Strictly confidential. Everything you have sent is treated as confidential under the agreement I signed and returned to you on 14 August 2026, and I am bound by it from my side whether or not it is ever countersigned. This page is unlisted and excluded from search engines, it names no author and carries no title of yours, and nothing links to it. Nothing you drop into it is uploaded: every check runs inside your own browser, which is why it keeps working with the network disconnected. Your files are held on one server, used only for this work, shown to nobody, and deleted the day you ask.

From 14 September 2026 this page is a demonstration only. The Nova typesetting system, its terms and the status of every one of the twelve levels now live in one place, the private repository github.com/MariusGithub13/nova-typesetting-system, which you open through the GitHub invitation sent to you. Where anything on this page differs from that repository or from my email of 14 September, the repository and that email are correct. The price, payment, timing and upkeep terms that used to be written further down have been removed from this page, so that they can never disagree with the repository again.

One page · and it runs here

Checking Before Typesetting

Checks that read an incoming submission and say whether it is complete, before anyone spends typesetting time on it. Every rule it applies is quoted from the guidelines you sent on 14 August, not from a general style guide, and where your documents are silent it checks nothing. It uses your trim size, your resolution minimum and your terms because they are yours. You can run it on a real submission right now.

Try it on one of your own chapters

The file is read inside your browser and is never uploaded. Nothing is sent anywhere, nothing is stored, and this page works with your network disconnected. The one exception, so that nothing in the Network tab is a surprise: until you choose a file, the page asks this site every few minutes whether a newer version of itself has been published, and stops asking the moment you drop anything. It sends nothing but that question.

This is still meant to be argued with. Thirty-one changes were made to it today, thirteen of them from your own testing, and two checks were withdrawn once your guidelines turned out not to contain the rule behind them. So please keep going: tell me what it should catch and does not, what it flags that it should not, and which of these matter least to you. None of it is fixed until you have said.

What it finds in your own four submissions

These are your files, run against the current version, not an example. The two you sent on 27 August came with your team’s own checklist of what they found by hand, which is the only honest way to judge this, so those go first.

Book B · 25542 · against your team’s checklist

12 findings that would stop typesetting, 45 queries for a person to rule on. Nothing below was told to it in advance.

On 27 August the Chapter 4 line was the one it missed, and I told you so. It is caught now, both halves of it.
Your Chapter 7 line, “Missing figure 17”, I cannot reproduce from the file, and I would rather say that than manufacture a check that appears to find it. That chapter contains seventeen captions, seventeen references to them in the prose, and seventeen pictures, numbered 1 to 17 with no gap. Whatever your team saw is not encoded in the document. What chasing it did find is below, and it is worth more than the line I was chasing: all seventeen of those figures are under your own 350 dpi minimum, running between 221 and 284 dpi. If you can tell me what was on the page in front of you, I will build the check for it.

Book A · 13177 · the one this page cannot open

794 MB across 169 files, which is four times what a browser tab can hold. The figures below come from the installed version, in a single pass, and that is the whole argument for it: read chapter by chapter, every check that compares chapters against each other is destroyed, and this is a nine chapter book.

The two you sent on 14 August

Unreadable forms are counted as a failure of the tool, not of the author. Nothing is ever reported as clean when it could not be read. The six title versions on Book 2 are a question, not a fault: with no contents page in that submission there is nothing to judge them against, and a vote among forms may not accuse one.
Measured 28 August 2026 against Revision 50 of this page, on all four archives you have sent. Books B, 1 and 2 were run in this browser and you can reproduce them by dropping the same zip here. Book A was run on the installed version because it does not fit a tab. Every figure carries the moment it was taken, because this page changes several times a day and a number without a timestamp quietly becomes a claim about the past.

What you actually receive, and why this page is the weakest version of it

Everything is written in Python, and you get the source. Not a licence, not a thing you rent from me. Plain Python files your own people can open, read, run and change, on your server or on a Windows machine in your office. If you would prefer not to run it at all, I will host it and look after it for a fixed monthly figure, but that is a convenience you choose, not the way it is sold. The source is yours either way, and it keeps working whether or not you are paying me. If you and I part company, it keeps working exactly as it did the day before, and any competent person you employ can maintain it.

