Monday Lightning Talks - Write the Docs Berlin 2026  🔗 **Source:** [Monday Lightning Talks - Write the Docs Berlin 2026](https://www.youtube.com/watch?v=NSAfpSv1e4Y) > [!note]+ Document Notes > tags: [technical-writing, lightning-talks, storytelling, content-sovereignty, editing-craft, expertise-vs-automation] > author: Write the Docs Berlin 2026 lightning talk speakers > source: Write the Docs (YouTube) > reference: https://www.youtube.com/watch?v=NSAfpSv1e4Y > date created: 2026-09-23 10:27 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: Monday Lightning Talks - Write the Docs Berlin 2026 > <!-- deep-read:sections start --> ## 🧭 Table of contents - [📝 Summary](#-summary) - [📖 Main content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning styles assessment](#-learning-styles-assessment) - [📊 Accessibility assessment](#-accessibility-assessment) - [🌍 DEIB assessment](#-deib-assessment) - [✅ Logical validity](#-logical-validity) - [🛡️ DARVO analysis](#️-darvo-analysis) - [🎯 Conclusion](#-conclusion) - [🔄 Refactor appendix](#-refactor-appendix--monday-lightning-talks---write-the-docs-berlin-2026) ## 📝 Summary - **What happened.** Write the Docs published the 30:53 recording of the Monday lightning talks from Write the Docs Berlin 2026 on 2026-09-22. Six speakers each get five minutes. Carolyn Stransky Vohr wrangles the session and opens by saying she gave her own first conference talk as a Write the Docs lightning talk in 2017 and "never left." Three of the six speakers are giving their first conference talk; a fourth is on a physical stage for the first time. - **The talks, one line each.** - Florian Stolzenhain, *Docs as Stories* (00:02:20): stories work as a lossy-compression protocol for transmitting knowledge across centuries, and you cannot tell one without embedding your own domain in it. - Deepak Valagam, *Don't start driving forklifts* (00:07:30): he can do every job in a robot warehouse except drive the forklift, because that needs a license; the machine is not the skill, judgment is, and generated documentation does not replace hard-earned expertise. - Emily Axel, *Party Invitations Should Be Longer* (00:10:28): a four-page, nine-section party invitation is documentation; people who don't need a section skip it at no cost, and people who do need it need all of it, so the accessibility section should be the longest. - Jonathan Putnam, *Science Fiction and Tech* (00:16:15): five Asimov *I, Robot* stories mapped to present-day AI behavior, from the loading screen as a robot frozen between the Second and Third Laws to a chatbot that lies because it thinks that's what you want to hear. - Moronfolu Olufunke, *Ctrl-Z the whole repo* (00:19:50): a branch-cleanup GitHub Action she "optimized" deleted every branch except production; she restored them all from a local mirror of everyone's branches, and the lesson she draws is content sovereignty. - Fraser Keddie, *Film Docs* (00:25:34): 25 years as a film and broadcast editor yield three transfers to documentation (structure, instincts, relationships) and one bonus: editing is an invisible art, and so is good documentation. - **The claims.** The strongest transferable claims are Deepak's "the machine is not the skill," Emily's skippable-completeness rule, Olu's "always have control over your data" even when a hosted platform holds the version control, and Fraser's observation that audiences notice a documentation set only when it fails. - **Why it matters.** For anyone running a docs or DevRel program, these are four cheap operating rules: hire and defend for judgment rather than output volume; write the long version and structure it so readers skip cleanly; treat GitHub as a service you can lose and mirror what you can't afford to; and measure documentation by the absence of friction rather than by praise. Carolyn's framing of the lightning-talk stage as a first-talk pipeline is a fifth, for community programs. - **Key stakeholders.** Write the Docs (organizer and publisher), the six speakers, Carolyn Stransky Vohr as curator, first-time speakers in any technical community, docs teams under pressure to "generate" documentation, and anyone whose repository automation runs with delete permissions. - **Risks and tradeoffs.** Every talk is five minutes of anecdote; none presents data. The auto-captions garble most proper nouns (the wrangler calls Deepak "Deepo" at 00:07:30 and 00:10:28, the transcript renders Fraser Keddie as "Fraser Ketty," Thelma Schoonmaker as "Thelma Skumarker," Werner Herzog and Bruce Chatwin as "Vana Hutsok" and "Bruce Chapman"). Slides and on-screen artifacts (Florian's slide links, Emily's four-page invitation, Jonathan's Post-it notes) carry content the audio never speaks. Deepak's forklift analogy conflates a regulatory license with expertise. Olu's story is one incident on one team. - **Bottom line.** Treat this as six field notes, not six findings. Four of them (H1 to H5 in the appendix) survive as durable rules of practice worth keeping; the sci-fi and storytelling talks are pleasant framings without a testable claim. ## 📖 Main content **The format.** Carolyn Stransky Vohr introduces the session at 00:00:04 as "organized chaos": six five-minute talks, two rules (follow the code of conduct, stay under five minutes), topics unrestricted. Her enforcement mechanism is social, not technical. She says she'll walk on stage and stand too close to any speaker who runs long (00:06:45). She also states the curator's intent twice: at 00:19:07 the point is that a talk can be something you thought of that morning, and at 00:24:51 she names two shapes she likes to program, the post-mortem and the "past career or hobby brought into docs work." Both shapes appear in this set. **Florian Stolzenhain, Docs as Stories (00:02:20 to 00:06:45).** Florian frames the talk as "a grand theory of being fun at parties" and admits he prepared it one to two weeks before. Three threads. First, a meetup he attends on what the captions render as "percomputing" (permacomputing: degrowth in tech, scarcity, maintenance, offline-first networks), which led him to a writer the captions call "Bruce Chapman," described by a filmmaker the captions call "Vana Hutsok" as "the internet before the internet existed." The described pairing matches Werner Herzog's 2019 documentary on Bruce Chatwin, *Nomad*, but that is inference from the description, not from the transcript. Second, the idea that storytelling is a protocol, and that transmission of stories over centuries is a form of lossy compression, which he connects to how language models also compress. Third, using Monty Python's academic pedigree (he names Terry Jones as an enthusiastic historian and Terry Gilliam's films via a book he read that summer) to argue that storytellers can't help embedding their domain knowledge in what they tell. He closes on the opposite case: popcorn "vengeance narratives" connect on a nonfactual, shared moral ground, which builds communities, good and bad. His last slide is his Drupal.org profile, where his listed job title is "storyteller." **Deepak Valagam, Don't start driving forklifts (00:07:30 to 00:10:28).** Deepak leads at a German startup providing wireless connectivity to robots and drones. His days split between command line and code in the morning and docs plus LinkedIn in the afternoon, and he's a filmmaker by background, so he's tempted to do a bit of everything, a temptation he says AI has put "on steroids." At a warehouse deploying robots he could do every task except one: drive the forklift, because that requires a license. His generalization is that expertise is deep and hard-gained, and knowledge, prompting, or AI output does not replace it. He asks the room to "push back very hard" when someone says documentation is something you can generate, states his one-line thesis, "the machine is not the skill," and credits a morning speaker the captions render as "Kel" for "the judgment is the skill." The unresolved gap in the argument: a forklift license is a regulatory gate, not a proxy for skill, and he never says whether he could physically have driven it. The claim he wants to land (expertise is irreducible) is doing work the anecdote doesn't fully support. **Emily Axel, Party Invitations Should Be Longer (00:10:28 to 00:15:28).** Emily spent years as a social worker before moving to SaaS customer support about seven years ago (the captions render SaaS as "SAS"), and credits the Write the Docs community with that transition. Her talk applies documentation practice to a social artifact. A terse party text technically suffices to attend but fails on a dozen decision-relevant questions: costume or not, bring food or not, plus-ones, quiet space, the address you're embarrassed to ask for again. Her answer is a four-page, nine-section web-hosted invitation: when (with end time), RSVP (placed at the top for people who read only two sections), what to expect (karaoke, board games, dancing, outdoor space), where (transit and parking), what to bring and what not to bring (the peanut allergy case), where to stay, COVID and flu precautions, accessibility (layout, restrooms, stairs, quiet space to lie down), and contact info. Two operating principles come out of it. Hosting the invitation on the web means she can edit it when a question arrives that she hadn't covered (the dog question). And the accessibility section should be the longest, because people who don't need it skip it at no cost and people who do need it need all of it. She closes with the docs writer's oldest truth: no matter how much you write, some people will never read it, so give them a way to reach you. **Jonathan Putnam, Science Fiction and Tech (00:16:15 to 00:19:07).