home/resources
drag or scroll sideways along the spines, then pick one
the shelf · updated july 2026
The sources behind my work in conversation design, documentation, UX writing, and testing content with research. Each one tagged, annotated, and traceable to a decision.
why this page exists
Every recommendation I make traces back to something here, so a decision can be reviewed by anyone.
Core changed a decision I make in production. Shelf is canon I open when practice runs out.
The craft layer: how a system takes turns, repairs itself, and holds a persona. The first four are where the discipline actually lives.
Erika Hall · A Book Apart · 2018
Conversation is the original interface, and most software is a poor conversationalist. Hall turns cooperative principles into design constraints.
What it isShort and argumentative: why humans read every exchange as social.
The vocabulary that aligns a mixed team in one meeting.
Diana Deibel & Rebecca Evanhoe · Rosenfeld Media · 2021
The most practical book here: persona, sample dialogs, error handling, with scripts concrete enough to lift into a spec.
What it isA practitioner handbook: research, persona, sample dialogs, testing, ethics.
Write sample dialogs before flows. A persona is a set of constraints.
Cathy Pearl · O'Reilly · 2016
The canonical voice text. Pearl's confidence thresholds became my tiers for when an agent acts, confirms, or hands off.
What it isHow recognition fails, and how an interaction recovers and keeps its place.
Graduated confirmation: act, confirm, or hand off is a design decision.
Robert J. Moore & Raphael Arar · ACM Books · 2019
IBM Research's Natural Conversation Framework. Its expandable sequences give every exchange room for the repair users actually need.
What it isA design framework built on conversation analysis, from IBM Research.
Design the expansion first. Users interrupt and change their minds.
Michael McTear & Marina Ashurkina · Apress · 2024
McTear is the academic authority here. His evaluation taxonomy sits under every eval plan I write.
What it isWhat large language models changed, from the field's textbook author.
An evaluation taxonomy that names what kind of evidence a study produces.
Victor Dibia · self-published · 2025
Dibia builds AutoGen at Microsoft Research. He names the reliability problem a generator-and-reviewer architecture creates for the person waiting.
What it isOrchestration patterns, and where autonomy multiplies failure.
More agents means more places to be wrong. Design what the person sees meanwhile.
Mezba Uddin · Apress · 2026
Tiered fallback, persona protection, and A/B testing on tone, shown at the level where they survive a release.
What it isA manual for the parts tool books skip: not understanding, persona drift.
Tiered fallback: a nudge, a guided menu, then a graceful exit.
Joseph Weizenbaum · Communications of the ACM · 1966 · read it
The original. Weizenbaum shows how little machinery it takes to be read as understanding, and then spends the rest of his life alarmed by it.
What it isThe DOCTOR script, keyword ranking and transformation rules, described by the person who wrote them.
Scripted answers are a design position, not a shortcut. It is why the assistant on this site generates nothing.
Sacks, Schegloff & Jefferson · Language · 1974
Where turn-taking was first described from real recordings. The finding with teeth: silence is a poor end-of-turn signal.
What it isThe founding paper of conversation analysis, from transcripts of real talk.
Cutting in at silence breaks a norm users feel as rudeness.
Herbert H. Clark & Susan E. Brennan · 1991
The theory underneath every confirmation step, and why a well-placed acknowledgement costs less than the misunderstanding it prevents.
What it isHow two parties establish that they have understood each other.
A budget: every acknowledgement costs attention, every skipped one risks a repair.
H. P. Grice · 1975
Quantity, quality, relation, manner. I run the four maxims as a lint pass over system prompts.
What it isHow people mean more than they say.
Four questions for any generated reply. Most bad agent copy fails the first.
Langevin et al. · CHI · 2021
Four evaluators, one day, a defensible review before you have a single user. The cheapest quality gate available.
What it isHeuristic evaluation adapted for agents, where general usability heuristics go quiet.
A method with a citation, which turns an hour of inspection into evidence.
Research methods applied to words: comprehension, task success, and dialog quality. This is the part of my practice people are most surprised is measurable, so the shelf leans towards method and statistics.
Marilyn Walker et al. · ACL · 1997
Task success and dialog cost in one model. Thirty years on, still the cleanest way to prove one design beats another.
What it isSeparates what the user achieved from what it cost them.
Fewer turns is not automatically better. This shows when it actually helped.
Jeff Sauro & James R. Lewis · Morgan Kaufmann
Which test, which sample size, which confidence interval. It lets me put numbers on a content change and state their limits.
What it isApplied statistics for small studies: which test, which interval, how many people.
