I re­flect about the re­cent books I have read and the things I (and you) can learn from them.

I want you to read that ti­tle as lit­er­ally as pos­si­ble. Harper is now ca­pa­ble of evo­lu­tion.

I’ve given dozens of talks over the years, yet I still feel like I have much to learn about pub­lic speak­ing. Preparing for my up­com­ing talk at WordCamp US has me stop­ping to re­flect on what I al­ready know.

I’ve got some ex­cit­ing new work I’m prepar­ing to dis­cuss later this week (hopefully). In the mean­time, I’ve col­lected some re­cent sight­ings of Harper in the wild.

Grammar check­ing can be cum­ber­some, es­pe­cially when its sloth gets in the way of your think­ing. That’s part of why so many pre­fer Harper over Grammarly or LanguageTool.

I was re­cently en­cour­aged to sub­mit a talk pro­posal to the AI track at WordCamp US. I was more than happy to oblige, and to my sur­prise, I got in.

It’s not easy, but I think it’s one of the best habits I’ve ever built.

Expression rules were de­signed for a cer­tain kind of er­ror. But there are many gram­mat­i­cal or lay­out rules which hard dif­fi­cult or down­right im­pos­si­ble to de­fine with an ex­pres­sion.

The end re­sult is a portable neural net­work that achieves ~95% ac­cu­racy on gram­mat­i­cally cor­rect text.

In an age where large lan­guage mod­els have ex­cep­tional con­text win­dows, it’s easy to fall into the trap of be­liev­ing a sys­tem is sim­pler than it is.

This ar­ti­cle guides con­trib­u­tors on writ­ing a Harper ex­pres­sion-based rule. It cov­ers defin­ing the rule’s ex­pres­sion, writ­ing test cases, and im­ple­ment­ing the match_to_lint func­tion for user sug­ges­tions.

This post ex­plains how to add sim­ple phrase cor­rec­tions to the gram­mar checker Harper—a be­gin­ner-friendly way to ad­dress com­mon, con­text-in­de­pen­dent er­rors and con­tribute to its gram­mar check­ing ca­pa­bil­i­ties.

This ar­ti­cle in­tro­duces con­tribut­ing to the open-source gram­mar checker Harper by writ­ing gram­mat­i­cal rules, out­lin­ing dif­fer­ent ap­proaches for var­i­ous skill lev­els.

A guide for adding a new pro­gram­ming lan­guage to the Harper lan­guage server.

The ti­tle of this post is some­what mis­lead­ing. Local-first soft­ware rarely needs to be scaled at all.

Failing to ac­count for this re­al­ity can slow down de­vel­op­ment and dis­suade con­trib­u­tors from stick­ing around.

It has been a slow cou­ple of months for my read­ing habit. Although, now that I’ve said that out loud, I re­al­ize I say that every time.

There’s a great deal of over­lap be­tween those who use Harper and those who use Firefox.

You know—that aw­ful (but strangely sat­is­fy­ing) space where the ma­jor­ity of your com­mits are small pedan­tic ed­its that don’t change the end-user-ex­pe­ri­ence at all?

The most com­mon type of ma­chine learn­ing out there takes the form of some kind of neural net­work. Inspired by how our own brains work, these sys­tems act as func­tion ap­prox­i­ma­tions. They are great, but they come with a few key pit­falls.

Harper is cur­rently un­der­go­ing some pretty rad­i­cal changes when it comes to its lan­guage analy­sis. These im­prove­ments will im­prove the out­put of our ex­ist­ing rule en­gine, in ad­di­tion to mak­ing en­tirely new cor­rec­tions pos­si­ble.

After re­ceiv­ing some feed­back re­lated to the sug­ges­tion box’s vi­sual un­pleas­ant­ness and dif­fi­culty to un­der­stand in­tu­itively, I’ve started mak­ing some mod­i­fi­ca­tions. Nothing dras­tic—I don’t want to con­fuse ex­ist­ing users.

SpellBolt is a Firefox ex­ten­sion that de­liv­ers Harper’s fan­tas­tic gram­mar check­ing straight into your Firefox browser.

There are a great va­ri­ety text ed­i­tors on the web, each used in a dif­fer­ent con­text. That’s a beau­ti­ful thing, but un­for­tu­nately our users ex­pect Harper to work in all of these places seam­lessly.

If there is one thing o3 is use­ful for, it’s find­ing canon­i­cal doc­u­men­ta­tion for com­plex sub­jects.

When our users in­stall Harper, they should ex­pect it to work any­where they do.

Spend more time on the in­tro­duc­tion than any­thing else.

Advice you can give your mom (or other non­tech­ni­cal per­son) for how to use ChatGPT wisely and ef­fec­tively.

An up­date on the long-awaited Harper Chrome ex­ten­sion, its cur­rent state, and what to ex­pect next.

Reflecting on the clar­ity pho­tog­ra­phy brings.

Reflecting on lessons learned from han­dling English edge cases in de­vel­op­ing Harper’s Chrome ex­ten­sion.

I was hon­ored to be a guest on Linkarzu’s YouTube se­ries.

Four years in Rust + WebAssembly taught me these 3 lethal foot­guns—don’t learn them the hard way.

Two months does not mean twice as many.

Reach fur­ther each day. The min­i­mum ef­fec­tive dose.

Don’t be­lieve the click­bait.

A new ap­proach to false-pos­i­tives.

A boon if you com­ment your code.

Your code is al­ways wel­come.

Of course it’s that easy.

🎉

Spoiler: there aren’t many.

The morn­ings are for deep work.

Wisdom from my grand­mother.

Bonus: why you need to do it too.

You might not ac­tu­ally need it.

I might be start­ing to sound like a bro­ken record.

Chronically un­der­rated, chron­i­cally over-pre­scribed

Where did Harper come from?

How I pre­view my Markdown quickly and re­li­ably.

It’s eas­ier than you think.

How I in­tend to live bet­ter.

It al­most feels wrong. Maybe it is.

Back in my day, we used math for au­to­com­plete.

How I built a soft­ware ren­der en­gine from scratch, and you can too.

It saved me enough time, I had some to share about it.

It did­n’t work for me, and if you read­ing this, it prob­a­bly won’t work for you ei­ther.

An ex­per­i­ment on how to live in a seem­ingly hope­less world.

A pro­posal.

Or: why do­ing things at the last minute ac­tu­ally saves time. I talk about the im­por­tance of shock­ingly fast it­er­a­tion cy­cles and lean man­u­fac­tur­ing.

We look at sev­eral in­ter­est­ing ways com­put­ers gen­er­ate ran­dom num­bers. It may fas­ci­nate you to know that some meth­ods are not truly ran­dom, but only an ap­prox­i­mate.

I built a Wordle solver in Rust to beat my grandma. Follow the jour­ney of how I did, and how I failed.

We go through step-by-step how to build a Discord bot us­ing the Rust Serenity Framework.

A key part of Rust is far bet­ter than what JavaScript has to of­fer.

In which I go through the process of de­sign­ing, build­ing, and test­ing a vec­tor graphic pen plot­ter.