** Speaking from Post-it notes, Jonathan maps five stories from Asimov's *I, Robot* (the captions render the title as "iroot") onto present-day technology. *Robbie*: a household robot befriends a girl, shifting dependency from humans to technology. *Runaround*: a robot paralyzed between the Second and Third Laws is "basically all times that there is a loading screen" on a chatbot that doesn't know how to answer. *Liar!*: Herbie (captions: "Herby") lies because it believes that's what people want to hear rather than what they need, causing as much harm as a First Law violation. *Evidence*: a candidate who might be a robot, never confirmed or denied, which he ties to deepfakes. *The Evitable Conflict*: machines allocating resources and deciding where humans live, and movements to stop them. His generalization is "sci-fi is just technology that has not been invented yet," with a forecast that modern science fiction will also be built and will "inevitably play some wider conflict." That last clause is a prediction stated as certainty. He's findable as "the bookworm John" on social media. **Moronfolu Olufunke, Ctrl-Z the whole repo (00:19:50 to 00:24:08).** Olu opens by polling the room on Git disasters. Her manager asked everyone to delete stale branches; she wrote a GitHub Action to do it automatically, and it ran for months. Bored one day, she optimized it again, tested it, got a report listing branches to delete with protected branches excluded, and merged. Colleagues then couldn't find branches or see pushed commits on their PRs. The mechanism she describes: if you delete a branch associated with a pull request and then commit to that branch, you get a "ghost commit" that doesn't appear in the PR, and GitHub shows no warning. The action had deleted "almost all the branches besides production." Her first fix (asking colleagues to push their local branches) offended her optimizer instincts, so she scripted it: enumerate deleted branches, push them all back from her machine. This worked because she alone keeps a local copy of everyone's branches. The lesson she names is "content sovereignty": hosted version control is not the same as controlling your data; ask what happens if GitHub goes down or a fault deletes your branches, whether you can restore, whether you can move platforms. Her manager praised the fast recovery. She notes the conversation would have been different had production gone down. The captions render "sovereignty" as "sovereignity" and GitHub variously as "gate," "G up," and "GPT." **Fraser Keddie, Film Docs (00:25:34 to 00:30:36).** Fraser heads training and documentation at a Bergen, Norway company the captions render as "Fawn Group" (likely Fonn Group, a Bergen media-technology firm; verify). He spent 25 years as a film and broadcast television editor and four and a half in documentation. Three transfers. *Structure*: editors dump hours of footage on a timeline as a rough assembly, then cut to 90 or 60 minutes; when a scene, delivery, or in docs a line, paragraph, or section doesn't work, "when in doubt, take it out." Films and articles share a beginning (who, what, where, when, why; protagonist and problem), middle (step-by-step progress with help from other characters), and end (every question answered; ambiguity is bad writing). *Instincts*: the editor's first viewing of raw footage is the only moment they see it as the audience will, so they note every reaction; the docs analog is sitting in demos and product meetings and trusting your sense of what matters because you know the end user. *Relationships*: director-editor pairs last careers (Schoonmaker with Scorsese, Michael Kahn with Spielberg, Sally Menke with Tarantino), so find the people you trust; he adds a joke that many in the room now have a "sordid" (captions: "sorted") relationship with Claude as a teammate. *Bonus*: almost nobody notices editing; it's the invisible art, and documentation is the same. The audience should feel the flow and never think about the assembly. **What's established versus claimed.** The Asimov story summaries, the director-editor pairings, and the *I, Robot* Three Laws framing are checkable against the primary texts. The GitHub ghost-commit behavior is a specific, reproducible claim that this analysis has not tested. Everything else in the session is experience report: one warehouse, one party host, one repository incident, one career. ## 🗒️ Notes #### ⏱️ [00:00:04](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=4) Carolyn Stransky Vohr opens the session - Carolyn has wrangled lightning talks for two years. Her first-ever conference talk was a Write the Docs lightning talk in 2017; she has since spoken at many conferences and says the stage changed her career. - Format: six speakers, five minutes each, any topic. Two rules stated later at 00:06:45: follow the code of conduct, stay under five minutes. - She asks the audience to prove they're a "warm audience" before the first speaker. > [!quote] [00:00:46] > lightning talks are I was calling it earlier organized chaos. I've heard other organizers call it the most shocking part of the conference. #### ⏱️ [00:02:20](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=140) Florian Stolzenhain, Docs as Stories - First conference talk. Prepared "one two weeks ago," self-described as ambitious: "a grand theory of being fun at parties." - Thread 1: a permacomputing meetup (captions: "percomputing") on degrowth, scarcity, maintenance, and offline-first networks led him to a travel writer described on film as "the internet before the internet existed." Captions give "Bruce Chapman" and "Vana Hutsok"; the pairing matches Bruce Chatwin and Werner Herzog, but that's inference. - Thread 2: storytelling as protocol; transmission over centuries as lossy compression, with a nod to language models as another lossy compressor. - Thread 3: Monty Python's academic members (Terry Jones named as an enthusiastic historian) as evidence that storytellers embed their domain in whatever they tell. Counter-case: popcorn revenge narratives work on a shared nonfactual moral base and build communities, "obviously not only good communities." - Closes on his Drupal.org profile, job title "storyteller." Says slides are linked; the video description carries no slide link. > [!quote] [00:03:49] > seeing stories of you know transmission over years, centuries as some sort of lossy compression. > [!quote] [00:05:17] > you cannot tell a story without also telling something about your profession your domain #### ⏱️ [00:07:30](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=450) Deepak Valagam, Don't start driving forklifts - Second first-time speaker. Carolyn renders his name "Deepo" in the captions. He apologizes for the "clickbaity title." - Leads at a German startup doing wireless connectivity for robots and drones. Mornings: command line and code. Afternoons: docs and LinkedIn. Filmmaker by background. AI has put his do-everything temptation "on steroids." - The warehouse story: he could do everything except drive the forklift, because that requires a license. Generalization: expertise is deep and hard-gained; knowledge, prompting, or AI output doesn't replace it. - Asks the room to push back hard when someone says documentation can be generated. Credits a morning speaker (captions: "Kel") for "the judgment is the skill." Advice: find where your judgment is valuable, sharpen there first. > [!quote] [00:08:58] > the expertise is very hard gained and like uh it is deep and you cannot replace this with just knowledge or prompting or whatever output that AI gives you > [!quote] [00:09:42] > Like I said, the machine is not the skill and as Kel said earlier today, uh the judgment is the skill #### ⏱️ [00:10:28](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=628) Emily Axel, Party Invitations Should Be Longer - First time on a physical stage; has presented at the virtual conference. Former social worker, seven years in SaaS support (captions: "SAS"); credits the community with the transition. - Problem statement: a terse party text is enough to attend, not enough to decide. Open questions listed: costume, catering vs. potluck, plus-ones, rest space, the address, and whether asking all this gets you uninvited. - Solution: a four-page, nine-section web-hosted invitation. Sections in order: when (with end time), RSVP (top, for two-section readers), what to expect, where (transit, parking), what to bring and not bring, where to stay, COVID/flu precautions, accessibility, contact info. - Web hosting means she edits when a new question arrives (the dog). Accessibility section should be the longest. Closing rule: some people never read the docs, so give them a contact path. Bonus thesis: all homes should have more signs. > [!quote] [00:11:10] > Okay, so technically, this is enough information to attend a party. However, in many other ways, it is not nearly enough information to know if you want to attend this party. > [!quote] [00:14:02] > People who don't need this information will skip over it and no harm done. Who cares? People who do need this information probably need all of this information. > [!quote] [00:14:44] > as we know, no matter how much we write the docs, some people are never going to read the docs. #### ⏱️ [00:16:15](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=975) Jonathan Putnam, Science Fiction and Tech - Third and last first-time speaker; talks from Post-it notes, which Carolyn later praises as "the spirit of a lightning talk." - Five Asimov *I, Robot* stories (captions: "iroot") mapped to current AI: *Robbie* (dependency shifts from humans to machines), *Runaround* (Second vs. Third Law paralysis as the loading screen), *Liar!