This page is a demonstration, not the tool. It is the same checks rewritten to run inside your browser so you could try them today without installing anything or sending a file anywhere. That convenience costs a great deal, and you should know exactly what, because some of it you have already run into:

The findings you have been reading came out of the version wearing all of those restrictions. That is the argument for the installed one, and it is the only argument I would make for it.

What changed at my end since we last worked together

You already know what I can do with LaTeX. That part is not new, and it is not what this proposal rests on. What is new is that I no longer only use the tools. I build them.

Python was the thread that ran underneath all of it, and it came back to the centre of what I do. Everything on this page is written in it: the checks, the reference verification, the build that compiles a book the same way twice. The page you are reading is not a description of a program, it is the program, running on your machine as you read.

Three things you can check rather than take from me:

• this page read your two submissions, all nineteen chapters and twenty-four copyright forms including the scanned ones, and produced findings you then corrected me on;
• the reference checker runs against a real book and reports what it cannot decide instead of guessing;
• the build environment is pinned to a fixed version, so a book compiled next year comes out the same as one compiled today.

That last one is the part that costs money, and it is worth saying plainly. Writing a check that works once is quick. Making it keep working when I am not there is the slow half: the tests, the pinned environment, the documentation, and the discipline of leaving alone anything the program cannot be certain about. The later levels put systems on your own machines, where they have to run for years without me. What you would be paying for is that, not the clever part.

I am telling you this because Levels 10 to 12 are a different order of work from checking references, and you should know who is proposing them before you read what they are.

The whole plan, so you can change it

Twelve levels. None of this is agreed, and the order is a suggestion, not a sequence you have to accept. It is written down so you can reorder it, delete from it, or tell me the thing that actually hurts is not on it at all. Five came from your own brief. Seven are mine, and those are the ones to be most sceptical about. They are numbered so that you and I can refer to the same thing by name in an email, not because they have to be taken in order: apart from the last three, which need the house rules encoded first, any of them can be built on its own.

Each is quoted and built separately, each is a fixed price agreed before it starts, and you can stop after any one of them. There is nothing you must renew to keep it working, and you own the source. (The upkeep arrangement further down, and my hosting it for you, are both optional and change neither of those.) I am not putting a number on the later pieces here: it would be a guess before the piece is scoped, and a guessed number is worse than none.

Level1

Reference checking · you asked for this · built and running

Every citation matched to a source and back again, in Word and in LaTeX, including references written into prose with no citation command, which standard tools cannot see. It only reads, so it can be judged against a book you have already published without risking anything.

You get: a report per chapter, and the checker itself.

Level2

Italicising references · you asked for this · not built

Your Chapter Guidelines say sources must be italicised and that bold must not appear anywhere in the references. The danger is never the reference it changes, it is the thing it touches that it should not: a species name already italic for another reason, a title inside a quotation, a foreign phrase. So it is built timid, and anything uncertain is left alone and listed for a person.

You get: the corrected file, and a line by line list of every change made.

Level3

The house rules in one place · your tasks 1, 2 and 3, as one job · not built

Trim and margins, front matter, and Nova styles applied to an author's manuscript are three names for the same question: where does the specification live. Encoded once, they apply the same way every time, instead of living in messages that whoever was not copied never saw.

You get: the rules as one file your people run, and change in one place.

Level4

Submission preflight · my recommendation · this page

Is the source there, are the figures and tables the text mentions actually present, does each chapter carry what the others carry, are the forms complete and matched to the right title. This is the piece you have been testing all day, and every change listed further down came out of that.

You get: one page per submission, saying go, go with queries, or go back to the author.

Level5

Chapter assembly and cross-checking · my recommendation · not built

Chapters gathered into a book in the order the contents page promises, with the numbering carried through and the cross-references between chapters checked. Your Book 1 already shows why: the contents page has no heading for Chapter 11.

