for a human to approve — and why a patch this small has an outsized effect on the whole shop.
The first post in a series on moving more work through a shop — from one station, to the office, to the shops you trade with. It starts with a roll of tape and three numbers, and most of what it finds doesn't need a line of code.
Most bad custom-software stories start with a contract shape, not a bad programmer. How we work — three ways to scope the work, fixed prices, contingency stated instead of hidden — and why two-week fixed sprints are the shape where the buyer and the builder want the same thing.
Four ways to pay for a shop system — rent it, buy it, adopt open source, or build on an open substrate — and the ten-year math behind each. Including the honest reason a from-scratch custom bid loses, told by the people who lost one.
No shame — it worked for twenty years. But the risks have changed, and there are three realistic ways out. Here's how to pick, including the two that don't involve hiring anyone like us.