Building Toward Type Nobody Can Take Back

Why an open platform works toward typographic independence, and how far that work has actually got

This paper is provided for engineering, operational, and research purposes and does not constitute investment advice or a solicitation to invest in any Woodfine direct-hold solution. Statements marked "planned," "intended," "targeted," "may," or "expected" are forward-looking and subject to change. Full disclosures appear at the end of this paper.

Working Paper PDS-018 · v0.1.0 · CC BY 4.0

Our thesis is that a platform built so that its customers do not depend on a supplier has to apply the same standard to its own lettering — and that we are building toward that rather than reporting it as finished. A typeface is not a possession. It is licensed software, counted and metered like any other, and a document that depends on one depends on that licence continuing to hold. When the licence lapses, or the terms change, or the business simply grows past what it paid for, the document cannot be reissued as it was. For most documents this is an irritation. For a document a person may need to produce again, unaltered, years after it was signed, it is a real defect in a real thing.

The direction we hold is straightforward: every typeface a document of ours depends on should be one that nobody can withdraw. Real work supports that direction and is worth being specific about. Every brand-facing typeface once specified from a proprietary vendor library now has a named openly-licensed replacement recorded in the published typography rules. The site publishing this paper holds its own body face and its own monospaced face as files in its own deployment, under a licence anyone may rely on, with the full licence text shipped beside the files and no request going out to an outside font service when a reader opens a page.

That work is evidence of a direction in progress, not proof the direction has been travelled. Four things are not finished and this paper says so in its own section rather than a footnote: no document-generation tool consumes the print typography rules yet; the retired proprietary names are still sitting in the live style fallback chains, ahead of their open replacements in two of three cases; this site's own heading face is named in its style variables with no font file behind it, so headings fall through to whatever serif the reader's machine has; and where the platform does rely on ordinary system faces, that is a choice about speed and third-party requests rather than an independence achieved. Those are different claims and conflating them would flatter us.

1. The thesis

A platform can be open in its code, open in its data, and open in its licence, and still have one thread running out of it to a supplier who can pull. Typography is a common place for that thread to be, precisely because it looks like a matter of aesthetics rather than of dependency.

Our position is that it is a dependency, and it should be removed the same way every other dependency on this platform is meant to be removed — not by promising to behave well, but by arranging things so that nobody has a lever to pull. A document set out in a typeface that anyone may use and redistribute cannot be broken by a commercial decision taken somewhere else. A document set out in a licensed commercial face can be.

The reason to care is not the appearance of the page. It is the reproducibility of the document. A great deal of what this platform is for is producing records that are meant to outlast the software that produced them: agreements, statements, filings, records attached to a building. A record you cannot reproduce in its original form is a weaker record than one you can. Type is part of the original form.

The composition is the claim, and it is a modest one. Openly licensed typefaces are ordinary and widely used; there is nothing novel about choosing one. Holding font choices as named entries in a governed file rather than as strings scattered through stylesheets is also ordinary. What we are arguing is that the two together should be treated as a dependency question rather than a design question — that "which face" belongs in the same conversation as "which licence" and "who runs the server," and should be reasoned about with the same seriousness.

2. The problem, in the reader's terms

Most people have never bought a typeface and reasonably assume that a font is something that comes with a computer, the way a ruler comes with a pencil case. It is worth being concrete about what a commercial typeface actually is, because the exposure is not obvious until you see the meter.

A typeface is a design; a font is the software file that draws it. Selling that file is a licensing business, and the licence is metered in ways that surprise people the first time they read one. A desktop licence is commonly counted by the number of people who may install it. A web licence is commonly counted by page views per month or per year. An application licence is counted by the number of applications the file is embedded in. A licence to embed the face inside a document that will be widely distributed — a printed report, a filed prospectus — is often a separate grant again. Exceeding a count is a breach, and the counts are the kind that ordinary business growth exceeds without anyone deciding anything.

Now think about what that means for a document. A set of statutory financial statements is typeset in a licensed face. Five years later, the business needs to reissue that document — because a regulator asked, because a buyer is conducting due diligence, because a dispute turns on what the original said. If the licence has lapsed, or the seat count has moved on, or the supplier has been acquired and the terms have changed, the business is in the peculiar position of not being permitted to reproduce its own document in its own form. The content is theirs. The lettering is not.