Honest confidence. Credibility rests on stating what a finding cannot prove.
Erika Hall · A Book Apart
Research sized for teams that ship. I hand the evaluative chapter to anyone who says there is no time to test.
What it isResearch inside real constraints, including the politics of asking.
Name the smallest useful study. That is usually enough to get one approved.
Jeffrey Rubin & Dana Chisnell · Wiley
The procedural baseline. When a study has to hold up to scrutiny, this says how it should have been run.
What it isThe full procedure, from writing tasks that do not lead to reporting.
Task wording. Using the interface's own words tests the label, not the concept.
Steve Portigal · Rosenfeld Media
Interviewing as a technique with failure modes. The material on silence improved my moderation more than any template.
What it isThe craft of the interview itself: rapport, the awkward moment, the tell.
Waiting. The useful sentence arrives after the pause you wanted to fill.
Tom Tullis & Bill Albert · Morgan Kaufmann
The metric catalogue. Useful when a stakeholder asks for one number and deserves a better answer.
What it isWhat can be measured, how each metric behaves, which survive as benchmarks.
Restraint in combining. One score hides the disagreement, which is the finding.
Mike Kuniavsky · Morgan Kaufmann
The wide survey of methods, strongest on choosing between them. For when the research question is still soft.
What it isA broad catalogue of methods, for teams deciding what to run.
Most weak studies are competent executions of the wrong method.
ISTQB · free download · istqb.org
Testing AI is a statistical problem: one passing example proves nothing. The eleven QA flows I run for hallucination, bias, consistency and security are built on these syllabi.
What it isFree syllabi on why probabilistic systems break pass-and-fail testing.
Language QA already speaks, which makes a bigger eval set easy to argue.
W3C · free · w3.org
Read as a content standard, it sets testable rules for link text, errors, and headings. Legible in an afternoon.
What it isAccessibility guidelines as testable criteria. A large share govern words.
A checklist: link text that stands alone, errors that say how to fix.
My first discipline, and the one changing fastest. The frameworks here settle arguments about structure; the newer entries deal with documentation that machines read as well as people.
Jared Bhatti et al. · Apress · 2nd edition
The canonical book of the profession. Two ideas earn their keep weekly: friction logs, and a content type taxonomy.
What it isThe full lifecycle, audience to maintenance, with the AI-assisted parts added.
The friction log turns documentation feedback into product evidence.
Daniele Procida · free · diataxis.fr
Tutorial, how-to, reference, explanation. It ends most arguments about where a page belongs.
What it isTwo axes, four modes, genuinely different jobs.
A shared referee. Once the mode is agreed, content arguments resolve themselves.
Diana Lakatos · Apress · 2023
Documentation strategy end to end. Its chapter on measures of success is where docs metrics stop being vanity numbers.
What it isDesign thinking for developer docs, including the operational side.
Measures tied to purpose: whether the reader finished the task the page existed for.
Andrew Etter
One sitting, and the clearest statement of docs-as-code: lightweight markup, version control, docs beside the thing they describe.
What it isA brief essay arguing docs deserve the tools code already uses.
The fastest way to get docs into the pull request workflow.
Christopher Gales & the Splunk documentation team
A working team writing about running a working team, including the politics of owning content nobody else wants.
What it isEssays from an in-house team on process, structure, and standing.
Most documentation problems are staffing and review-cycle problems in costume.
open proposal · free · llmstxt.org
A convention for telling language models what a site contains. I implemented it here, which is how you find where it is thin.
What it isA markdown file giving a model a curated map of your content.
Writing for a reader and for retrieval pull the same direction more often than teams expect.
Google · free · developers.google.com
The default I inherit on most teams. Its value is being decided: a house style that already answered the small questions.
What it isA complete searchable guide: terminology, formatting, code samples, tone.
Speed. It moves the argument from what is correct to what we agreed.
Microsoft · free · learn.microsoft.com
learn.microsoft.com/style-guide
Stronger than Google's on voice, bias-free language, and the vocabulary of AI features. The better reference when the product speaks.
What it isAimed at product experience as much as docs: voice, accessibility, inclusive terms.
The most concrete published position on describing what a model did.
Robin S. Adams & Junaid A. Siddiqui (eds.) · Purdue University Press
One corpus of design reviews read through many lenses. It gives review a literature, which matters when your assistant takes part in one.
What it isMany teams analysing the same recorded design reviews, each through one lens.
It gives review the standing of an object of study.