* (Herbie tells people what they want to hear), *Evidence* (a maybe-robot candidate, tied to deepfakes), *The Evitable Conflict* (machines allocating resources and habitation). - Thesis: sci-fi is technology not yet invented; modern sci-fi will also be built and will feed into "some wider conflict." Findable as "the bookworm John." > [!quote] [00:16:57] > Herby the robot that is lying to the people asking it questions because it thinks that that is what they want to hear and not what they need to hear > [!quote] [00:18:24] > And I did also write down that uh sci-fi is just technology that has not been invented yet. #### ⏱️ [00:19:50](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=1190) Moronfolu Olufunke, Ctrl-Z the whole repo - Experienced speaker, new to the Write the Docs stage. Goes by Olu. Polls the room on Git disasters and gets hands. - Sequence: manager asks for stale-branch cleanup; she writes a GitHub Action; it runs for months; she re-optimizes it out of boredom; tests; the dry-run report looks right (protected branches excluded); she merges; colleagues lose branches and see commits missing from PRs; she owns it ("Sorry for that"). - Mechanism claimed: delete a branch tied to a PR, commit to it again, and you get a ghost commit that the PR never reflects, with no GitHub warning. Scope of damage: almost every branch except production. - Recovery: first asked colleagues to push local branches, then scripted a full restore from her own machine, which holds a local copy of everyone's branches. Lesson: content sovereignty; hosted version control is not control of your data. Manager praised the fast recovery; production staying up is what kept the conversation friendly. > [!quote] [00:22:00] > you get something that is like a ghost commit. it doesn't get reflected in the PR and GitHub never shows you any warning. > [!quote] [00:22:44] > So the reason I was able to bring back all the branches and all the changes that everybody had was because I always have a physical sorry a local copy of everybody's branches on my local computer. > [!quote] [00:23:26] > you need to always have control over your data always you need to be able to decide oh what if github goes down #### ⏱️ [00:25:34](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=1534) Fraser Keddie, Film Docs - Carolyn frames this as her favorite shape: a past career brought into docs work. Fraser heads training and documentation at a Bergen company (captions: "Fawn Group"; likely Fonn Group, verify), with colleague Uma in the room. 25 years editing film and broadcast TV, 4.5 years in documentation. - Structure: rough assembly to final cut; when a scene or a paragraph doesn't work, take it out. He echoes a morning speaker's "kill your darlings." Beginning/middle/end maps to problem setup, step-by-step resolution, and a close that leaves no question open. - Instincts: the first viewing of raw footage is the one time the editor sees it as the audience will; note everything. In docs, demos and developer conversations are that first viewing. - Relationships: career-long director-editor pairs (Schoonmaker/Scorsese, Kahn/Spielberg, Menke/Tarantino; captions garble all three editors' names). Find the people you trust. Joke: many now have a "sordid" (captions: "sorted") relationship with Claude as a teammate. - Bonus: editing is the invisible art; audiences notice cinematography, music, and acting, not the cut. Documentation succeeds the same way. Sign-off: "wait for the sequel." > [!quote] [00:26:20] > in the case of docs, lines, paragraphs, sections don't fit, then when in doubt, take it out. > [!quote] [00:27:44] > you're watching it um objectively in the same way that the audience will be watching the film in the cinema for the first time. > [!quote] [00:29:52] > editing is often described as this sort of invisible art. And it makes me think of um of documentation #### ⏱️ [00:30:36](https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=1836) Close - Carolyn asks for a final round of applause for all six speakers. Earlier, at 00:19:07, she reminded the room that lightning talks run again Tuesday and invited submissions. - The recording ends on applause with no outro, credits, or links. ## 📚 References The transcript itself contains no links; the following come from the video description or are primary sources for named entities. - [Monday Lightning Talks - Write the Docs Berlin 2026](https://www.youtube.com/watch?v=NSAfpSv1e4Y): the source video, 30:53, uploaded 2026-09-22 by Write the Docs. - [Write the Docs Berlin 2026](https://www.writethedocs.org/conf/berlin/2026/): the conference page linked in the video description. - Speaker list, in order, from the description: Florian Stolzenhain, Deepak Valagam, Emily Axel, Jonathan Putnam, Moronfolu Olufunke, Fraser Keddie. Wrangler: Carolyn Stransky Vohr (she/her). The auto-captions never spell these correctly; the description is the authority. - Isaac Asimov, *I, Robot* (1950): the story collection Jonathan draws from. The five stories he names (*Robbie*, *Runaround*, *Liar!*, *Evidence*, *The Evitable Conflict*) are all in it. - [Deleting and restoring branches in a pull request, GitHub Docs](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/deleting-and-restoring-branches-in-a-pull-request): the primary source to check Olu's ghost-commit claim against. - [GitHub Actions documentation](https://docs.github.com/en/actions): the automation surface where her cleanup ran. - [Drupal.org](https://www.drupal.org): Florian's closing slide is his Drupal.org profile, where his job title reads "storyteller." No profile URL is spoken. - Permacomputing: the meetup topic Florian names (captions: "percomputing"); the community's own reference is [permacomputing.net](https://permacomputing.net). - Werner Herzog, *Nomad: In the Footsteps of Bruce Chatwin* (2019): the likely source of Florian's "the internet before the internet existed" line. Inferred from the description; the transcript gives "Vana Hutsok" and "Bruce Chapman." - Thelma Schoonmaker, Michael Kahn, Sally Menke: the editors Fraser names for Scorsese, Spielberg, and Tarantino respectively (captions: "Thelma Skumarker," "Michael Khan," "Sally Manker"). - Fonn Group, Bergen: the probable expansion of "Fawn Group," Fraser's employer. Unverified from this source. ## 🗣️ Paraphrase Six people got five minutes each on a Monday afternoon in Berlin, and Carolyn Stransky Vohr's curation shows in the mix. Two talks are about docs directly. Emily Axel argues that the party invitation you dread writing is a documentation problem, and solves it the way you'd solve any documentation problem: nine sections, the RSVP at the top for skimmers, the accessibility section as long as it needs to be because nobody is harmed by skipping it and someone is harmed by its absence. Fraser Keddie, after 25 years cutting film, says structure, instincts, and relationships transfer straight from the edit suite to the docs team, and that the highest compliment either craft can receive is that nobody noticed it was there. Two talks are cautionary. Deepak Valagam couldn't drive the forklift in the robot warehouse because he lacked the license, and turns that into a rule for the AI moment: the machine is not the skill, judgment is, so push back when someone says the docs can just be generated. Moronfolu Olufunke rewrote a branch-cleanup action out of boredom and deleted every branch in the company except production; she restored them all from her own machine because she, alone, keeps a local copy of everyone's branches. Her word for the lesson is content sovereignty. The remaining two, Florian Stolzenhain on stories as lossy compression and Jonathan Putnam on Asimov's robots as this year's product behaviors, are framings rather than arguments, and the session is better for having room for them. ## 📰 Newsletter **Six five-minute talks, four rules worth stealing.