microsoft.public.word.numbering

Long-form reference · 3,143 words · about a 14-minute read

A group named after a failure mode

Support hierarchies normally name their groups after products, and sometimes after tasks. The Word corner of Microsoft's news server mostly followed the pattern: there were groups for mail merge, for tables, for page layout, for document management, each mapping onto something a user might be trying to do. microsoft.public.word.numbering broke the pattern in an instructive way. It was named after a single feature, and it existed because that feature generated enough grief, steadily, version after version, to sustain a discussion group of its own for well over a decade. Nobody convenes a standing support group for a feature that works.

The subject was automatic numbering in Microsoft Word: numbered and bulleted lists, outline-numbered headings, chapter and caption numbers, and the multilevel machinery underneath all of them. The feature's failure modes were few and well known. That consistency, sustained across Word versions and across the group's whole life, is what this page documents: a small, dense case study in what happens when a powerful feature is also an opaque one, and when the people who understand it best are unpaid volunteers on a vendor's server.

microsoft.public.*: a hierarchy outside the Big Eight

The group belonged to microsoft.public.*, a hierarchy Microsoft ran itself from the mid-1990s as a peer-support channel. Unlike the Big Eight, where creating a group meant a public request for discussion and a vote, Microsoft created groups administratively whenever a product team wanted one — the normal arrangement for a vendor hierarchy, but pursued here at unusual scale. By the company's own accounting when it wound the system down in 2010, it was hosting more than 2,000 public newsgroups covering, in its words, virtually all of its products, alongside more than 2,200 private newsgroups reserved for partners, certification candidates and MVPs. The Word family alone splintered into dozens of subgroups, and that automatic numbering warranted an address of its own, rather than a corner of a general formatting group, says a good deal about the feature.

Vendor hierarchies were not a Microsoft invention — companies and projects had long run their own, the Free Software Foundation's gnu.* being the classic early example — but microsoft.public.* became probably the largest and most consequential of them, simply because so much of the world's office work ran on the products it covered. For a stretch of the late 1990s and 2000s it was many ordinary users' first and only contact with Usenet: they arrived not out of interest in the medium but because a knowledge-base article or a colleague had pointed them at a newsgroup, asked their question, and left. The regulars stayed.

The infrastructure was the public news server msnews.microsoft.com: free NNTP access, no account required, any newsreader welcome. The groups were unmoderated — Microsoft said so plainly when explaining why it preferred forums, and the spam load of an unmoderated hierarchy was one of its stated complaints — and the company's servers kept articles for only ninety days, a retention policy whose consequences for the historical record are taken up at the end of this page. The hierarchy also travelled beyond Microsoft's own machines: plenty of ISP news servers and public services carried microsoft.public.* alongside the canonical hierarchies, and Google's Usenet gateway archived it, which is the main reason the discussion outlived its host.

This directory reached the group the same way it reached everything else, through the mail gateway described on the front page; the mechanics are its story, not this one.

The MVPs: volunteers on a corporate server

The bargain that made the hierarchy work was simple and slightly odd: Microsoft supplied the server, and volunteers supplied most of the answers. The busiest of them were recognised through the Most Valuable Professional programme, which grew out of exactly this kind of unpaid answering: some of the earliest MVPs were the most active helpers on CompuServe and Usenet, and one widely repeated account traces the award's origin to Microsoft's CompuServe FoxPro forum, where a running tally of who answered the most questions was eventually formalised into recognition. The award lasts a year at a time and is given for the previous year's contributions, which kept it honest as a measure of current activity rather than past glory; among its perks were the private newsgroups Microsoft ran for MVPs alongside the public ones. The reward, in other words, was status, insider access and event invitations — never payment.

That distinction was tested once, memorably. On 22 October 1999 a Microsoft executive announced the programme's cancellation — a decision that may have been a response to a lawsuit then facing AOL, whose volunteer community leaders had argued they deserved wages for their hours online. The protest from the community was immediate, some of it aimed directly at Bill Gates and Steve Ballmer, and Microsoft rescinded the cancellation three days later. The reversal settled the model for the following decade: the company got official-quality support for free, and the volunteers got a title and kept answering.

The Word MVPs institutionalised their answering unusually well. They maintained their own FAQ site, and its numbering section functioned as the group's reference shelf: an anatomy essay explaining how the machinery actually worked, a survey of methods for restarting list numbering, and a cookbook of field-based techniques for the cases the machinery could not be trusted with. The anatomy essay, Word's numbering explained, contributed by the Word MVP John McGhie, opens by warning the reader that the subject is a real brain-breaker, and that Word makes it worse by hiding complexity behind simplistic explanations — candour of a kind the official documentation did not offer, from people doing the vendor's support work for nothing.

