DEV Community: Céline

How Anthropic Replaced Git with Scion Using Claude

Céline — Tue, 20 Oct 2026 14:04:18 +0000

On 13 October 2026, Anthropic published Redesigning source control at Anthropic. Its monorepo no longer runs on Git. A team of seven built an in-house version control system called Scion, plus a code review tool called Antique, and finished the migration in five months, with Claude writing a lot of the code.

The Hacker News thread passed 895 points and 1,000 comments. I read the post, the two Mononoke posts that followed it, and the biggest sub-threads. Here's how Scion works, where Claude actually fit, the credit fight, and what plain Git already gives you.

TL;DR

Anthropic went from about 2,500 merged changes a day (May 2026 average) to a record 30,401 (9 September), and says Claude's code is why. Another Anthropic page says Claude authored more than 80% of the code it merged as of May 2026.
Scion moves the work to the server: commits, merges, rebases, blame and history. Clients are thin and assumed to be online.
Anthropic reports a 200x smaller CPU and memory footprint than its old Git mirror fleet and a 25% CPU capacity win in CI. Its text says checkouts got two orders of magnitude faster; its own chart shows a median CI checkout going from 31 s to 3.5 s (x8.8).
An early Claude Mythos Preview snapshot rewrote the Git importer with no further instructions: 20x less memory, 4x faster.
Much of the design has prior art in Meta's Mononoke and Google's Piper. Anthropic expanded its credit to Mononoke on 15 October.
Anthropic's own verdict: Scion "is by no means a suitable replacement for git for general use." It isn't open source.

The scale that broke Git

The numbers from the post:

What	Figure
Main branch history	almost 100 GB
Compressed working copy	over 25 GB
CI checkouts	on the order of 25 million a day
Naive Git traffic for that	over 600 PB a day
Git mirrors at peak	several hundred

The 600 PB line took heat. HN user nh2 pointed out that it assumes a full clone per checkout, "which of course nobody should do". The author agreed: "The point is that it is a non-starter to operate this way but git doesn't hold your hand to do the right thing here."

The real story is the pile of workarounds: shallow clones, sparse checkouts, promisor remotes, mirrors, and repo snapshots ("seeds") in blob storage. Seeds drifted further from main between refreshes. Diff and blame on partial clones needed more and more care.

Then came the observation that changed the design: "nearly all of our clients were agents and jobs that are online by definition." Git's best property, a complete offline repo in every clone, had turned into dead weight.

How Scion works

The server does the work

In Git, every clone can act as a full server. Scion is deliberately asymmetric. The server answers history and blame questions, creates commits, and performs merges and rebases. The client keeps a bounded object cache, lays out files and watches the filesystem.

Every query the client can make is also an API. A CI job that needs the diff against its merge base just asks for it, with no checkout at all. Anthropic says its automated bisect and revert pipelines "wouldn't be practical at scale without this cheap, programmatic access to source control."

There's no local commit and nothing to push. Refs exist, but the workflow discourages them except for main and release branches, because global state that every server has to agree on doesn't scale.

Changesets, not trees

A Git commit points at a full tree. A Scion changeset is an entry in a write-ahead log. It lists the files changed relative to its first parent, with content-addressed pointers to compressed contents in blob storage.

That makes writes cheap: no packfiles to accept, no locks, no cross-server coordination. Moving a ref is a compare-and-swap on a single Spanner row, separate from creating changesets.

Derived data

A changeset only knows what changed, so on its own it can't answer "who last touched this line". Scion computes derived data for that: blame structures, path history, skip lists over the commit graph. Each type is defined recursively from a changeset's parents, can be computed eagerly or deferred, and is invisible to the client. So the team keeps adding new indexes without touching the client.

On HN, the post's author also cited Martin Kleppmann's Turning the database inside out as an inspiration, alongside Mononoke. That essay describes indexes, caches and materialized views as data derived from the original.

A fetch protocol made for caches

This is the part I'd steal. Git's fetch negotiation revolves around what the client already has, so the server can compute a packfile of what's missing. Scion's client only says what it wants. It starts from a root tree, checks its local store, then requests missing trees and blobs recursively.