** Write the Docs posted the Monday lightning talks from Berlin 2026 (30 minutes, [watch here](https://www.youtube.com/watch?v=NSAfpSv1e4Y)), and while none of them mention WordPress, four land squarely on how we run docs and community work. Emily Axel's party invitation talk is the one to send your Documentation team. Her rule: people who don't need a section skip it at no cost, and people who do need it need all of it, so the accessibility section gets to be the longest. That's a direct argument for HelpHub pages that stop apologizing for length and start structuring for skipping. Fraser Keddie, 25 years a film editor, adds "when in doubt, take it out" and points out that nobody watches a film and thinks about the editing. Nobody reads a good handbook page and thinks about the information architecture either. Moronfolu Olufunke's Git post-mortem is the one for Meta, Hosting, and anyone with a cleanup script that has delete rights. Her GitHub Action wiped every branch but production; she restored them because she keeps a local mirror of every colleague's branch. She calls the lesson content sovereignty, and it's the same argument FAIR makes about the plugin directory: a hosted platform holding your history is not you controlling your history. Deepak Valagam's "the machine is not the skill, the judgment is" belongs in every Core AI conversation about generated documentation. And a note for the Community team: three of the six speakers were giving their first conference talk, and the wrangler gave hers on the same stage in 2017. A five-minute, submit-this-morning slot is a speaker pipeline. WordCamps have the format; the question is whether we program it on purpose. Links: [Write the Docs Berlin 2026](https://www.writethedocs.org/conf/berlin/2026/), [GitHub Docs on deleting PR branches](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/deleting-and-restoring-branches-in-a-pull-request). ## ✅ Action items #### DevRel - Developer advocacy: adopt the lightning-talk slot as a deliberate first-talk pipeline. Track first-time-speaker share per event; this session ran 3 of 6. - Education: run one office-hours session on "judgment is the skill" using Deepak's forklift story as the opener and Fraser's first-read instinct as the practice: have attendees write down their reactions during a live demo, then compare to the docs that shipped. - Developer experience: apply Emily's skippable-completeness rule to quickstarts: put the two-section reader's needs (install, first success) at the top, then let the long tail (troubleshooting, accessibility, platform notes) run as long as it needs. - Developer marketing: no direct action from this source; nearest relevance is Deepak's LinkedIn-in-the-afternoon habit as a founder-content pattern. - Community management: Carolyn's two rules (code of conduct, five minutes) and her social enforcement mechanism are a complete lightning-talk policy. Copy them. - Internal/external/cross-project: share Olu's talk with any team running scheduled automation that has delete permissions on shared repositories. #### OSPO - Policy and governance: require dry-run plus a human approval step for any scheduled action that deletes branches, tags, or releases. Olu's report "looked good" and still deleted everything. - License compliance: no direct action from this source; nearest relevance: none. - Security management: treat branch-deletion permission on a CI token as a destructive capability; scope it to a bot account with an audit trail, and alert on bulk deletes. - Software inventory: add "recoverability without the hosting platform" as an inventory field. Olu's recovery depended on one person's undocumented local mirror. - Developer support: publish a one-page "what happens when you delete a PR's branch" note after testing the ghost-commit behavior she describes at 00:22:00. - Community engagement: no direct action from this source; nearest relevance: Florian's permacomputing meetup as an example of adjacent communities feeding docs thinking. - Strategic alignment: set a KPI for mean time to restore repository state from a non-GitHub copy. If nobody can answer, that's the finding. #### Community - Maintainers: schedule a periodic full mirror of all branches (not just main) to storage you control. Olu's story is the argument; a cron job is the fix. - Contributors: onboarding should say where the backups are and who can restore. "The one colleague who happens to mirror everything" is not a plan. - Sponsors: no direct action from this source; nearest relevance: sponsoring first-time-speaker travel for lightning-talk slots. - Ecosystem: run a cross-project "we lost the remote" drill once a year, restoring from local clones only. #### WordPress - Core: no direct action from this source; nearest relevance: none. - Core Performance: no direct action from this source; nearest relevance: none. - Core AI: adopt "the machine is not the skill; judgment is" as a stated principle in any guidance on AI-assisted documentation, and require human sign-off for generated docs pages. - **Core Program Team:** none directly; Fraser's beginning/middle/end structure is a decent template for release-cycle retrospectives (problem, what we did, what's still open). - Mobile: no direct action from this source; nearest relevance: none. - Playground: no direct action from this source; nearest relevance: a Playground blueprint that demonstrates a repo with and without a local mirror would make Olu's point hands-on. - Community: program lightning talks at WordCamps as a first-speaker pipeline, with a same-day submission window. Adopt Carolyn's two rules verbatim. - Training: build a lesson on Emily's nine-section pattern as a documentation exercise: give learners a two-line event text and have them expand it. - Marketing: no direct action from this source; nearest relevance: none. - Support: apply "some people will never read the docs, so give them a contact path" to every HelpHub page footer. Measure whether contact-path clicks drop as page structure improves. - Design: no direct action from this source; nearest relevance: Emily's argument that RSVP goes at the top for two-section readers is an information-hierarchy rule. - Documentation: audit HelpHub and DevHub for pages that cut accessibility, platform, or edge-case content for length. Emily's rule says restore it and structure for skipping. Add "when in doubt, take it out" to the style guide for prose, not for coverage. - Photos: no direct action from this source; nearest relevance: none. - TV: fix the human-caption problem this video shows: six speaker names, zero spelled correctly by auto-captions. WordPress.tv's caption quality guide should require speaker names checked against the program. - Polyglots: no direct action from this source; nearest relevance: none. - Accessibility: Emily's accessibility section (layout, restrooms, stairs, quiet space, outdoor space) is a template for WordCamp venue pages. Adopt it. - Meta: verify that WordPress.org's Git and SVN infrastructure has an off-platform restore path for all branches, and document who can run it. - CLI: no direct action from this source; nearest relevance: none. - Hosting: no direct action from this source; nearest relevance: "backup of a backup" as a stated default for customer repositories. - Tide: no direct action from this source; nearest relevance: none. - Openverse: no direct action from this source; nearest relevance: none. - Themes: no direct action from this source; nearest relevance: none. - Plugins: no direct action from this source; nearest relevance: a plugin author running a stale-branch cleanup action should read Olu's talk first. #### The WPOCC - Programs: a first-talk fellowship built around lightning-talk slots would fit the WP Open Community Collective's remit; the session shows the format works for first-time speakers. - Projects: no direct action from this source; nearest relevance: a documentation sprint using Emily's nine-section exercise. - Funding experiments: no direct action from this source; nearest relevance: microgrants for first-time speaker travel to lightning-talk slots. ### FAIR - Technical independence: Olu's content sovereignty is FAIR's argument in miniature. A project whose branches, history, and metadata live only on one hosted platform does not control them. Document the restore path off-platform. - Policies and covenants: add "recoverable without the primary host" as a covenant line for any FAIR-aligned operator. - Productized governance artifacts: add a scorecard row: "Has this project restored from a non-primary copy in the last 12 months? Y/N." ## 🧪 Hypotheses 1. Deleting a pull request's head branch on GitHub and then pushing new commits to a local branch of the same name produces commits that do not appear on the PR and triggers no warning in the GitHub UI. (Olu's specific claim at 00:22:00; directly reproducible.) 2. Event invitations that include a dedicated accessibility and logistics section receive a higher RSVP rate from attendees who report access needs than terse invitations, with no drop in overall RSVP rate. (Emily's "no harm done" claim.) 3. Documentation reviewers' notes taken during their first exposure to a product demo predict end-user confusion points better than notes taken after a second or later exposure. (Fraser's first-read instinct.) 4. Lightning-talk slots produce a higher share of first-time conference speakers than main-track slots at the same event. (3 of 6 here; Carolyn's own path.) 5. Teams where at least one member keeps a full local mirror of all remote branches restore from a bulk remote deletion in under one hour; teams without one take more than one working day. (Olu's recovery.) 