Document accessibility remediation & NIMAS conversion
Machines included. We take the document work that has to be right and can't be done by hand at volume — and hand every judgment call back to a person.
Your files never leave your building
Accessibility remediation · NIMAS · conversion
The five above, in detail — four systems and the shop they run in. Everything here exists so that a person who couldn't read the book, can.
Perceivable · operable · understandable
Reads a .docx and finds what fails: untagged headings, images with no alt text, tables with no header row, contrast under AA. Proposes each fix with the reason. You approve the ones you want, and it writes a conformance report that states its own limitations instead of hiding them.
Point it at a folder. It sorts every file into already fine, a machine can fix this, and a person has to look at this — with the evidence behind each call and an hours estimate for the last pile. Most shops find two thirds of the pile never needed a human.
A copyeditor's pass that never touches the file. Every suggestion comes back as a tracked change with a note saying why, bound to the exact run of text it came from. If the text moves before you approve, the suggestion expires rather than landing in the wrong place. Read it left to right; every mark is a unit of meaning.
a11y · one source, every reader
A student who reads by touch, by ear, or at 24 point needs the same book as everyone else, on the same day, on the same page. Mate takes curriculum source — InDesign, Word, legacy XML — and assembles the NIMAS fileset those formats are made from: real structure, images with alt text worth reading, MathML where there's math, and print pagination preserved so a class can work together. It validates before it ships and hands you the report, so a rejection is something you find here rather than downstream.
Under every system is one orchestration layer of our own — OpenClaw. It reads the job and routes each step to whatever handles it best: a frontier model for judgment, a local model for bulk, plain code for the parts that don't need a model at all.
You don't choose. You don't tune. You don't get a bill that swings when a vendor changes their pricing. The models are ours to worry about; the outcome is yours to keep.
We don't put specific vendor and version names on the page, because they change and the system doesn't. There's no secret in it: ask and we'll tell you exactly what ran, on what, and when. Every run is logged, and the log is yours.
Audit · propose · approve · apply
The system reads your files and says what's wrong. It changes nothing yet.
Each fix comes back as a suggestion, with the reason and the exact spot it touches.
You keep, skip, or edit each one. Skipping is a normal answer, not a failure.
Approved changes are written, re-checked, and logged. The original is never overwritten.
Nothing we build applies a change you haven't seen. That isn't a setting. That's the whole of it.
These systems don't replace your editors. They hand your editors a shorter, better-evidenced list — so the judgment stays with the people who were always doing the judging.
WCAG 2.1 AA · ADA Title II · Section 508
Send us a representative slice of the backlog. You get findings, severity, and an honest estimate of the work — including the parts a machine can't do. No fixes, no obligation, no call required. Most engagements start here, and some end here.
We quote a system off your audit, not off a rate card. The number reflects the risk we take off your desk. If the audit says the pile is easier than you feared, the quote says so too.
The systems on your own hardware, your own model keys, your own network rules. Includes the specs, the test corpus, and every limitation we know about. Licensed per production team.
Tell us what's on your desk. We'll tell you what it takes.
info@finelinesediting.comWhat we don't sell: a certificate of compliance, a legal opinion, or interactive form remediation. Anyone selling you the first two by the file is selling you a feeling.