Sharon Burton (ed.) · XML Press · 2026
Fifty years of the profession told by the women who built it, from typewriters to touchscreens. The subtitle is the method: a history by the women who did the work.
What it isA collection of first-hand accounts of careers in technical communication across five decades of tooling change.
Every tool panic in this field has a precedent, and somebody wrote down how it went. I stand on that record when I write the voice of an agent.
Words inside the interface, where the constraint is the flow.
Michael J. Metts & Andy Welfle · Rosenfeld Media · 2020
Writers belong where flows are decided. Its material on error states is what I quote most often in reviews.
What it isWords as a design material: process, collaboration, accessibility, responsive systems.
Timing. Copy written after the flow can only label someone else's decisions.
Torrey Podmajersky · O'Reilly · 2019
Voice charts. The artefact that makes tone reviewable by people who are not writers, which is what gets it enforced.
What it isA method for copy you can defend: voice, patterns, then measurement.
The voice chart makes tone something a team can review.
Sarah Winters · Content Design London
Answering a user need with the format the evidence supports, which sometimes means a table and sometimes means deleting the page.
What it isThe founding statement, from public-sector work where people have no alternative.
Start from the user need in the user's words. Format is what the evidence decides.
Nicole Fenton & Kate Kiefer Lee · New Riders · 2014
Voice and tone for people having a hard day. It ages well because it is about judgement.
What it isWriting for the web with care: voice, tone, difficult messages.
The separation of voice from tone, which keeps a brand consistent across registers.
Janice (Ginny) Redish · Morgan Kaufmann
Writing for people who scan, backed by usability evidence. Its before-and-after rewrites are the most convincing teaching device I know.
What it isWriting for people who arrive with a task, from a plain-language authority.
The before-and-after format converts stakeholders faster than any stated principle.
Eric Meyer & Sara Wachter-Boettcher · A Book Apart
Stress cases in place of edge cases. The person reading your error message is often having the worst day of their year.
What it isDesigning for people in crisis, written after the authors met those cases.
The rename. A stress case sounds like a person, and gets prioritised.
Linguistics and HCI: the layer I open when a project asks a question that practice cannot answer. These reward being read slowly, once.
Bing Liu · Cambridge University Press · 2015
The reference text. Its chapter on mining intentions is the most rigorous treatment of intent I have found.
What it isThe standard survey, with chapters on intention and deceptive opinion.
Two cases I design for: frustration shaped like a request, and a third party's want.
Herbert H. Clark · Cambridge University Press · 1996
Language as joint action. The full theory behind grounding, for when a design argument needs it.
What it isLanguage use as something two people do together, in real time.
A reply is a contribution, so the question is what it lets you do next.
Peter Auer (ed.) · Routledge
Repair strategies and filler words are language dependent, so a fallback tuned in English quietly stops working elsewhere.
What it isHow bilingual speakers move between languages, and what that accomplishes.
A conversational product does not localise by translating its strings.
Zellig Harris · University of Chicago Press · 1951
The distributional hypothesis, written in 1951. The direct intellectual ancestor of embeddings.
What it isDescribing a language from distribution alone, decades before the computation existed.
It explains why a linguist reads a vector space as familiar territory.
Card, Moran & Newell · 1983
Where GOMS comes from. The empirical floor under any claim about cognitive load, including mine.
What it isThe book that made HCI quantitative, modelling the user as an information processor.
Predict before you measure. It catches a failing design before anybody builds it.
Bill Moggridge · MIT Press · 2007
Oral history from the people who built the field. Read for how long the current arguments have been running.
What it isInterviews with the designers behind the mouse, the PC, early mobile.
Perspective. The now-obvious decisions were contested, which is worth remembering.
Rogers, Sharp & Preece · Wiley · 6th edition
The textbook. A fair check on whether a study design is standard or improvised.
What it isThe standard survey: cognition, data gathering, prototyping, evaluation.
Checking a plan against the textbook finds the step I was about to skip.
Nothing here matches that combination yet.
This list is revisited weekly, and it grows the way it was built: a title enters after it changes a decision, and leaves when I stop reaching for it. If something belongs here and is missing, or you want the argument for why one of these earned its place, write to me.
The practice these feed into is documented in the case studies on the work.
Not craft, but the conditions craft needs: how teams are shaped, how safe it is to speak, and why structure outlasts intent.
Erin Meyer · PublicAffairs · 2014
How feedback and decisions actually travel across cultures. It matters the moment a team spans several of them.
What it isA map of how cultures differ on feedback, trust, and decisions.
Why the same sentence lands as blunt in one office and vague in another.