The traditional comparison is exact and old. A printer setting metal type owned the type. The trays of sorts sat in the shop and were the shop's property; a printer who kept his type could reprint a job in twenty years. Photo-typesetting and then digital typesetting turned the trays into a subscription — which brought enormous benefits and also quietly converted a durable asset into a recurring permission. An openly licensed typeface returns the tray to the shop. Not because the open face is better than the commercial one, which it often is not, but because it is held rather than permitted.

There is a second, quieter version of the problem, and it belongs to the web rather than to print. A great many sites do not hold their fonts at all. They load them, when a reader opens a page, from a large third-party font service. That is convenient and free and it means every reader's browser makes a request to a company that is not the one whose page they thought they were reading. Two things follow: the page stops working the way it should if that service is unreachable, and the reader's visit is visible to a party they did not choose. Holding the font files on the same server as the page removes both, and costs a few hundred kilobytes.

3. What is actually being governed

This section sets out the mechanics plainly. A reader comfortable with font licensing and fallback chains may go to Section 4.

The licence is the thing that matters, and it has a standard name

The openly licensed typefaces used throughout this platform are published under the SIL Open Font License, version 1.1 [sil-ofl-1-1]. It is worth reading what that licence actually does, because "open" is a word that covers a range.

It grants permission, free of charge, to use, study, copy, embed, modify, redistribute and sell copies of the font — with no fee, no seat count and no page-view meter. Three of its conditions matter here. The font files may not be sold on their own, though they may be bundled, embedded, redistributed and sold inside any software. A modified version may not use the original's reserved name, so an altered face cannot be passed off as the original, which protects a reader's expectation that a named face is the face they think it is. And each copy distributed must carry the copyright notice and the licence text with it, which is a small housekeeping duty rather than a restriction.

Two absences matter as much as the conditions. There is no clock and no conversion date; the permission is not withdrawn later and does not require renewing. And the licence says in terms that the requirement for the font to stay under this licence does not reach any document created with it — so a document set in such a face carries no licence obligation of its own, which is exactly the property a record meant to be reissued years later needs.

Licences of this kind are identified by short standard codes maintained in a public registry, so an organisation's compliance team can look up exactly what any named licence permits rather than taking a supplier's characterisation of it on trust [spdx-license-list]. That matters for the thesis: "this face is safe to depend on" is a checkable statement, not an assurance.

A font choice is a token

The platform records its font choices the way it records every other design decision: as a named entry in a governed file, with a value and a written description of when it applies, in the standard vendor-neutral interchange format maintained by the Design Tokens Community Group under the World Wide Web Consortium [dtcg-w3c]. A companion paper on this site explains that mechanism in full.

The practical consequence is that a font decision is addressable. There is a name for "the formal serif for institutional print," a name for "the condensed sans for dense financial tables," and a name for "the workhorse body face." Documents and stylesheets refer to the name. The face behind the name can be changed in one place, and everything that meant that role follows — which is precisely the mechanism that made the substitution described in the next section a change to a handful of entries rather than a hunt through every document ever produced.

What a fallback chain is, and why the order in it is a fact

One more piece of mechanics, because the honest limit in Section 6 turns on it.

A stylesheet rarely names a single face. It names a list — a fallback chain — and the machine displaying the page walks the list from left to right and uses the first face it actually has available. The list exists because the machine showing the page may be any machine, and the author cannot know what is installed on it.

The order of that list is therefore not decorative. It is an instruction about preference, evaluated on the reader's own computer. A name sitting first in the list is the face that will be used wherever it is present. A name sitting third is a face that will be used only where the first two are missing. Which means that removing a face from a design and removing its name from the chain are two different acts, and only the second one is mechanically true.

There is a second, separate instruction that matters just as much, and the two are easy to confuse. Naming a face in the chain says which face is wanted. A different instruction elsewhere in the stylesheet says where the file for that face may be fetched from. If the second instruction is missing, the first is merely a wish: the browser looks for the named face on the reader's own machine, does not find it, and quietly moves down the chain to something it does have. Nothing announces the substitution. The page looks entirely reasonable, and it is set in a face nobody chose. Both halves of the honest accounting in Section 6 turn on this distinction.

4. What has actually been replaced

The substitution is real, completed as a decision, and recorded where anyone can read it. It is presented here as evidence supporting a direction, not as a claim that the direction has been fully travelled.