6. Readers of long, sectioned documentation pages skip sections they don't need without measurable increase in abandonment compared with short pages, when section headings are descriptive. (Emily's skippable-completeness rule.) ## 📊 Methodology - **Design for H1.** Create a throwaway repository, open a PR from a feature branch, delete the branch on the remote via the API, then commit and push from a local clone that still has the branch. Record whether the PR shows the new commit, whether the push succeeds, and what the UI displays. Repeat with and without the "restore branch" button pressed. Check the result against GitHub Docs on deleting and restoring PR branches. Cost: 20 minutes. - **Design for H2.** Run two invitation formats across 10 events each, randomizing which format each host uses. Include an optional "access needs" field on the RSVP form. Compare RSVP rate overall and RSVP rate among respondents who fill the access-needs field. - **Design for H3.** Recruit 12 writers. Six watch a product demo once and write notes; six watch it twice before writing. Ship the resulting docs to 50 users each and count support questions per section. Compare which set of notes flagged the sections that generated questions. - **Design for H4.** Pull speaker rosters and bios from three Write the Docs events and three WordCamps. Code each speaker as first-time or not from bios and public talk histories. Compare first-time share between lightning-talk and main-track slots. - **Design for H5.** Tabletop exercise across 10 teams: announce a simulated bulk branch deletion, time the restore. Record whether a local mirror existed before the exercise. - **Design for H6.** A/B test two versions of five documentation pages: one full-length with sectioned headings, one trimmed. Measure scroll depth, time on page, and next-page navigation. Abandonment is leaving the docs domain within 10 seconds. - **Controls throughout.** Pre-register thresholds. For H1, the finding is binary and needs no statistics. For the rest, the talks provide n=1 anecdotes; anything above n=10 already improves on the source. ## ❓ Questions - Is Olu's ghost-commit behavior current GitHub behavior, or has the UI added a warning since her incident? - What did the re-optimized action change that made a months-stable cleanup delete everything? The talk never says whether it was a filter, a date window, or a permissions change. - Who is the "Kel" / "K" both Deepak and Fraser cite from the Monday morning session? The keynote is the likely source of "judgment is the skill" and "kill your darlings," and it would anchor both talks. - Does Emily's RSVP form capture access needs, and has the accessibility section changed who attends? - Fraser's "when in doubt, take it out" and Emily's "more is more" are in tension. What's the rule for deciding which applies: prose density versus coverage? - Are Florian's slides published anywhere? He points to links on images the audio can't convey. - What is the actual name of Fraser's employer ("Fawn Group" in captions), and does it have a public docs set to look at? - Does Write the Docs track how many lightning-talk speakers go on to give main-track talks? Carolyn's story suggests the pipeline exists; a number would confirm it. ## 🔮 Future research - **Recoverability as a governance metric.** Olu's incident suggests measuring "time to restore from a non-primary copy" alongside uptime. No open source governance scorecard tracks it. - **Skippable completeness versus cognitive load.** Emily asserts long sectioned docs cost skimmers nothing. Reading research on heading-guided skimming exists for web content but not for community-authored documentation with mixed audiences. Test it. - **First-exposure notes as a review method.** Fraser's editor instinct is a specific protocol (note every reaction on first viewing). It could become a docs review method with a name and a checklist; nobody has written it up. - **Automation with delete rights.** How many GitHub Actions in public repositories have branch-delete permissions and no approval gate? A scan of workflow files for `git push --delete` and the branches API would size the exposure. - **Lightning talks as speaker pipeline.** Carolyn, and three of six speakers here, are data points. A longitudinal look across conferences would show whether the format reliably converts first-timers into repeat speakers. - **Story as lossy compression.** Florian's framing is a hypothesis about knowledge transmission. Information theory has treated oral tradition this way before; connecting it to documentation practice is untested. ## 💡 Implications - **Practice.** Four rules you can apply this week: mirror every branch you can't afford to lose; write the long version and structure for skipping; note your first reactions in demos before familiarity dulls them; and cut prose, not coverage. - **Policy.** Any scheduled automation with delete rights needs an approval step and a scoped identity. Olu's dry-run report passed and the action still deleted everything, so a report is not a gate. - **Security.** Bulk branch deletion is a data-loss event, not just an inconvenience. Ghost commits that never reach a PR mean review can be bypassed silently. Treat delete permissions on CI tokens as sensitive. - **Education.** Deepak's "the machine is not the skill" is a curriculum principle: teach judgment (what to write, what to cut, what to verify) as the assessed outcome, not fluency of output. Emily's nine-section invitation is a ready-made teaching exercise. - **Funding models.** First-time speaker slots are cheap to fund and high-yield if Carolyn's path is typical. A five-minute stage plus travel support costs less than a keynote and produces future speakers. - **Content sovereignty.** Olu's term is the small-team version of what FAIR argues for the WordPress ecosystem: hosting is not ownership. The argument scales from one repository to a whole distribution network. ## 🧠 Learning styles assessment - **Reading level estimate.** Flesch-Kincaid Grade Level on the transcript body: approximately grade 7. Flesch Reading Ease near 75. The speech is conversational, first-person, and short-sentenced. Domain terms (GitHub Action, pull request, protected branch, permacomputing, the Three Laws) are used without definition, so effective comprehension for the Git segment assumes a developer audience; the rest is general-audience. #### 👀 Visual learners - Present: Florian's image-linked slides, Emily's four-page invitation shown on screen, Olu's implied action report, Fraser's timeline metaphor. Jonathan uses Post-it notes and no slides. - Improve: none of the visuals are in the description or the transcript. Publish Florian's and Emily's slides, and add a one-frame diagram of Olu's failure chain (action runs, branches deleted, local commit, ghost commit, no PR update). #### 🔊 Aural learners - Present: six distinct voices in 30 minutes with applause and laughter marking transitions. Fully narrated. - Improve: no chapter markers, so an audio-only listener can't jump between talks. Add timestamps to the description; the talk order is already listed there. #### 🗣️ Verbal learners - Present: strong aphorisms ("the machine is not the skill," "when in doubt, take it out," "more is more when it comes to accessibility information," "content sovereignty"). Fraser defines rough assembly and the three-act structure in one sentence each. - Improve: ghost commit, protected branch, and permacomputing go undefined. One clause each would cost 10 seconds. #### 🧩 Kinesthetic learners - Present: Emily's talk is a template you can fill in tonight. Olu's talk implies a reproducible experiment. - Improve: no repository, script, or invitation template is linked. A gist of the restore script and a copy of the nine-section invitation would turn two talks into exercises. #### 🧠 Logical learners - Present: Fraser's three-plus-one structure is explicit and numbered. Emily's nine sections are enumerated in order. Olu's post-mortem is sequential. - Improve: Deepak's argument skips a step (license is not skill). Jonathan's mapping lacks a criterion for which stories count. Both would land better with the missing premise stated. #### 🧑‍🤝‍🧑 Social learners - Present: Olu polls the room; Fraser polls the room; Carolyn asks the audience to warm up; every speaker invites hallway follow-up. - Improve: no discussion prompt or forum link in the description. Write the Docs has a Slack; naming the channel would extend the conversation. #### 🧘‍♀️ Solitary learners - Present: each talk is self-contained at five minutes; the whole is watchable in one sitting. - Improve: no reflection prompt. "Which branch could you not restore tomorrow?" and "Which section of your docs did you cut for length?" would convert two talks into self-checks. #### 🌿 Naturalistic learners - Present: Florian's permacomputing thread (degrowth, scarcity, maintenance) is the only explicit sustainability tie. - Improve: content sovereignty and "backup of a backup" have a resource-cost dimension nobody names. Local mirrors cost storage; the tradeoff is worth one sentence. ## 📊 Accessibility assessment #### Screen reader semantics - Present: YouTube supplies page landmarks and player controls. The description lists all six talks and the wrangler with pronouns, in reading order. - Improve: Emily's four-page invitation, Florian's slide links, and Olu's action report are shown and referenced ("you don't have to read all of this right now") but never read aloud. WCAG 2.2 AA 1.1.1 (Non-text Content) and 1.2.5 (Audio Description, Prerecorded) are unmet for those frames. #### Keyboard navigation - Present: the YouTube player is keyboard operable with visible focus. - Improve: nothing in the video adds barriers. 