You get: the assembled book, and a list of what did not line up.

Level6

Table and figure fitting · my recommendation · not built

Your guidelines give a maximum width of 4.5 inches portrait and 6.5 inches landscape, and figures at 350 to 750 dpi. Checking that is arithmetic, and arithmetic is where a machine should always win. Fitting what does not fit is the slower half, and it is the half worth automating.

You get: what will not fit, named, before anyone lays out a page.

Level7

Index consistency · my recommendation · not built

The same name indexed under two spellings splits an index silently, and no reader of the proof will catch it. On one 412 page scholarly work this found 61 names indexed twice. It is also where a naive tool does real damage: converting a manuscript can delete every index entry while reporting success, which I have measured.

You get: the entries that disagree with each other, before the index is set.

Level8

Copyright page and CIP · my recommendation · not built

The Cataloging-in-Publication block compared against the Library of Congress record character for character, as they require. It also tells real CIP data from the permitted fallback wording.

You get: a pass, or the precise difference, per title.

Level9

Putting it in your people's hands · you asked for this · not built

You said your primary goal is a system you operate in-house. That is not a footnote, it is a piece of work: your templates migrated, the documentation your production people follow without me, and whoever needs it able to run all of it on their own machine. Anything that needs me on a Tuesday to process a manuscript is a service, not a system, and it is not what you asked for.

You get: it running on your machines, documented, with nobody depending on me.

Level10

Your LaTeX books building themselves, identically, on your own server · my recommendation · the engine exists, the deployment does not

A private repository in Nova’s own GitHub account, holding the house class file and your templates, deployed to a server you control. A chapter goes in, a typeset PDF comes out, and it is the same PDF whoever runs it.

This answers something you wrote to me. On book 23443 you said “I am not sure if the corrections are causing the errors or if it is an incompatibility between our latex programs.” You were right, and the cause is mechanical: nineteen typesetters means nineteen LaTeX installations quietly diverging. The build here runs inside one fixed, version-locked environment, so there is nothing left to diverge. It also removes a step nobody knew was mandatory: the 6x9 master as distributed does not compile, on your own 2011 toolchain or on a current one, so every typesetter has had to repair it locally before they could build anything, each in their own way, with nothing recording what they changed.

You get: the repository, in your account. The build, on your machines. Every change to the house spec dated, attributed and reversible, instead of living in a message someone was not copied on.

Level11

The same for Word, which is what most of your authors actually send · my recommendation · the largest piece here

Nova styles applied to an author’s .docx and laid out to your trim, on the same server, from the same repository. It is not an alternative to the piece above: your own books are mixed, Word chapters and LaTeX chapters inside one title, so a book needs both halves of one pipeline.

One thing I will not do, and it is why this piece needs care rather than speed. The obvious tool for converting Word silently deletes every index entry while reporting success. I have measured it on a real scholarly work: seventeen thousand index entries to zero, with a clean exit. So this route does not use it. It applies your styles to the document itself, and it counts the index entries before and after and refuses to continue if the number moves.

You get: the typeset chapter, the original untouched beside it, and a refusal rather than a quiet loss whenever something does not add up.

Level12

One book out of both, which is what you actually ship · my recommendation · the last piece, and the largest

The two pieces above each produce half a book. Your books are mixed, so neither half is the thing you send to print. This is the piece that takes the LaTeX chapters and the Word chapters and makes them one volume: continuous pagination across both, one contents page, one index merged from both sources, running heads throughout, and pages a reader cannot tell apart.

I used to do this by hand for you. My own invoices from the time describe the job in those words: “Mixed book, six Word chapters and three LaTeX”, and another, “Mixed Biomedical book, ten Word chapters and one LaTeX”. Reconciling two typesetting routes into one volume is not a new idea I am proposing to you. It is the step that was always there, done by a person, and it is the last one still done that way.

The hard part is not the merging, it is the index. Word index entries and LaTeX index entries have to become a single alphabetised index with page numbers that do not exist until after the merge. That is why the index check earlier in this list matters more than it looks: a name indexed under two spellings quietly splits an entry, and at this stage it splits it in the finished book.