The platform's brand typography was originally specified from a well-known proprietary vendor library — the kind of collection an agency licenses and a corporate identity is built on. Four roles were involved, and each now names an openly licensed face instead.

The formal serif — the face for a white-paper cover, a legal block heading, the title of a formal document — was a slab serif from that library. Its replacement is Zilla Slab, a humanist slab serif under the open licence described above, carrying the same institutional weight the role was chosen for.

The condensed sans for dense data — the face for a financial ledger, a tight table, a capitalised micro-heading where horizontal room is scarce — was a condensed grotesque from the same library. Its replacement is Barlow Condensed.

The body face — the workhorse behind ordinary paragraphs, memoranda and operational text — was a geometric sans from that library. Its replacement is Nunito Sans.

And a fourth role, a high-contrast editorial serif for pull quotes and secondary markers, is filled by Sahitya, which was openly licensed from the start rather than replacing anything.

The stated reason for the exercise is the one this paper is about, and it is recorded in the rules themselves rather than reconstructed here: every asset embedded in a document the platform produces must be freely distributable, so that no font-licensing review can ever stand between the platform and producing a document. The replacements were not chosen because they were free. They were chosen because a licence that cannot be withdrawn removes a dependency.

The same principle runs through the site serving the pages you are reading. Its body face is Inter, in five weights; its face for code, command examples and identifiers is JetBrains Mono, in two. Both are published under the same open licence described above. Both are held as files inside the deployment's own static assets rather than fetched from an outside font service when a reader opens a page, so no request leaves for a third party. And the full text of the licence for each ships in the same directory as the files it covers, naming the face's own authors — which is precisely the housekeeping duty described in the previous section, done in the most ordinary way available.

That arrangement is also, in its small way, the second kind of evidence this paper can offer, because it replaced something broken. The stylesheet that now declares those faces records on its own face what it corrected: the two faces named there previously were named and never loaded. There was no instruction anywhere to fetch them, so every reader had been seeing a substitute all along while the file confidently declared otherwise. A font name in a stylesheet with nothing behind it is not a typographic decision. It is a decision that looks like one.

5. Where ordinary system faces are used on purpose

Not every surface on this platform carries a brand typeface, and some deliberately carry none. The design system's own global typography rules name only generic categories — the reader's system interface sans, the reader's system monospace — and no custom face at all. Several of the document families in the print rules do the same, specifying ordinary system or metric-compatible faces for their body text rather than a brand face.

It is worth being precise about why, because it would be easy and wrong to present this as independence achieved.

The reason is speed and the absence of a third-party request. A generic system face is already on the machine. Nothing is downloaded, nothing is waited for, and no text flashes invisible while a file arrives. On an operator's working screen — a console, a panel, a tool somebody uses all day — that is simply better, and it would be better even if brand typefaces cost nothing and carried no licence at all. The print families follow the same logic for a different reason: a set of statutory financial statements is meant to look like a set of statutory financial statements, and the plain, universally available faces that register uses are the correct answer rather than a compromise.

So this is a simplicity choice, made for its own reasons. It happens to carry no font dependency, which is a welcome side effect. It is not the same thing as having removed a dependency that was there, and it should not be counted twice. A platform that used only system faces everywhere would have no typographic dependency and would also have made no typographic decision; that is not the position we are arguing for, and it is not the position we are in.

6. What is not finished

Four things, stated plainly.

No document-generation tool consumes the print typography rules yet. The open substitutions described in Section 4 specify what a print and PDF pipeline should use once one exists. They are not, today, a live mechanism embedding those faces into shipping documents. The platform's own published documentation says so in as many words, and that is the accurate description: a decision made and recorded, waiting on the machinery that will act on it.

The retired names are still in the live fallback chains. The platform's live style variables name three font roles, and each chain still contains the proprietary name it was meant to leave behind. In two of the three — the display role and the serif role — the retired name sits ahead of its own named open replacement in the order. Since a fallback chain is evaluated left to right on the reader's own machine, a machine with the retired face installed will use it in preference to the open substitute that was chosen to replace it. The first name in each chain is an openly licensed face, so the common case is unaffected; but the retired names have not been removed, and the ordering means the substitution is not yet mechanically complete in the place that decides what a reader actually sees. We found this by reading the file while writing this paper, which is roughly the point of writing it.