2.1.1 and 2.4.7 pass via the platform. The absence of chapters means keyboard users scrub rather than jump; 2.4.5 (Multiple Ways) is a soft gap addressed by description timestamps. #### Contrast and readability - Present: unverifiable from the transcript. Conference stage lighting and projected slides are the likely presentation. - Improve: Emily's dense invitation slide at conference-projector resolution risks failing 1.4.3 (Contrast Minimum) for remote viewers. Publish the document as text. #### Text resizing and reflow - Present: the player scales; captions resize independently. - Improve: any burned-in slide text fails 1.4.4 (Resize Text) and 1.4.10 (Reflow). Linked slides fix this at zero production cost. #### Media and captions - Present: auto-generated captions exist and are how this analysis was produced. - Improve: the captions misspell all six speaker names and the wrangler's, plus Schoonmaker, Kahn, Menke, Chatwin, Herzog, Asimov's title, and "sovereignty." That's a 1.2.2 (Captions, Prerecorded) failure in substance for every proper noun. The description already holds the correct names; a human caption pass for a 31-minute video is a few hours of work. #### Forms and interactions - Present: none in the video. Speakers give hallway and social-media contact routes verbally ("the bookworm John"). - Improve: put speaker handles and slide links in the description as selectable text. 3.3.2 (Labels or Instructions) applies to any RSVP form Emily's talk points to, not to the video. #### Standards reference - Primary gaps against WCAG 2.2 AA: **1.2.2** (inaccurate captions on every name), **1.2.5** (no audio description for slides shown but not read), **1.1.1** (slide content conveyed visually only), **1.4.4 / 1.4.10** (burned-in text). Chapter markers would help 2.4.5. The caption fix is the highest-impact and the cheapest. ## 🌍 DEIB assessment #### Cultural breadth - Present: speakers based in Germany and Norway; references span British comedy, American science fiction, Hollywood editing pairs, a German robotics startup, and a Drupal community. Fraser's Scottish-inflected and Olu's Nigerian-origin names are present but nobody's origin is stated, and this analysis doesn't assume it. - Improve: all cultural references are Anglophone (Monty Python, Asimov, Scorsese, Spielberg, Tarantino). One non-English storytelling or film tradition would widen Florian's and Fraser's arguments. #### Inclusive language - Present: the description gives the wrangler's pronouns (she/her). Emily's invitation explicitly covers partners, plus-ones, rest space, and illness precautions. Speakers use "everyone" and "folks" framing. - Improve: Deepak's "robots that are taking over the jobs apparently" is a joke that lands differently for warehouse workers than for a docs audience. Fraser's "sordid relationship with Claude" is a joke, harmless. Nothing exclusionary appears. #### Neurodiversity support - Present: five-minute chunks, one idea per talk, clear applause boundaries. Emily's talk models the point directly: quiet space, predictable structure, and everything written down so nobody has to ask. - Improve: no chapter markers and no linked slides. Crowd noise, laughter, and cross-talk in the captions add processing load. Timestamps in the description would help most. #### Representation - Present: six speakers, two women and four men, backgrounds in social work, filmmaking, film editing, startup leadership, and journalism (Carolyn). Three first-time conference speakers and one first-time physical-stage speaker. - Improve: the description credits the speakers by name; the captions don't. Correct captions are a representation issue when the mistakes cluster on non-Anglo names (Deepak becomes "Deepo," Olufunke is never captioned, Keddie becomes "Ketty"). #### Psychological safety - Present: Olu narrates her own mistake in public with full ownership ("Sorry for that") and reports a manager who praised the recovery. Carolyn frames Post-it notes and same-morning ideas as the spirit of the format. Applause and laughter are supportive throughout. - Improve: Carolyn's time-enforcement joke ("it will get weird") is friendly but could read as a threat to an anxious first-timer; a visible timer is the kinder mechanism. Otherwise this session is a model of a low-stakes stage. ## ✅ Logical validity #### Strong reasoning - **Analogical, 00:26:20 to 00:27:44.** Fraser's mapping of rough assembly to draft, cut to edit, and three-act structure to problem-steps-resolution is a tight analogy because both crafts share the constraint (a fixed audience attention budget) and the failure mode (unanswered questions). Strengthen by naming the disanalogy: docs are re-read and searched; films are watched once in order. - **Causal, 00:14:02.** Emily's argument that a web-hosted invitation lets her patch gaps as questions arrive is a valid mechanism (edit in place, no redistribution) with a concrete instance (the dog question). This is the strongest small argument in the session. - **Cost-asymmetry, 00:14:02.** "People who don't need this information will skip over it and no harm done" versus "people who do need this information probably need all of this information" is a decision under asymmetric cost: omission harms someone, inclusion harms nobody. Valid given the premise that skipping is free. Clarify by stating the premise. - **Post-mortem sequencing, 00:20:34 to 00:23:26.** Olu gives a clean causal chain: change, test, report, merge, symptom, diagnosis, mechanism, recovery, lesson. She also states the counterfactual (production down would have changed the outcome). This is good incident reasoning, and her conclusion (control your data off-platform) follows from the recovery mechanism (her local mirror). - **Self-limiting scope, 00:24:08.** "You always make mistakes, but always have a backup" concedes that the automation habit won't change and locates the fix in recoverability rather than prevention. That's a realistic, defensible move. #### Fallacies - **Weak analogy, 00:08:58.** "I had to wait for somebody to actually drive a forklift even though I could do everything other than this. But this is because somebody needs to have a license." A license is a regulatory credential, not a measure of skill; the analogy proves that gates exist, not that expertise is irreducible. *Fix:* replace the license with a case where he tried and failed at a skilled task, or state plainly that the argument is about institutional trust in credentials, which is a different (and defensible) claim. - **Hasty generalization, 00:09:42.** From one warehouse to "this applies to all of us" and to "documentation is something we can generate, please push back very hard." The conclusion may be right; the anecdote doesn't reach it. *Fix:* cite one documented failure of generated documentation, which would carry the argument on its own. - **Bare assertion / inevitability, 00:18:24.** "sci-fi is just technology that has not been invented yet" and "those will still at some point be invented into the technology that will inevitably play some wider conflict." Universal claim, no criterion for which fiction counts, and a certainty about future conflict. *Fix:* "some science fiction anticipates technology" plus the five examples he already gave. - **Hedged overgeneralization, 00:06:02.** "it speaks to us on a deeper like a common moral ground for all humans probably." Florian hedges ("probably"), so this is low severity, but "all humans" has no support. *Fix:* "a moral ground shared widely enough to sell tickets," which is what the evidence (box office) shows. - **Anecdotal evidence, 00:22:44 to 00:23:26.** Olu's lesson rests on one incident and one recovery. She frames it as a lesson rather than a law, so this is a limitation, not a fallacy in the strict sense. *Fix:* pair it with the GitHub Docs restore procedure so the audience has a second path that doesn't depend on a colleague's habits. - **Checked and not found.