Every request names an immutable, content-addressed object, so any cache can answer it:

job -> scn client -> scion-daemon (per node) -> scion-cache (per region) -> scion-server -> Spanner / cloud storage

All tiers answer identical RPCs, so the client doesn't care about topology. Each tier keeps a cache hit rate over 99%. Sparse and virtualized checkouts need no special support, because the client simply doesn't ask for trees it doesn't want. Transfers are resumable too, since there's no multi-gigabyte packfile response.

Anthropic's results: a 200x smaller footprint than the Git mirror fleet, checkouts "two orders of magnitude faster", and a 25% CPU capacity win for CI.

Read the post's own median-time chart before you repeat the middle one, though. Git vs scn:

Command	git	scn	Faster
Checkout in a CI job	31 s	3.5 s	x8.8
Checkout at the start of a compute job	175 s	22 s	x8.0
New worktree, agent environment	7.1 min	5.1 s	x84
Status, developer environment	3.9 s	19 ms	x207
Commit, developer environment	1.4 s	1.1 s	x1.3
Total source control time per change request, agents	34 min	3.7 min	x9.2

Big wins everywhere, but "two orders of magnitude" holds for a few commands, not for CI checkouts. Same caution for the merge chart: its callout reads "30,401 merged on Sep 9". The "over 30,000 per day" in the text is the record day, and the 7-day average line sits well below it.

Where Claude fit (and where it didn't)

Seven people built Scion and Antique, "not to mention half a dozen related systems," in five months, while merge volume grew 10x. What the post and the author's HN comments say about Claude:

The importer. The first Git-to-Scion ingestion pipeline was "largely handwritten". An early internal snapshot of Claude Mythos Preview rewrote it "with no further instructions": 20x less memory, 4x faster, with the ordering and data-structure invariants preserved.
Before that model, the author says, the code was "only moderately assisted by Claude with a lot of human-still-typing-things-in-an-editor." Mythos Preview brought "a noticeable, step function increase."
Engineers said no. Everyone on the team "would now rate Claude as better at writing code than them." But very often "the role of the engineers was to tell Claude 'no.'" One example: systems that worked out when a change merged by linearly reading the commit log instead of asking the code review tool. Adversarial review also pushed Claude toward over-engineering.
Review changed. Humans still review every change request, but they check its intent instead of reading every line.
Operations. Claude instances nicknamed "superfans" each sit in a Slack channel, through Claude Tag, watching one component's telemetry and logs.
No global merge lock. Hundreds or thousands of changes can land while tests run, so testing moved post-merge and pre-deploy. Conflicts happen, and "Claude can typically fix these conflicts faster than humans can even identify them."

Two caveats. Mythos-class models are limited to a small set of vetted organizations, so most of us can't reproduce that rewrite. And the 20x and 4x numbers come with no benchmark, which HN noticed.

The Mononoke question

Here's the timeline, from primary sources:

13 October. Anthropic's post credits Google's Piper and Meta's Sapling as prior art.
14 October. Rain, who worked on Mononoke at Meta, publishes Anthropic's Scion is heavily inspired by Mononoke on sunshowers.io (426 points). Mononoke is Sapling's server, and its Bonsai changesets look a lot like Scion's changesets. The charge: Meta's team "came up with many of the insights that Claude dutifully applied almost a decade later."
15 October. A second post, More on Anthropic's Scion ripping off Mononoke (254 points), makes the point explicit: "The problem is attribution, not copying." It also notes that "Scion" reverses the syllables of "Bonsai" and drops the "B", adding: "This might be a coincidence, or might not." On HN, citing private correspondence, Rain writes that "Claude was pointed directly at Mononoke and asked to re-engineer it." Nobody outside can verify that.
15 October, 17:59 UTC. Nick G., the Anthropic post's author, replies. Sapling was credited because that's how Meta refers to the whole system. "We also did not just tell Claude to build us a version control system." A footnote gets added. Asked what share of the codebase was directly inspired by Mononoke: "Very little, overall."
Same day. The post now reads "Updated October 15", with a footnote saying Mononoke "pioneered many of the techniques we use in Scion, in particular the Bonsai data model." Rain replies: "I now believe the attribution is sufficient."