You get: the volume, print ready, from chapters that arrived in two different formats. And a refusal rather than a quiet loss: if an index entry or a reference goes missing between the chapters and the book, it stops and tells you which.

How the last three fit the first nine, because that is the whole design. Everything from 1 to 9 only reads. None of it can alter a manuscript, which is why the free first step is safe to accept. 10, 11 and 12 are the pieces that produce something, and they are built to sit behind the others rather than beside them:

A submission arrives and preflight (4) is the gate. If it fails, it goes back to the author and never costs anyone typesetting time. If it passes, the house spec (3) is what the build applies, reference checking (1) runs inside the build so a chapter with an undefined citation cannot reach page proof, assembly (5) turns built chapters into a book, and the copyright page check (8) reads the front matter that came out. Your own blocked-book log has a title where everything was finished, index and all, and the book could not ship for one undefined reference. That is the exact failure this arrangement makes impossible.

Two ways to run it, and the choice is yours, not mine.

On your own machines. A server at Nova, or a rented one, or a single Windows machine for the Word half. This is what you said you wanted: a system you operate independently in-house. The source is yours either way, so this option is always open to you, and it stays open even if you take the other one first and change your mind later.

Or I run it for you. If nobody at Nova wants to look after a server, I will host it on mine and keep it running: the environment maintained and version-locked, the rules updated as you change them, submissions processed as they arrive. A fixed monthly figure, no timesheets. This changes nothing about ownership. You still receive the full source, it still runs on your own machines the day you want it to, and if you ever stop paying me the system does not stop working, it simply stops being my problem to keep alive. I would not offer you anything that takes itself away when the invoice does.

One honest note if you choose the second: your manuscripts and your authors’ signed forms would sit on my machine while the work runs. That is a bigger standing commitment than the confidentiality agreement we have already signed covers, so I would want it written down properly first, on your terms rather than mine.

What 10 and 11 need from you that the others do not: somebody at Nova who can run a server, or a rented one. If neither is available, piece 11 still works on a single Windows machine, it simply will not run unattended overnight.

Why this order. Reference checking is first because it only reads, so you can judge it against a book you have already published without risking anything. If it does not earn its place, nothing is owed and nothing has been disturbed. Then the two you named as most tedious, then the ones I am recommending, and handover last because it should cover everything by then. But it is a suggestion. If the preflight questions you asked are the ones that hurt, we start there instead, and if something on this list is worthless to you, say so and it comes off.

What I would want from you on this page more than anything else: tell me which of the twelve you would cut. A list that survives contact with the person who does the work every day is worth more than a longer one I wrote alone.

The first step, at no charge

Nothing needs to be agreed before that, and if it does not earn its place then nothing is owed and no harm is done. No meetings.

Price, payments, timing and upkeep

These terms are no longer kept on this page. The current ones are in the file docs/PRICE-AND-PAYMENTS.md in the repository, and the reasons for the changes made on 14 September, including a fixed price, payment for each level you accept, the six week delivery, a twelve month warranty and an optional support plan, are set out in my email of that day. Keeping the terms in one place means your team can never again find two versions of the same offer from me, which is what happened on this page.

How this is built, and who controls it

Three rules I work by. They are not concessions written for this proposal. They are how everything I build is put together, and they are the reason the rest of this page can be stated without qualification.

One more thing, which matters to the people who will actually use it. This does not measure anybody. It looks at submissions as they arrive from authors, before your production people have touched them, and its whole purpose is to stop work landing on someone’s desk that should never have reached it. It keeps no per-person statistics and produces no record of who did what. I have been the person on the receiving end of that queue, and a tool that quietly turns into a way of watching the staff is worse than no tool at all. The report says what a submission is missing. It never says who.

Where the line sits

It flags. It never decides. A query is a question put to an editor, not a verdict passed on an author.

What I would need

You have already sent the first two of these, on 14 August, and both have been run. What follows is what is still outstanding.

