For sites with data, code, or a tool behind them · one-time licence · unlimited sites
Every bad link you ever paid for had one thing in common. Somebody had to say yes.
A vendor taking your order. A site owner accepting a pitch. A marketplace holding your
balance. handsofflinks deletes that somebody. It is a link pipeline you buy once and own
outright: it builds real assets, publishes them through platform APIs, then re-fetches
each published page and reads the link attribute off the served HTML before any row may
say LIVE. No outreach email exists in the system.
Nobody has to reply.
An average guest post costs $459 as of August 2026, and that count comes from adsy.com,
itself a marketplace that sells guest posts. Rented links start at $175 apiece. Retainers
open near $5,000 a month with a three-month commitment. People pay those prices, run one
Semrush check, and find sites built solely for selling backlinks. The failure under all
of it is the same failure. A counterparty profits before you do and disappears after.
Remove the counterparty and the racket has nowhere to live. That removal is this product.
$99once, and it is yours, unlimited sites
vs
$175 to $450for one rented link, prices as of August 2026
One payment via Stripe. handsofflinks-v1.0.5.zip, 415,822 bytes of full source, arrives
the moment you pay. Thirty days, full refund, no reason required, and since you received
source code, you keep it either way.
181failed or undiscoverable. Published anyway, below
$0paid for placement. 0 manual actions
No outreach email. No PBN. No paid placements: $0 went to placement across everything I
built. No promise about your rankings, ever.
The problem
“You actually paid to get de-valued by Google.”
“I paid almost $13,000 and only saw a decline in clicks and impressions.”
That is a buyer reviewing a link building agency in May. He is not rare. Another watched a
keyword fall from rank 18 to 97 after the order went in. A third opened Search Console and
found the placement he had paid for dragging thousands of spam tag links behind it. His
summary: “You actually paid to get de-valued by Google.”
Here is what those buyers found out after the invoice. Most of what this category sells
as outreach is inventory: sites “built solely for selling backlinks”,
“super low-traffic sites with built-up DA”, the same template repeated across a
network that exists for no reader. You paste the placements into Semrush and the story
collapses. One reviewer put the entire market in seven words: “What, do they think we
wont check?”
You checked. That is how you got here. So I will not ask you to trust anything on this
page.
I do not sell links. I sell the machine that builds them. handsofflinks is a pipeline
you buy once, $99, full source, delivered the moment you pay, that writes short articles
from a CSV of your own numbers and publishes them under your own accounts through platform
APIs. No outreach email exists in the system. Nobody has to reply. Before any row in its
ledger may say LIVE, it re-fetches the published page and reads the rel attribute off the
served HTML, then confirms a crawl path exists. The audit you already run in Semrush,
performed before anything is counted.
I ran this method on my own 16 sites for 37 days, on my own money. 413 links built. 232
passed audit. The other 181 are in my ledger too: 130 published but undiscoverable, 51
dead. Average yield was 7.25 referring domains per site, 9 at the most, and then it
flattens. The rewrite you would actually be buying is newer than that history and has
produced 2 independently verified live links so far. Two is the honest number, so two is
the number I print.
It is macOS only. It runs from a plain terminal with no AI subscription; hands-off
autonomous operation needs an agentic CLI such as Claude Code. It is capped at 12 links per
site per week on purpose, and I do not promise rankings at all. If you want a big number in
Ahrefs by Friday, we are the wrong vendor. You can also keep doing what most operators in
this market do, which is nothing; nothing is free, and nothing cannot make your site worse.
This page is only for you if paying to go backwards is the thing you refuse to do twice.
The gate values behind that cap, exact-match anchors at 15 percent and a
dofollow band of 78 plus or minus 7, are enforced in code; the
penalty answer in the FAQ walks the whole mechanism.