The best exchange came next. durhamg, "the longest serving and most senior member (now ex-member) of the Facebook source control team", wrote: "I had no problem with your post." Rain answered that "this belief is not unanimous among the current source control team at Meta."

Why not just run Mononoke? HN user fcarraldo had tried: "Open source Mononoke and EdenFS don't really work outside of Meta."

For context, here's who got here first:

System	Who	Since	The idea Scion shares
Piper	Google	2012	A central repo. The toolchain "requires developers to be online" (CACM, 2016)
Mononoke and EdenFS	Meta	2017 and 2018	Sapling "transitioned to a client-server architecture" (Sapling docs)
VFS for Git, then Scalar	Microsoft	Scalar in Git 2.38	Git "slightly more coupled to a central server for missing objects" (Microsoft, GitHub)

That same 2016 paper counts 16,000 human changes and 24,000 automated ones on a typical workday in Google's monorepo. So the scale isn't new either.

My read: the ideas came from people who solved this a decade ago, and since 15 October the post says so plainly. What Claude changed is the cost of rebuilding those ideas around your own infrastructure. That's a smaller claim than "Claude designed a VCS". I think it's also the more useful one.

What Git already gives you

Anthropic isn't asking anyone to leave Git. It wants forges to offer server-side RPCs for history and diffs, and it wants Git's sparse and shallow features to "compose well".

GitLab answered the same day with Scaling Git for the agentic software factory. It quotes Nick G.: "We see room to expand what Git servers can do for these workloads without departing from the overall Git model." The two have been collaborating since GitLab's June event, where Nick G. joined them on stage.

On the client side, a lot already ships:

# Blobless clone: full history, file contents fetched when needed
git clone --filter=blob:none https://github.com/your-org/your-repo.git

# Only materialize the directories you work in (cone mode is the default)
git sparse-checkout set services/api packages/ui

# Or let Scalar do it: commits and trees only, a sparse checkout,
# and background maintenance. The worktree lands in your-repo/src
scalar clone https://github.com/your-org/your-repo.git

Before any of that, measure:

git rev-list --count --since=30.days.ago HEAD

Divide by 30. My own side project's two repos hold 1,273 commits on main in total, with 30 on the busiest day. At its record pace, Anthropic merges that many in about an hour. Git is not my bottleneck. HN user chris_money202 put it bluntly: "I have had zero issues using git with agentic coding."

What the headlines got wrong

The coverage ran ahead of the post in a few places:

AI Weekly shows an "Original headline: Anthropic Open-Sources Scion". The post never says that, and AI Weekly's own body says "it will not be open-sourced".
AlphaSignal: "Anthropic says Claude produced most of Scion's implementation". The post gives no share at all.
SecurityOnline: CPU and memory down "to a mere 0.05% of previous levels". 200x smaller is 0.5%.
Arize: "With traditional code review gone". The post: "Our team members still review every change request for the system".
The Register: these problems spread "over the next few months". The post says "over the next six to 12 months".

The part that applies to everyone

The post opens with the real takeaway: at Anthropic, "coding is no longer the bottleneck". When code gets cheap, the pressure moves to whatever comes after it: review, CI, deploys. Anthropic expects this "class of problems" to become more common "over the next six to 12 months".

So if you run agents in parallel, watch your review load and your CI queue before you worry about your VCS. Scion is what that pressure looks like on a 30,401-merge day.

Drafted with AI assistance (Claude). Every quote and number links to its source.

124 "Dofollow" Directories, Fact-Checked: 39 Pass Nothing

Céline — Wed, 14 Oct 2026 16:40:28 +0000

My SaaS, Awiser, had a low Domain Rating.

Every growth thread says the same thing: submit to directories, collect dofollow links, watch your DR climb. Before spending a dollar, I wanted to know which of those lists were telling the truth.

So I took 124 directories, launch sites and dev platforms from those lists and had AI agents read every pricing, submit and FAQ page. Then I checked them against five published audits of real listings' HTML, all from September and October 2026. Where no audit covers a site, I quote its own page.

39 of them pass nothing to a new listing. You get a nofollow or ugc link, a redirect, or no link at all. Product Hunt is on that list. So are G2, Capterra and Indie Hackers.