This site's own heading face has no file behind it. The previous section described a correction: two font names declared with nothing instructing a browser to fetch them. Exactly one instance of that defect is still present on this site, and it is in the paragraph above's own subject matter. The style variable for display and heading text names a serif face; there is no rule anywhere in the stylesheet telling a browser to load it, and no file for it in the directory holding the others. The heading you are reading is therefore set in whatever serif the machine in front of you happens to provide, by way of the fallback the same variable declares. Nothing is broken and nothing is misrepresented to a reader, but a named face that is never loaded is a decision recorded and not taken, and it is the sort of thing a paper of this kind should say rather than leave for a reader to discover.

The independence claim is a direction and should be read as one. What is true today is that the decisions have been made, recorded and published; that the platform serving these papers holds its own faces rather than borrowing them; and that the roles that would otherwise depend on a proprietary library have named open replacements waiting. What is not true is that a document produced by this platform today has been demonstrated end to end, from source text to finished file, using only faces nobody can withdraw — because the tool that would produce that file does not exist yet. We intend to build it. Saying it is done would be the easiest sentence in this paper to write and the one a reader could most simply disprove.

7. What this changes for the reader

The first change is what a reader can check. Typography is unusual among the independence claims a platform might make, because it is inspectable without any special access: the licences of the named faces are public, the font files served by a page can be seen by anyone who looks at what a page loads, and the rules recording the choices are published text files. A reader who wants one concrete, verifiable instance of whether this platform's independence talk is real can start here, and will find both the substitution and its unfinished edges.

The second change is about the durability of documents. A record produced entirely from openly licensed faces can be reproduced in its original form indefinitely, by anyone, with no permission needed from anyone. For a class of record meant to outlive the software that made it — an agreement, a statement, a filing — that is a property worth having and cheap to obtain, and it costs a reader nothing to ask their own suppliers about.

The third change is for readers of the pages themselves. A page whose fonts come from its own server does not report the visit to a third party and does not stop looking right when that third party is unreachable. That is a small benefit, freely available, that a great many sites decline for convenience.

The trade-offs are real and belong in the same paragraph. Openly licensed faces are genuinely constrained: the commercial libraries are deeper, better hinted in places, and carry weights and widths that do not always have an open equivalent, and a designer who needs a specific one is being asked to give something up. Holding font files yourself means carrying a few hundred kilobytes and updating them by hand rather than having a service do it. And the honest headline of this whole paper is that the completed part of the work is a set of recorded decisions and a self-hosted web platform, while the document pipeline those decisions were made for has yet to be built.

8. An open invitation

This paper states a position we hold and several questions we would rather have answered than assumed.

To researchers in digital typography and type design: the replacements in Section 4 were chosen for aesthetic correspondence to the faces they replace and for licence terms. Whether they hold up as durable choices over a ten- or twenty-year horizon — actively maintained, covering the character sets that bilingual and multi-jurisdictional documents need, rendering well on the kinds of hardware a printed regulatory document eventually passes through — is a genuinely open assessment we are not qualified to make about our own choices.

To specialists in document preservation and archival practice: the argument that an openly licensed face makes a document more reproducible assumes the font file itself remains available and unchanged. Fonts are versioned; metrics shift between versions; a document reproduced with a later version of the same named face is not necessarily identical. What the right practice is — embed the exact file with the document, record a fingerprint of it, something else entirely — is a question the preservation field has thought about far longer than we have.

To font-licensing and compliance specialists: we would value a rigorous reading of what an organisation actually has to do to stay clean when it redistributes openly licensed faces inside its own product and inside the documents that product generates. We believe the obligations are light. We would rather be told precisely where they are not than assume it.

To anyone who maintains typography across a large body of published material: both defects in Section 6 are the same shape — a name recorded in a file that does not match what a reader's machine actually does — and both were found by a person reading the file rather than by anything that reports it. We would be glad to learn what genuinely catches this class of drift in practice: a face named but never fetched, or a retired name still sitting ahead of its replacement in a preference list.

And to designers who have moved a real brand identity off a commercial library: we are interested in what was lost, not what was gained. The gains are easy to list and we have listed them. The losses are the part a paper written by the party making the change is least likely to see clearly.