Why numbering, of all features

Numbering earned its own group because Word's design made it genuinely hard to reason about. Word 97 introduced the list-template architecture that underpinned automatic numbering in every release through Word 2003, and those versions supplied the bulk of the group's traffic. Under that architecture, list formatting lives in per-document list templates that interact with paragraph styles, with direct formatting and with each other — and almost none of that state is visible in the interface. Two lists that look identical on screen can behave differently, because they hang off different hidden templates; an innocent edit in chapter two can rewrite the numbering in chapter nine. The model was powerful enough to do nearly anything and opaque enough that few users could tell what it was doing, a combination that guaranteed the same questions arrived every week from new people.

Part of the confusion was that "numbering" in Word is not one feature but several stacked on top of each other, and the MVP literature spent real effort just separating them. The reference essay's own taxonomy distinguishes field numbering — explicit counters placed in the text — from simple list numbering and from outline list numbering, the multilevel scheme that ties numbers to a hierarchy of levels; and it distinguishes, separately, the two ways any of it can be applied, directly to paragraphs or through styles. Page numbers, footnote numbers, chapter numbers, caption numbers and table-of-contents entries each ride on different parts of this stack. A user who asked why their numbering was broken was, roughly half the time, asking about a different feature from the one they thought they were using, and the first job of an answer was often taxonomy rather than repair.

The interface made its own contribution. The toolbar numbering button applied list formatting directly, on top of whatever the underlying styles specified, and direct formatting layered over numbered styles was treated by the group's regulars as the root of most evils: it multiplied hidden list state, and it meant a document's visible numbering no longer corresponded to anything a style sheet could predict or repair. The right-click choice between restarting a list and continuing the previous one generated its own steady traffic.

Word 2007 rebuilt the surface — the ribbon, multilevel list galleries — but left the underlying machinery substantially intact, so the group's accumulated diagnoses carried over to the new release rather than expiring with the old ones. That continuity is worth pausing on: it meant a decade of archived troubleshooting stayed current, which is rare in software support and part of why the archive of this particular group kept being cited long after the group closed.

The clinic model

A vendor peer-support group of this kind had the rhythm of a walk-in clinic. A thread would open with a symptom, usually in the subject line; the standard intake questions followed — which version of Word, whether the numbering was linked to styles or applied from the toolbar, whether the document had travelled through other hands — and the diagnosis came from a fairly short differential. The failure modes that brought people to such a room were remarkably consistent:

• Lists that restarted at one unbidden, or refused to restart when asked.
• Outline numbering that detached itself from heading styles, so that chapter numbers vanished or every heading became 1.1.
• Numbering that changed after a document made a round trip through a colleague's copy of Word.
• Documents assembled by pasting from several sources, which imported one another's list formatting along with the text.
• Indents and number alignment drifting as a long document accumulated edits.

The answers were nearly as consistent as the questions, and they amounted to a doctrine. Abandon the toolbar buttons. Define a multilevel list once, link each of its levels to a style — for outline-numbered headings, to the built-in heading styles — and then leave it alone; let every numbered paragraph get its numbering from its style and nothing else. Much of this was written up at length on the Word MVPs' site, and a large fraction of threads ended with a pointer to one of those articles rather than a bespoke reply. In practice the group functioned as triage plus a routing layer over a small canonical literature, which is one reasonable way for a support community to spend volunteer hours.

For documents whose list formatting had corrupted beyond salvage, the standard prescription was surgical: select everything except the final paragraph mark, copy it into a fresh document, and rebuild the numbering there. The trick works because Word treats that last paragraph mark as a container for a layer of document-level formatting, so leaving it behind leaves the corruption behind with it. It says something about the machinery that the community's most reliable repair was, in effect, evacuating the building.

SEQ, LISTNUM and the escape hatches

A second strand of traffic concerned the escape hatches. Word's field codes offered numbering that behaved predictably, at the price of doing everything by hand: SEQ fields for named counters, LISTNUM for list numbers placed inline, AutoNum for the simplest cases. Fields were the standing recommendation for the problems the automatic machinery handled worst — captions numbered by chapter, appendices running to their own scheme, cross-references expected to survive heavy editing, and running example numbers of the kind linguistics papers depend on. Threads on these subjects read less like troubleshooting than like instruction in a small programming language, which in fairness is what Word's field codes are: a field is an instruction evaluated when the document is updated, it takes switches that modify its behaviour, and it can be combined with bookmarks and cross-reference fields so that "see example (12)" keeps pointing at the right example after the twelfth example moves. Learners had to absorb the discipline that came with it — fields only recalculate when told to, so the ritual of selecting everything and pressing the update key before printing became part of the workflow — but the numbers a field produced were the numbers it was told to produce, which after a fight with the automatic system felt like a luxury.

The MVP reference shelf tracked this strand too: the FAQ site's numbering section carried worked recipes for building numbered lists from SEQuence fields, for captions that include chapter numbers, and — a reminder that not all long documents are academic — for producing sequentially numbered invoices. The pattern across all of them is the same trade the group kept explaining to newcomers: the automatic system does everything for you until it does something to you, and the fields do nothing for you but also nothing to you.