Below are the 39 to skip, the 11 free ones with no strings and the 13 that cost $15 or less. Then a 10-second DevTools check, so you can test any list yourself. This one included.

TL;DR

124 checked. 52 can give you a dofollow link. 39 give a new listing nothing. 29 more I couldn't verify, and 4 are press platforms where a journalist decides.
Only 11 directories give a free dofollow link with no badge, no fee and no top-3 lottery. The best of them, SaaSHub (DR 80), only follows your link once you verify the listing.
Most of the 52 that work want something. 29 want money. 9 want a badge on your site.
"Free dofollow" usually comes with fine print: a top-3 ranking, an upvote threshold, a badge you can never remove, or a queue of up to 60 weeks.
13 directories promise a dofollow link in writing for $15 or less.
Don't trust any list, this one included. The snippet below checks a live listing in 10 seconds.

What "dofollow" actually means

There is no rel="dofollow". A link is followed when its rel attribute is empty, or holds only browser flags like noopener and noreferrer. Three values switch it off.

`rel` value	Meant for	Followed?
empty, `noopener`, `noreferrer`	Normal links	Yes
`nofollow`	Links the site doesn't vouch for	No
`ugc`	Comments, forum posts, user profiles	No
`sponsored`	Ads and paid placements	No

Google says links marked nofollow, ugc or sponsored "will generally not be followed". Ahrefs is blunter: only followed links increase DR, and only the first one from each domain.

A clean rel can still hide a dead link:

A redirect. The button points to the directory's own short link, which forwards to you. If that hop is nofollow or noindex, like Dang.ai's, nothing reaches you. A plain 301 may pass value, but no page check can prove it.
A JavaScript button. No <a href>, nothing for a crawler to follow.
A page-level block. A noindex listing page drops out of Google, which eventually stops following its links. A nofollow robots meta tag switches off every link on the page at once.

The 39 that pass nothing

These show up on almost every "best dofollow directories" list. Here is what a new listing on each one actually gets you, as checked in September and October 2026. Two can pay off later: Hacker News follows a story from about 12 points, and AlternativeTo follows some established apps.

Site	DR	What a new listing actually gets
Trustpilot	94	`rel="nofollow noopener"`
SourceForge	93	Your URL as plain text. Paid listings get a `nofollow` redirect
Product Hunt	91	`rel="ugc"` on all 20 product pages checked, and 11 of 20 launch-day pages were `noindex`
Hacker News	91	`nofollow` on every story under 9 points, a plain link from about 12
G2	91	No link at all on free profiles. Paid ones go through a `nofollow` click tracker
Capterra	91	No link on free profiles. The paid button has no `href`
Crunchbase	91	`nofollow`
Awwwards	89	`nofollow`, after a $65 submission fee
Wellfound	87	`nofollow`
F6S	83	`nofollow`
Dang.ai	82	Its free plan requires a crawlable link to dang.ai. Visitors come back through a redirect served as `noindex, nofollow`
Indie Hackers	82	`rel="nofollow noopener"`
AlternativeTo	81	`nofollow` on new listings
Peerlist	77	Only the weekly top 5 get a link, and it's `ugc`
Futurepedia	73	`nofollow` tool links

Writing platforms don't save you either. Links inside Medium articles and GitHub READMEs are nofollow (checked in July), and Hashnode and HackerNoon mark article links nofollow or ugc (checked in September).

The other 20 include review sites that show no link at all, like GetApp, Software Advice and SoftwareSuggest. Others are closed: I found no public way to add a new tool to StackShare, CodeProject is archived, and Gartner Peer Insights only lists enterprise software. All 39 are in the full table linked at the end.

The fine print behind "free dofollow"

Most directories that do give a followed link attach a condition. It sits in their pricing table, not on the list that sent you there. I found five patterns.