• The chapters of the conference volume. I have its copyright forms and have run them, but not the manuscripts they belong to, so nothing about that book's chapters has been checked. It is also the submission where the forms and the book would first be checked against each other: with the contents page beside them, the tool can say which of the six book titles on those forms is the correct one instead of only reporting that they disagree.
• The rest of your list of what goes wrong, untidied. The messy original is more useful than a clean summary, because the exceptions are where the real specification lives. The two corrections you have already sent each changed the design rather than adding a rule.
• One book you have already published, in the files you worked from. I had guessed the PDF inside Book 1 might already be this, and I was wrong, so I am asking properly rather than leaving the guess standing. That file is an eight page editor agreement, not a book.

And one thing I have taken off this list rather than ask you for. I was going to ask you to rule on degrees, because your two documents do not agree: the Guidelines for Review of Final Proofs allow “Only MD or PhD, or other Doctorate degree, no Master or Bachelor degrees”, while your Chapter Guidelines say to list only the highest degree received and then print “John M. Anderson, BA, BS” as the correct form.
Your book settles it without you. Every author across all nineteen chapters and the contents page carries MD and nothing else, which is acceptable under every reading of both documents. So I check nothing where they disagree, nothing is blocked, and this costs you one line whenever a later book actually needs it. I would rather spend your attention on the two items above.

What has changed, and why

Revision 51 · 14 September 2026 · 68 changes, the terms now live in the repository, not on this page

Everything below dated 14 August came out of your own testing. You ran two real submissions, told me it caught nothing, then told me exactly what it was getting wrong, six numbered points at a time. Eleven of these changes are yours by name, and the ones that are not are faults your files exposed. The rules it now checks are the ones you sent on 14 August, not anything I inferred.

This list is here because you should be able to see how a supplier behaves when something of theirs is wrong, rather than take a description of it on trust. Every entry says what changed, what caused it, and who found it. The entries where the answer to that last question is you are the ones I would read first.

Each change has a fixed number. C1 is the oldest and the newest is at the top. A number is never reused and never changes, so you can write “C7 is not quite right yet” or “C3 should also cover tables” and we will both know exactly what is meant. Newest first below.

What each change was tested against. Everything from C8 down was tested on cases I rebuilt from your description. Everything from C9 up was tested on the two sets you sent on 14 August, the nineteen chapter clinical title and the folder of conference copyright forms. Book 1 is the one quoted in the figures below. The proceedings volume has not been run end to end yet.

What is still not right

• Book 1 now reports 8 blocking findings and 43 queries, against 14 and 105 when you first ran it. The two newest are the figure and table widths, which had never been measured: 69 figures wider than your maximum and 7 oversized tables. The rest are one blank form, 4 forms with no book title, the contents page with no Chapter 11 heading, and three chapters whose reference lists have lost their spaces.
• The conference copyright forms have now been run. 25 of the 30 were read, and the report is 5 blocking findings: 2 blank forms, 13 with the Book/Journal Title left empty, 6 different versions of the book title in circulation, and 2 forms this browser could not read at all. The chapters of that volume have not been run, because I have not been sent them.
• The formatting checks are built from your current Chapter Guidelines: bold inside running text, bold in the references, captions with no descriptive text, caption placement, figure and table widths, body text size, chapter length, picture alignment and italicised sources. Still not checked, deliberately: Chicago reference style, split tables repeating column headers, and non-English sources translated in square brackets. The first needs judgement, the second cannot be known before page layout, and the third needs language detection I do not trust enough to put in front of an author.
• A signature left inside a figure, and misspellings inside a figure image, are not detected.
• Whether a signature is genuine is not something this will ever answer. It can tell you a line is empty. It cannot tell you a person signed it, and I would not want it to pretend otherwise.
• A form whose fields are filled in by hand rather than typed will read as empty, because handwriting is not recognised. That is deliberate: a guess about handwriting is worth nothing here.

Built and run by Marius Andronie. Prepared for Nova Science Publishers and not published. The figures on this page come from Nova’s own production records; chapter and author identifiers have been removed throughout. Please treat the page as confidential to the people it was sent to.