The long-document people

The group's constituency was defined less by profession than by page count. Short documents rarely exposed the machinery; long ones exercised it until it failed. A dissertation is precisely the document that exposes the model — hundreds of pages, years of intermittent editing, numbered headings, captions, tables, appendices and examples all expected to hold to a faculty's formatting rules — and its author is precisely the user with no time to learn list-template internals when the numbering collapses close to a deadline. The same shape recurs outside the university: contracts, where clause numbers are load-bearing because other clauses and other documents cite them; technical manuals, with numbered procedures and revision cycles; corporate templates expected to number reliably in hundreds of hands. What unites them is that the numbers are structural, not decorative: when a contract's clause 14.2(b) silently becomes 14.3(b), or a manual's step numbering restarts halfway through a procedure, the document has not become uglier, it has become wrong. The group's traffic skewed accordingly towards people for whom the failure was urgent — which explains both its patience with newcomers, who arrived in a hurry and in trouble, and the regulars' insistence on teaching the doctrine rather than patching the symptom, since a patched document would be back.

Linguistics deserves its own sentence, because it explains who still follows links to this address. Numbered example sentences are the working currency of a linguistics paper: data is presented as example (1), (2), (3), and the argument refers back to it by number, so a numbering system that silently restarts or renumbers is not a cosmetic nuisance but a citation-integrity problem running through the whole text. Dissertation-writing linguists fought Word over exactly this, and their community — the readership of the LINGUIST List, among others — passed around pointers to the threads and MVP articles that solved it, which is how a Word support newsgroup ended up cited from an academic linguistics archive.

Some of academia solved the problem by leaving. LaTeX generates its numbering deterministically when the document is compiled, and stable automatic numbering was one of the standard reasons long-document academics gave for learning it; the TeX world's own newsgroup is preserved in this directory at comp.text.tex. The two groups are, in a sense, the same clinic with different treatment philosophies.

Neighbours in the hierarchy

microsoft.public.word.numbering sat inside a large family. The Word groups alone covered mail merge, tables, page layout, document management and more, with parallel families in other languages; the numbering group was among the more specialised, which is exactly the point of it. Beyond Word, the hierarchy stretched across the whole product line, from consumer support to developer plumbing.

This directory preserves several of its other corners, mostly from the administrative and developer side: the two WMI groups at microsoft.public.dotnet.framework.wmi and microsoft.public.windowsxp.wmi, and the scripting mainstay microsoft.public.scripting.vbscript. Each tells its own story; together they give a fair cross-section of what a vendor-run hierarchy looked like from the inside — the same server, the same MVP economy, very different rooms.

Closure and afterlife

The end came for the whole hierarchy at once. In the spring of 2010 — the tech press reported it that May — Microsoft announced that it would close its public newsgroups and migrate users to its web forums: Microsoft Answers for consumers, TechNet and MSDN for professionals and developers. The closure began in June 2010 with the lowest-volume groups and proceeded in phases, with notice promised in each affected group well before its close date and the company anticipating completion by the autumn. The stated reasons were centralising content, better spam management, the forums' ability to mark answers as answers and, with disarming frankness, what the announcement called perhaps the most appealing benefit of all: "we own the app". The company's own figures framed the decision: newsgroup traffic had halved over the preceding fiscal year while unique visitors to the forums had more than doubled.

For the community the trade had a real shape to it. The forums promised things NNTP never delivered: contributions kept visible far longer than the ninety days a newsgroup post survived, answers that could be marked and found as answers, moderation, and profiles that identified whether a reply came from an MVP, from Microsoft staff or from a passer-by. What was lost was equally concrete — the choice of client, the threading conventions of two decades, offline reading, and the odd, productive adjacency of a vendor's support groups sitting on the same protocol and often the same servers as the rest of Usenet. The MVP economy itself survived the move largely intact; the volunteers followed the questions, as volunteers do.

Because msnews.microsoft.com kept articles for only ninety days, Microsoft itself never held an archive worth the name; the historical record is entirely third-party. The most accessible copy is Google Groups, which gatewayed and archived microsoft.public.* and still holds this group's threads in readable form — the collection descends from Deja News, which had been archiving Usenet since 1995 and whose archive Google bought in 2001 — though the service stopped supporting new Usenet content in February 2024 and its archive comes with the usual caveats: patchy coverage of the early years, honoured deletion requests, and search that has grown weaker over time. What survives is substantial but not complete, which is the normal condition of Usenet history.

The questions, meanwhile, outlived the group. They moved to the web forums after 2010, where they continue in recognisable form, because modern Word still runs on recognisably the same list machinery the group spent its life diagnosing. That gives the archive an unusual property: the advice preserved in it has aged well, and linking numbering to styles and leaving it alone remains, more or less, the procedure that works. Most preserved newsgroups are historical documents; this one is still, inconveniently for everyone who hoped the problem would be fixed by now, practical reading.