Win your day. Open-Launch, Startup Fast and Aura++ share one pricing template: "Dofollow Backlink only if" you finish in the top 3 of the day and display their badge. LaunchIgniter and LaunchVault run the same rule weekly. For a new product with no audience, that means nofollow.
Hit a score. On Uneed, a free launch needs an upvote score of 20 for the dofollow link, or 10 with the $14.99 Fast-track. Of 60 launches from October 3 to 5, 48 were followed. All 12 nofollow ones sat at 8 points or less.
Pay. DevHunt's FAQ says it plainly: free listings are nofollow, paid launches get a permanent dofollow link. 90 of 92 free pages checked were nofollow. All 97 paid pages were followed.
Show their badge, forever. Startup Fame, Fazier, Twelve Tools, Turbo0 and ToolPilot give a real dofollow link for free if you link back. ToolFame and SaaSFame unpublish you when the badge comes off. That's a link exchange, and Google's spam policies list excessive link exchanges as link spam.
Wait. OpenHunts' free queue is about 60 weeks. SaaSFame reviews free listings "within 180 days". Huzzler's free page waits 72 days and ships noindex.

The rules also move under you. Uneed's free line closed on August 17 and reopened on September 11 with new rules. BetaList now charges for every submission, and its Visit Site button goes through a plain 301 redirect, which SaaSCity counts as followed.

The 11 directories that give a free dofollow link, no strings

No badge, no fee, no ranking contest. Some make you wait, and one makes you verify.

Directory	DR	The catch	Evidence
SaaSHub	80	Verify the listing with an email on your domain, then reconfirm each quarter. Unverified listings stay `nofollow`	29 of 30 verified listings followed, 0 of 13 unverified
ShowMeBestAI	76	None. Reviewed in 24 hours	Free plan lists a "Dofollow Backlink"
Launch Llama Tools	74	About four weeks of review	Official, plus a sampled listing
Supertools	72	Editors pick what gets in	Live audit
PitchWall (ex-BetaPage)	68	At least 30 days of waiting	"Do-follow backlink" on every plan, free included
Web Design Inspiration	66	Curated, so your site has to look good	Live audit
MicroLaunch	64	Pick the free option after signing up	69 pages checked, free and paid links identical
DodoDirectory	58	None	"Free dofollow backlink" on its homepage
AItoolnet	56	30 to 60 day queue	Free plan lists a "Do-Follow SEO backlink"
SaaSpo	56	SaaS websites only, curated	Live audit
Launching Next	55	Reviewed daily, no fixed date	Live audit

The biggest free followed link isn't a directory. DEV (DR 91), the site you're reading, keeps links inside an article's body followed, while the website field on your profile is ugc. Links in Substack posts and the Homepage field of a real PyPI package are followed too (tested here).

If you'll pay: 13 dofollow links for $15 or less

Each of these promises a dofollow link in writing on its own pricing page. One-time prices, checked in October 2026.

Directory	DR	Price	What the pricing page says
Aitop10	66	$6.90	"12 high-quality Dofollow backlinks", no backlink required
SaaSGrow	62	$9	"DoFollow Backlink", instant approval
OpenHunts	72	$9.90	"Guaranteed dofollow backlink"
LaunchVault	59	$9.99	"Guaranteed dofollow backlink", permanent
Similarlabs	70	$9.99	Dofollow links in six languages
See What New AI	63	$10	"Get a dofollow link", no link exchange
Uno Directory	70	$10	"lock in a DR 70 dofollow"
StartupTrusted	54	$11	"All your dofollow links are permanent"
Open-Launch	72	$12	"Guaranteed dofollow backlink from a DR 71 domain"
LaunchIgniter	75	$12	"Lifetime dofollow backlink (DR 75)"
SaaSFame	71	$14.90	"Get 3 dofollow links", listed immediately
ToolFame	76	$14.90	3 dofollow links, no badge
StartupBase	73	from $15	"Every paid launch includes a permanent dofollow backlink"

DR is Ahrefs via saascity.io, except Open-Launch, StartupTrusted and StartupBase (noonlaunch.com, September 2026) and Similarlabs and Uno, which are the sites' own claims.

Two catches. Aitop10's 12 links sit on one domain, and Ahrefs only counts the first link from each domain. Uneed ($14.99) nearly made this table, but its Fast-track link still needs 10 upvotes on launch day.

The badge route costs nothing but a link back: Startup Fame (DR 84), Fazier (84), Twelve Tools (83), Turbo0 (81) and ToolPilot (79). If you take it, put the badges in one "Featured on" row in your footer and stop at five.