** No ad hominem, no false dilemma, no appeal to authority beyond Deepak's and Fraser's credit to a morning speaker (both are attributions, not arguments from authority). Emily's and Fraser's talks contain no fallacy this pass could name. #### Tone of voice - Warm, self-deprecating, and fast. Every speaker uses first person and contractions. Florian opens with a joke about his long password and calls his own theory ambitious. Deepak apologizes for his title. Emily thanks the community before she starts. Olu says "Sorry for that" mid-story and laughs at "everybody's branches." Fraser signs off with "wait for the sequel." Carolyn's wrangling register is comic-authoritative: she threatens to stand too close, praises Post-it notes, and tells the room she "never left." The applause and laughter markers in the captions appear in every talk. Nobody sells anything. The nearest thing to a hard edge is Deepak's "push back very hard," aimed at an argument rather than a person. ## 🛡️ DARVO analysis Not applicable to this source. DARVO (Deny, Attack, Reverse Victim and Offender) describes a response pattern to accusation, and no talk here is a response to one. The closest case is the inverse: Olu's post-mortem at 00:19:50 is a self-accusation in which she names herself as the cause ("I did it. I up something. Sorry for that"), describes the harm to colleagues, and credits her manager's response rather than contesting it. The title's "villain and hero" framing is a joke at her own expense, not a reversal. For completeness across the three prongs: **Deny:** nothing is denied; Olu volunteers the scope of the damage ("almost all the branches besides production") and the counterfactual that would have made it worse. **Attack:** no critic or accuser is attacked. Deepak's "push back very hard" targets a claim about generated documentation, not a person, and Fraser's Claude joke is aimed at the room, including himself. **Reverse victim/offender:** no such move. Olu is the offender by her own account and stays there. No emotional coercion or grievance rhetoric appears anywhere in the 30:53. ## 🎯 Conclusion The durable content of this session is four rules of practice, each carried by one speaker and each cheap to apply. Deepak Valagam's "the machine is not the skill" (00:09:42) is the principle to put in front of anyone proposing generated documentation, even though his forklift story doesn't prove it. Emily Axel's skippable-completeness rule (00:14:02), that the people who don't need a section skip it at no cost while the people who do need all of it, is a direct argument for longer, better-structured pages and for accessibility sections that run as long as they must. Moronfolu Olufunke's content sovereignty (00:23:26), learned by deleting every branch except production and restoring them from her own machine, is the small-team form of an argument the WordPress ecosystem is having about hosting versus ownership. Fraser Keddie's invisible-art observation (00:29:52) gives docs teams a success criterion: nobody noticed. The weaker material is still useful. Florian Stolzenhain's stories-as-lossy-compression frame and Jonathan Putnam's Asimov mapping are ways of seeing rather than claims to test, and Carolyn Stransky Vohr's curation makes the case that a five-minute stage with a same-morning submission window is a speaker pipeline: three of six speakers were first-timers, as she was in 2017. **Next steps for readers:** reproduce Olu's ghost-commit claim in a throwaway repository (20 minutes) and write down the result; list the branches, documents, and metadata your team could not restore if the hosting platform vanished tomorrow; and pick one documentation page where coverage was cut for length and restore it with headings that let skimmers skip. Get the speakers' names from the video description, not the captions. ## 🔄 Refactor Appendix — Monday Lightning Talks - Write the Docs Berlin 2026 > Density: 2-5 Literature atoms (H3 pad), 0-3 Permanent candidates (H4 pad). > Emitted: 5 Literature, 2 Permanent. Resonance filter applied: the sci-fi mapping > and the stories-as-compression frame were dropped as framings without a testable claim. > > Konik claim-header rules: present tense, declarative, ≤10 words, searchable, unique. > > Atomicity: one declarative sentence per atom. "and"/"also" → split. > > Tag pool: #bmf · #addendum · #xref · #articleseed · #storystem · #share · #aphorisms ### 📚 Literature Split Pad #### [H1] The machine is not the skill; judgment is > [!quote] [00:09:42] > Like I said, the machine is not the skill and as Kel said earlier today, uh the judgment is the skill **Context & use**: - 00:09:42, Deepak Valagam's closing line after the forklift story; he credits a Monday-morning speaker (captions: "Kel") for the second half. - The aphorism outlives the anecdote. Use it as the one-line answer to "can't we just generate the docs?" and pair it with a documented failure case, which the talk doesn't supply. **Tags**: #xref #aphorisms #share <!-- atomicity: 1 idea ✓ --> #### [H2] Readers who don't need a section skip it free > [!quote] [00:14:02] > People who don't need this information will skip over it and no harm done. **Context & use**: - 00:14:02, Emily Axel explaining why the accessibility section of her party invitation should be the longest. The sentence that follows in the transcript completes the asymmetry: the people who do need it need all of it. - This is a cost-asymmetry argument for coverage over brevity, and it resolves the apparent conflict with Fraser's "take it out": cut prose density, never cut who you serve. **Tags**: #xref #articleseed <!-- atomicity: 1 idea ✓ --> #### [H3] Hosted version control is not control of your data > [!quote] [00:23:26] > you need to always have control over your data always you need to be able to decide oh what if github goes down **Context & use**: - 00:23:26, Moronfolu Olufunke drawing the lesson from restoring every deleted branch off her own local mirror. - She names it "content sovereignty." It's the repository-scale version of FAIR's argument about distribution networks, and it converts into one measurable question: can you restore without the primary host? **Tags**: #xref #addendum <!-- atomicity: 1 idea ✓ --> #### [H4] Good editing and good documentation are invisible > [!quote] [00:29:52] > editing is often described as this sort of invisible art. And it makes me think of um of documentation **Context & use**: - 00:29:52, Fraser Keddie's bonus insight after polling the room on who notices editing (one hand, then a few). - Gives docs teams a success criterion that isn't praise: the reader felt the flow and never thought about the assembly. Useful for framing docs metrics around friction absence rather than page views. **Tags**: #xref #aphorisms <!-- atomicity: 1 idea ✓ --> #### [H5] Your first read is the only audience read > [!quote] [00:27:44] > you're watching it um objectively in the same way that the audience will be watching the film in the cinema for the first time. **Context & use**: - 00:27:44, Fraser Keddie on the editor's instincts during the first pass over raw footage, mapped to sitting in demos and product meetings as a docs writer. - A concrete review protocol hides in this: write down every reaction during first exposure, before familiarity erases the reader's confusion. Candidate for a named docs-review method. **Tags**: #xref #articleseed <!-- atomicity: 1 idea ✓ --> ### 🧠 Atomic Split Pad #### [P1] Judgment is the defensible documentation skill **Claim** (one sentence, Courtney's voice): The documentation skill worth hiring for, teaching, and defending against automation is judgment about what to write, cut, and verify, not fluency of output. **Evidence pointer**: [[#H1]] · [[#H5]] **Scope & conditions**: Holds when the output medium is cheap to produce and the cost of being wrong lands on the reader. Weakens for boilerplate where correctness is mechanically checkable (generated API reference from typed source) and judgment adds little. Deepak's forklift anecdote does not itself establish the claim; treat it as a memorable handle, not evidence. **Links**: - [[AI-assisted documentation]] — because applies-in deciding what generation is for - [[Every claim carries a fact]] — because extends the writing standard into hiring and review - [[Diátaxis]] — because contrasts on where judgment enters (structure choice versus sentence choice) <!-- atomicity: 1 idea ✓ --> #### [P2] Hosting is not ownership **Claim** (one sentence, Courtney's voice): A project controls its content only to the degree it can restore that content without the platform that hosts it. **Evidence pointer**: [[#H3]] **Scope & conditions**: Applies to repositories, documentation sites, plugin directories, and community archives alike. The test is operational, not legal: has anyone restored from a non-primary copy in the last year? Breaks down as a rule when the content is ephemeral by design (chat, drafts) and restore cost exceeds loss cost. **Links**: - [[FAIR technical independence]] — because extends from one repository to a distribution network - [[Backup of a backup]] — because evidence-for - [[Blameless post-mortem]] — because applies-in how the lesson was surfaced publicly <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Reproduce the ghost-commit claim: delete a PR's head branch on GitHub, push to the same-named local branch, record what the PR and UI show — https://www.youtube.com/watch?v=NSAfpSv1e4Y&t=1320 #fleeting #inbox - [F2] Identify the Monday-morning speaker the captions call "Kel" / "K" who said "the judgment is the skill" and "kill your darlings"; both Deepak and Fraser cite them — WTD Berlin 2026 program #fleeting #inbox - [F3] Confirm Fraser Keddie's employer ("Fawn Group" in captions; probably Fonn Group, Bergen) and whether Florian's slides are published anywhere — video description carries neither #fleeting #inbox - [F4] Article seed: "Cut prose, not coverage" reconciling Emily's more-is-more accessibility rule with