Jason Fried & David Heinemeier Hansson · Harper Business · 2018
The case for calm work: fewer meetings, fewer hours, less manufactured urgency.
What it isAn argument for calm, focused work inside chaotic organisations.
Permission to treat constant urgency as a defect rather than a standard.
Amy C. Edmondson · Wiley · 2018
Psychological safety, with the research behind it. Teams do not report what they do not feel safe reporting.
What it isThe research case for psychological safety, and how to build it.
Nobody flags the problem they can see until it is safe to say it.
Matthew Skelton & Manuel Pais · IT Revolution · 2019
Team shape decides how feedback flows. A structural problem is rarely solved by trying harder.
What it isFour team types and three interaction modes for designing structure.
Autonomy and feedback flow are designed deliberately, not hoped for.
Donella H. Meadows · Chelsea Green · 2008
Feedback loops and structure shape behaviour long before anyone's intentions do.
What it isA primer on stocks, flows, feedback loops, and leverage points.
Structure explains more behaviour than intention. Change the loop, not the person.
Public systems I study for how they hold voice, content, and components together. Read them as worked examples, not as rules.
Paysafe
A payments product where precision in tone is a safety feature, not a style preference.
What it isPaysafe's content and visual system for a financial product.
Precision is what reads as trustworthy once money is involved.
Slack
Personality that stays usable. The reference for having a sense of humour without costing clarity.
What it isVoice and copy guidelines with a distinct personality.
Playful only survives when the copy underneath it is already clear.
Zendesk
Voice and tone tied to interaction design rather than parked in a separate document.
What it isA combined voice, tone, and interaction system.
Tone documented beside the pattern is tone that actually gets used.
Wise
Inclusive, accessible patterns in the content as much as in the visuals.
What it isContent and design patterns built around accessibility.
Accessibility written into the pattern, not bolted on at review.
NHS
Public service writing where the reader has no alternative and the stakes are health.
What it isContent standards and patterns for public healthcare services.
Plain language treated as a clinical safety measure.
Girlguiding
Values-led messaging carried by bold design, with neither one apologising for the other.
What it isA brand and content system led by organisational values.
Stated values give tone something to be accountable to.
Atlassian
UX writing embedded in the components themselves, which is where guidance gets followed.
What it isA design system with voice and content guidance inside each component.
Guidance attached to a component outlives guidance kept in a style guide.
Amplitude
One editorial tone holding product, docs, and marketing together.
What it isAn editorial voice system spanning several channels.
A single editorial standard is what stops channels drifting apart.
DHL
Iconography and language built for readers who do not share a first language.
What it isA global system for icons and plain language across markets.
Designing for translation from the start beats localising afterwards.
Who a system serves, who it can be turned against, and who gets to imagine what comes next. One book for the argument, three decks for the table: these are used with a team, not read alone.
Sasha Costanza-Chock · MIT Press · 2020
Community-Led Practices to Build the Worlds We Need. It moves the question from whether a design is good to who it distributes benefit and harm to, and who was in the room.
What it isA framework and a set of principles for design led by the communities that bear its consequences.
Two questions I now ask of a conversation design: who is this voice comfortable for, and who was never consulted about it.
Shruthi Sai Chivukula, Shikha Mehta, Colin M. Gray & Ritika Gairola · translate-ethics
Twelve cards that pair a manipulative role with a value-centered one, so a team plays out both intentions instead of assuming its own are good.
What it isA role-play method for surfacing designer intentions during ideation, evaluation and ethical dialogue.
Play the anti-hero against your own flow and it names, out loud, the pressure your copy could apply.
Daylight Security Research Lab, UC Berkeley · Nick Merrill, Narmin Mammadsoy, Cristián Maturana Álamos, Inderpal K., Akhilesh Pandita, Lily Bhattacharjee, Jasmine Z., Janani Sridhar
A role-playing guide that asks who your adversaries are, what they want, and what they would be willing to go through to get it. Threat modelling written for designers.
What it isAn improvisational game that builds personas for attackers, with description, motivation and capability.
A guarded topic is only guarded once somebody has played the person who wants around it.
Christina N. Harrington & Jennifer S. Roberts, with Kirsten Bray & N'Deye Diakhate · buildingutopiadeck.com
An Afrofuturist tool that uses speculative design to imagine new worlds and co-design better futures, rooted in justice and community visioning.
What it isA deck and futuring guidebook for community-led speculation, designed with and for Black and Indigenous communities.
Speculation has a method, and blank cards a participant fills in beat a future written for them.