Worth paying more for, if they fit your product: There's An AI For That ($49, DR 78) and Toolify ($99, DR 73, live in 48 hours) for AI tools. CSS Design Awards ($50, DR 80) if your design is strong, since a jury decides. AI Agents Directory ($49, DR 75) follows only from its $49 tier: its $19 Fast Track is explicitly nofollow.

Check any listing in 10 seconds

Open a live listing on the directory you're considering, open DevTools, and paste this into the console. Set domain to the site being listed: yours, or the product whose listing you're inspecting.

// Paste into DevTools on a directory listing page.
const domain = "yourdomain.com";
const robots = document.querySelector('meta[name="robots"]')?.content || "none";
const rows = [...document.querySelectorAll("a[href]")]
  .filter(a => a.href.includes(domain))
  .map(a => ({
    direct: new URL(a.href).hostname.replace(/^www\./, "").endsWith(domain),
    rel: a.rel || "(empty: followed)",
    href: a.href,
  }));
console.log("robots meta:", robots);
console.table(rows.length ? rows : [{ result: "No link to your domain: it's a redirect or a JS button." }]);

How to read the result:

direct is true and rel is empty, noopener or noreferrer: followed. That's the one you want.
rel contains nofollow, ugc or sponsored: not followed.
direct is false: the button goes through the directory's redirect first.
No rows at all: no crawlable link to you, only a short link or a JavaScript button.
The robots meta says noindex or nofollow: the whole page is out, whatever the link says.

Run it on two or three existing listings before you pay anyone, and once more on your own listing when it goes live.

The snippet can't see an X-Robots-Tag response header, and that's how Dang.ai's redirect blocks crawlers. For a redirect, check the headers in the Network tab of DevTools.

What directories won't do for you

They split their value. Ahrefs passes each linking domain's DR divided by the number of unique domains it links to. Its own example: a DR 26 site linking to five sites can move your DR more than a DR 73 site linking to 5,000. A directory with thousands of listings is that DR 73 site.

Google is not Ahrefs. Google's spam policies list low-quality directory links, and paid links that pass ranking credit, as link spam. A paid dofollow listing will count in your DR. Google may simply ignore it.

Directories get you off zero, and a few dozen clean links will do it. After that, what moves a site is stuff people link to because they want to, like data nobody else has published.

The full table

The full 124-row table, with each directory's DR, the exact rel attribute found, price, wait time and a source, is linked from the original post. Free, no signup.

Found a directory that moved your DR, or one I got wrong? Drop the listing URL in the comments. I'll run the snippet on it and add it to the table, with credit.

Sources

How this was checked: AI agents read each site's pricing, submit and FAQ pages in October 2026 and checked them against the audits below. Where no audit covers a site, the claim quotes its own page. DR is Ahrefs' figure as reported by saascity.io on October 9 and 10, 2026, unless noted.

Link audits:

SaaSCity: SaaS directories ranked by DR, plus its per-site checks of SaaSHub, Product Hunt, Hacker News, Uneed, DevHunt, G2, Capterra, Peerlist and BetaList, October 7, 2026. SaaSCity runs a competing directory and says so.
HackerNoon: I checked the rel attribute on 20 platforms, checks run September 17 and 18, 2026
NoonLaunch: 23 startup launch directories, data refreshed September 6, 2026
Witscode: AI tool directories, checked October 1, 2026
Stork: where to submit your AI tool, October 2026

How DR and link attributes work:

Originally published on awiser.co.

DEV Community: Céline

How Anthropic Replaced Git with Scion Using Claude

TL;DR

The scale that broke Git

How Scion works

The server does the work

Changesets, not trees

Derived data

A fetch protocol made for caches

Where Claude fit (and where it didn't)

The Mononoke question

Other pushback worth reading

What Git already gives you

What the headlines got wrong

The part that applies to everyone

124 "Dofollow" Directories, Fact-Checked: 39 Pass Nothing

TL;DR

What "dofollow" actually means

The 39 that pass nothing

The fine print behind "free dofollow"

The 11 directories that give a free dofollow link, no strings

If you'll pay: 13 dofollow links for $15 or less

Check any listing in 10 seconds

What directories won't do for you

The full table

Sources