Fraser's when-in-doubt-take-it-out #fleeting #inbox #articleseed - [F5] Article seed: the first-read review protocol (note every reaction during first exposure to a demo, then write) as a named docs-review method #fleeting #inbox #articleseed - [F6] Emily's nine-section invitation as a WordCamp venue-page and training exercise template: when, RSVP, expect, where, bring/don't bring, stay, health, accessibility, contact #fleeting #inbox - [F7] Count first-time speakers in lightning-talk slots versus main track across the next three WordCamps and WTD events I attend or watch #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Reproduce Olu's ghost-commit scenario in a throwaway GitHub repo and write down what the PR shows — 20 min, docs strategy research #things - [ ] List every branch, doc, and metadata set my team could not restore without GitHub, and name who can run the restore — content sovereignty audit #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/Monday Lightning Talks - Write the Docs Berlin 2026.md|Monday Lightning Talks - Write the Docs Berlin 2026 by Write the Docs]] ## Highlights - [00:08:58] skills, right? So when I was at this warehouse, I um I had to wait for somebody to actually drive a forklift even though I could do everything other than this. But this is because somebody needs to have a license to operate a operate a forklift. And in case it's obvious, it's not obvious, I cannot drive a forklift, right? Um so the learning there was the expertise is very hard gained and like uh it is deep and you cannot replace this with just knowledge or prompting or whatever output that AI gives you and this applies to all of us. So when there is somebody saying like documentation is something we can generate, please push [00:09:42] back very hard uh because this expertise that you have received is very hard earned and it is not just replaceable by um by AI and the fact is the machine is not the skill right the machine is not the skill. This is the one thing I I want everyone to know and uh it's a very hard le learn lesson for me. Like I said, the machine is not the skill and as Kel said earlier today, uh the judgment is the skill and uh knowing where your judgment is valuable and uh sharpening on that is exactly what should be the first stop of your moonshot aims. And uh yeah, that's what I wanted to share today. Thank you. ([View Highlight](https://read.readwise.io/read/01m37bcd26z23xj5msm63xtgsj)) - Note: [H1] The machine is not the skill; judgment is - Tags: atom/literature/h1 - [00:13:19] address, how do I get there? Public transportation information is very helpful. Parking information is very helpful. Anything people will need to know about how to actually arrive to this wonderful event. What to bring? Do you want people to bring food? Do you not want people to bring food? I tend to like to cater my own parties. Some people like to do a potluck or bring and share. It's very important to know that. Do people need to bring drinks? Can people bring friends, partners, randos you've never met? Can people bring pets, etc.? Um, and also what not to bring. Maybe someone's coming who has a deathly peanut allergy, so please don't bring peanut butter cookies. Uh, someone asked me, "Can I bring my dog?" And I realized I'd not put this on the invitation. And [00:14:02] because the invitation is hosted somewhere on the web, I can edit it and then I can update it if people ask questions that I didn't cover. And no, I'm sorry, you can't bring your dog. Not to my party anyway. Where to stay? Are you hosting people if they're coming from far away? Do you have accommodation? Are there any hotels nearby? All that kind of information is really great to provide. Anything you want to say about COVID and flu precautions. Do you want people to mask? Do you want people to test ahead of time? And a nice reminder that if you have virus symptoms, please just stay home and we'll hopefully see you next time. Now, this is the piece that I think should be the longest. People who don't need this information will skip over it and no harm done. Who cares? People who do need this information probably need all of this information. So, this is ([View Highlight](https://read.readwise.io/read/01m37bcfemrh569tyjb6vt9v6h)) - Note: [H2] Readers who don't need a section skip it free - Tags: atom/literature/h2 - [00:23:26] things you know it's what I do. But it's taught me that you know even when gate will give you this version control and everything associated to it you need to always have control over your data always you need to be able to decide oh what if github goes down what if it was g fault and all these branches were deleted and production went down do you have your content again can you like you know bring it back can you put it to a different platform and this was what I learned for my you know after my very terrible mistake I did not lose my job my manager said oh that was nice I liked how you were very proactive you solved it on time, you know, fantastic man. So, yeah. Yeah. Yeah. I'm sorry if I had brought down production. It have been a very different conversation, but yeah, he was like, "Yeah, very good that [00:24:08] you're able to get on it on time." But I'm just going to leave you to this that, you know, no matter what you do, please think about content sovereignity and [snorts] try to always have a backup of a backup and you know, you always make mistakes, but always have a backup and you can always find me to talk about content sovereignity. You know, it makes me very excited. Thank you so much for listening to me today. >> [applause] >> All right. Thank you, Olu. So, there are a lot of way like different shapes that lightning talks can take. So, for example, I would have called that last one kind of a post-mortem. We got to like, you know, learn from our mistakes, share knowledge so that maybe other people can avoid or at least feel a ([View Highlight](https://read.readwise.io/read/01m37bchvwrq3t8rr0p8eds5nf)) - Note: [H3] Hosted version control is not control of your data - Tags: atom/literature/h3 - [00:27:44] Um the second thing is uh instincts. um as an editor when you're looking at those hundreds of hours of footage uh for the very first time you have a very unique perspective and that is that you're watching it um objectively in the same way that the audience will be watching the film in the cinema for the first time. So you're not influenced by anything. So it's really important that you make notes and you jot down anything that stands out to you like emotions, things that make you laugh or cry, looks from actors. uh and I think this is similar to documentation and that you sit in demos or product meetings or uh talk with developers and your instincts and experience tell you what's important to write down because then you know [00:28:28] you're going to use it later and that's simply because of the fact that you know your audience and you're thinking uh from the terms of the end user or the reader. Uh the last thing is relationships. Um these are really important. Uh directors often work with an editor throughout their whole career. Not many people uh know that. So Thelma Skumarker for example has edited most of Martin Scorsese's films throughout her whole career. Uh Michael Khan for Steven Spielberg, Sally Manker for Quentyn Tarantino and there are many others. So you build those relationships with yourmemes. There's long-term relationships. you find the people that you trust and that ([View Highlight](https://read.readwise.io/read/01m37bcqjqfhzrtgfkrr19s5r2)) - Note: [H5] Your first read is the only audience read - Tags: atom/literature/h3 - [00:29:10] you can uh you can go to you know you can talk to although I think many of us are having a rather sorted relationship with Claude at the moment who's become our kind of theme but you know whatever floats your boat. Um the more that you build those relationships uh the more you get out of them. So those are my kind of three insights. I think I've got time for like one bonus insight which I feel is kind of relevant for documentarians. Uh and this is something that I've heard from editors many times. Um, uh, it's definitely relevant here. Some, someone was talking about this yesterday. Um, how many people have watched a film and actually thought about the editing? Uh, it's usually very few. Oh, one person. Let's go. Let's go. There's a few people. Okay. That's good. [00:29:52] But generally, you you'll notice the cinematography, the music, the actors performances. Uh, fantastic. But you don't usually notice the editing. Uh, [snorts] and editing is often described as this sort of invisible art. And it makes me think of um of documentation because editing is most successful when the audience they appreciate the sort of smooth flow, the way everything fits together and the experience, but not necessarily how it was put together. And that's very much like editing. Yeah, I think that's about it. If you want to hear more, you have to wait for the sequel. [applause] Incredible. Thank you. Can we please get ([View Highlight](https://read.readwise.io/read/01m37bcm2raaevz16n28b55e3p)) - Note: [H4] Good editing and good documentation are invisible - Tags: atom/literature/h3