The gap
The software running your column is probably very good at what it was designed to do. The problem is the scope of what it was designed to do. A cryo-EM session is a chain that starts in a dewar and ends in a set of 2D classes you decided to keep — and almost every tool in this field owns exactly one link of it.
The chain
This is the actual path a sample takes. Pick the profile closest to your setup, then correct it stage by stage — whatever stays unmarked is being carried by people, spreadsheets, and the assumption that someone remembers.
Now place your own
Now correct it. Click any stage your current software actually owns.
On a typical vendor console these 5 stages have no system of record. They are being held together by people, spreadsheets, and the assumption that somebody remembers:
Where you are now
None of these are bad software. Each one is very good inside its own boundary — and the boundary is the point. Find yours below, and read the third column as a list of the work your facility is currently absorbing by hand.
Shipped with the column
Turnkey, supported, and genuinely good at the thing it was built for: driving one manufacturer’s microscope through one session. It begins when the grid is already in the autoloader and ends when the movie lands on disk.
What it leaves to you
Total control, one machine
Scripting workstations can offer very deep instrument control. In many facilities, that control lives in macros on one computer, written in a language only a small number of people maintain.
What it leaves to you
Right idea, earlier decade
A real distributed architecture with a real database behind it — the correct instinct, implemented when throughput was orders of magnitude lower and the browser was not a serious option.
What it leaves to you
Held together by someone
Best-of-breed parts, a shared drive, and a set of scripts that move data between them. It works — because a specific person keeps it working. The glue became the product, and nobody was ever assigned to own it.
What it leaves to you
Note Most facilities are running two of these at once, with a shared drive in between. That seam is not a workflow. It is an unbudgeted job.
The cost of the seams
We are not going to quote you a percentage. You already have the numbers — they are in your session logs and in how your week actually went. These are the three that tend to decide whether a facility is running smoothly or quietly re-doing work.
01 / ask yourself
Count them from the dewar to the 2D class averages. Every boundary is a place where identity, metadata, and intent get re-entered by hand — or quietly lost.
02 / ask yourself
If defocus drifted, ice went thick, or the picker was mis-tuned, the gap between the mistake and the evidence is the amount of session you paid for twice.
03 / ask yourself
Not approximately — exactly. Same presets, same filters, same thresholds, same template, same order. If that answer depends on one person’s memory, it is not a protocol yet.
What you’re missing
Not features bolted onto an acquisition tool. Consequences of a single data model that spans sample logistics, the column, the compute, and the facility around them.
A grid arrives in a dewar, sits in a puck, gets loaded, imaged, screened, collected, processed, and classified. Today that journey crosses four or five systems and at least one spreadsheet. It should be one record — shipment and chain of custody through atlas, square, hole, exposure, movie, CTF, particles, and the 2D classes you kept.
Motion, CTF fit confidence, drift rate, ice thickness and contamination are all measurable while the session is still going. If they only reach you as a chart the next morning, they are a post-mortem. They should be a signal the queue can act on — continue, recheck, skip, pause.
A collection strategy is institutional knowledge. Stored as a macro file or a screenshot of a settings dialog, it leaves with its author. Stored as a versioned workflow with typed parameters and a snapshot taken at commit time, it stays with the facility.
Auto-selection is only as trustworthy as its explanation. For any hole on any square, you should be able to ask why it was selected or rejected — which filter, which threshold, which score, whether a human overrode it — months after the session closed.
Instruments, nodes, teams, projects, customers, and bookings are the actual unit of work in a core facility. Software that models one microscope and one operator pushes all of that onto calendars, email threads, and whoever answers first.
Your lab’s reviewed method should reach the operator as a first-class step: declared through a typed extension contract, submitted at session scale, and shown in a view designed for its result — not a script someone runs by hand afterward.
Pixel size, matrix calibration, beam shift, dose, apertures, detector references, preset alignment. Automation inherits all of it. A visible preflight can show whether each item is loaded and current; a trained operator must still decide whether the instrument and plan are safe to run.
The person most likely to be at the console at 3 a.m. is the newest member of the lab. If the tool assumes years of accumulated convention, the facility pays for that assumption in supervision, in re-runs, and in sessions nobody wants to admit to.
What changing actually involves
The realistic objection to any platform is the migration. So the sensible order is to add the missing links first and let the overlap resolve itself once the record is already in one place.
Import an existing session and watch the whole chain assemble itself — atlas, squares, holes, exposures, results — without touching the microscope.
Learn more 02 / stepMotion correction, CTF, picking, and classification at session scale, with live progress and result views. Your acquisition software does not have to change yet.
Learn more 03 / stepCalibration, presets, atlas, targeting, templates, and the run queue — with the target tree already populated by the work you did in the first two steps.
Learn moreGet started
Compare the capability matrix, or open the bundled demo session and follow one grid from import to selected classes.