9. Conclusion

A platform built so that its customers do not depend on a supplier should be built toward typographic independence too, and that is the direction we hold. The reason is not aesthetic. It is that a document whose typeface licence can lapse is a document that may one day be impossible to reissue in its original form, and a great deal of what this platform exists to produce is records meant to outlast the software that made them.

The work supporting that direction is real and specific: every brand-facing role once filled from a proprietary library now names an openly licensed replacement in the published rules, and the site serving these papers holds its own body and monospaced faces under a licence nobody can withdraw, in its own deployment, with the licence text beside them and no outside request. The work remaining is equally specific: no document-generation tool consumes those print rules yet; the retired proprietary names are still sitting in the live fallback chains ahead of their replacements in two of three roles; this site's own heading face is named with no file behind it and resolves to whatever serif the reader already has; and using ordinary system faces on an interface is a choice about speed rather than an independence achieved. We would rather publish a direction with its distance still to run than a completion a reader could disprove by opening one file.

References

SIL International. SIL Open Font License, Version 1.1. https://openfontlicense.org/

Linux Foundation. SPDX License List — standardised licence identifier registry. https://spdx.org/licenses/

Design Tokens Community Group. Design Tokens Format Module. World Wide Web Consortium Community Group. https://tr.designtokens.org/format/

Contributors

Prepared by Woodfine Management Corp.

How this paper was produced

This paper is grounded in the operating and development work the preparing staff do themselves, and in what they have learned from the people they work with routinely: the graphic designers, web developers, and software developers and engineers who build the platform alongside them, and the architects and structural, building-services, and civil engineers they develop buildings with. No outside professional reviewed or approved this paper, and nothing in it is professional advice. It was drafted and edited with AI assistance under editorial direction, and the analysis and conclusions are Woodfine's own.

Disclosures

Woodfine Capital Projects Inc. ("Woodfine") is the author of record. PointSav Digital Systems, which builds the platform described, is currently a trade name of Woodfine, planned to become a wholly-owned Woodfine subsidiary upon incorporation; PointSav does not itself offer, sell, or solicit any security. The typography rules and the platform described are our own work, so this paper argues for an approach we have a commercial interest in. This work was funded internally; no external research funding was received. The descriptions of how font licensing operates are given in general terms to explain a design and are not legal advice on any party's rights or obligations under any font licence. No claim is made or implied about the terms, conduct or quality of any typeface supplier; the faces replaced are described by role rather than by vendor because the argument is about dependency structure, not about any particular firm. This paper's content is provided for engineering, operational, and research purposes and does not constitute investment advice or a solicitation to invest in any Woodfine direct-hold solution. Some statements above describe planned or intended future work; in particular, the document-generation pipeline that would consume the print typography rules is intended and does not exist, and the removal of the retired proprietary names from the live fallback chains is intended and has not been done. Language such as "planned," "intended," "targeted," "may," and "expected" marks this forward-looking content, which is subject to change and does not constitute a commitment regarding future performance.

Data and reproducibility

Everything this paper reports was read directly from the published files in preparing it, rather than from a summary. The four typeface roles, the faces they now name, and the proprietary faces each replaced are recorded together in one short published document in the design-system repository, along with the stated reason for the exercise. The platform's own global typography rules, which name only generic system categories and no custom face, are a separate published file in the same repository. The live style variables reported in Section 6 are three lines in a single published stylesheet file, and the ordering described there — the retired name sitting ahead of its open replacement in two of the three chains — can be confirmed by reading those three lines. The font files this site serves are in its own static assets directory, visible to anyone who inspects what a page loads, and the full licence text for each sits in that same directory as an ordinary text file naming the face's authors; the licences are also public documents identified by standard registry codes [spdx-license-list], so a reader can confirm what each one permits without relying on this paper's description. The heading-face gap reported in Section 6 is confirmable the same way and in the same stylesheet: the variable naming that face is present, the instruction that would fetch it is not, and no file for it exists among the others. That stylesheet's own comments record the earlier version of the same defect, described in Section 4, in its own words. The statement that no document-generation tool consumes the print rules is the platform's own published current-state description, not an inference drawn here. No independent party has audited these typeface choices, verified these licence terms, or reviewed these claims.

PointSav Digital Systems™ and Woodfine Capital Projects™ are trademarks of Woodfine Capital Projects Inc.