The mechanism
Counterparty-Zero Publishing
Counterparty-Zero Publishing means the pipeline only touches surfaces where publishing
is a right, not a favor. Agents build an asset your site can credibly stand behind, a Go
module, a dataset, a short technical article grounded in your own numbers, and submit it
through the platform's public API. No pitch. No template email. No reply to wait for. Then
the step this category skips: the verifier re-fetches the published page with a browser
user agent, reads the rel attribute off the served HTML, and confirms a crawl path reaches
the page at all. Until that passes, no row may say LIVE.
You already audit link vendors after the fact, with Ahrefs or Semrush, usually to
discover what you actually bought. This machine runs that same audit before anything is
recorded, on every link, every time. None of the alternatives I audited publishes a
failure rate or any discoverability claim at all. Mine is printed further down this page,
denominator first.
You could build a rough copy of the publishing half for free; agent skills that attempt
it circulate on GitHub, and the buyers I want already know that. What the failures paid
for is the rest of it: a 116-entry channel registry where every reject keeps its recorded
reason, inline gates that cap exact-match anchors at 15 percent, a grounding rule that
refuses to write any figure it cannot trace to your own CSV, the post-publish verifier,
and 94 tests. You get full source. Grep it and check every claim in this paragraph
against the code.
How it works
Four steps, one run a week. Outreach is not one of them.
01
Research
Agents study your site and your data for an asset it can credibly own: a conversion
table, a benchmark, a small library, a reference article. Credibility is the filter.
Volume is not.
02
Build, for real
The engine composes from a CSV of your own numbers and refuses to invent a figure it
cannot trace to that file. If your data is thin, it writes nothing. You can watch it
refuse before anything real happens: the dry run needs no account and no real key, one
placeholder config line, then plan, build, and run with zero remote writes.
03
Publish through APIs
Eight channels are active: Mataroa, GitHub Pages, docs.rs, pkg.go.dev and the rest,
drawn from the 116-entry registry. Each submission goes through the platform's own
public interface, under an account you created and control.
04
Verify, then write it down
Re-fetch the served page, read the rel attribute, confirm a crawl path, and only then
ledger the row as LIVE, undiscoverable, or failed. The dead rows stay in the file. A
dead-link rate that only ever improves is a dead-link rate nobody is measuring.
Anyone who guarantees you no penalty is claiming a relationship with Google they do not
have. The gates exist to keep the machine boring: weekly volume capped, exact-match
anchors capped at 15 percent, spacing between channels enforced. I guarantee the software
runs and that I repair channels as they break. Never rankings.
Two prerequisites are hard walls. It runs on macOS, and macOS is the only system I have
verified. It operates from a plain terminal with no AI subscription; hands-off autonomous
operation needs an agentic CLI such as Claude Code, and if you do not already pay for one,
that is a recurring cost you would pay its vendor, not me. It is five commands, but it is
five commands.
The ledger, my own properties, 25 Jun to 31 Jul 2026
Every link built, day by day. The dead ones stay on the chart.
Thirteen publishing days. Two front-loaded bursts put 310 links on the board inside three weeks,
then it flattens, because the binding constraint is how many channels genuinely fit a
site, not how fast the pipeline can run. The tall bar is 26 June: 91 links, of which 58 came
back live, 8 had no crawl path and 25 failed outright.
Counts are from my own data/ledger.csv, the file the pipeline writes itself,
in the same format ./hol run writes on your machine. My ledger is my operating
record and is not distributed with the package; what ships is the pipeline that produces one, so
the numbers you can recompute after day one are your own.
Proof
I ran it on 16 of my own sites first. The failures print at the same size as the wins.
413
Links the ledger logged
16 sites, 25 Jun – 31 Jul 2026
232
That actually count
56% of what was built, the number I sell on
130
Published but undiscoverable
no crawl path, sandboxed host, or not server-rendered
10
Referring domains earned
distinct hosts, across the whole portfolio
75%
Dofollow, of the 232
174 dofollow, 58 nofollow, measured and reported every run
$0
Paid for placement
0 accounts bought, 0 manual actions
413 links built in 37 days, on my own 16 sites, with my own money. 232 passed audit,
which is 56 percent of what was built. 130 published but undiscoverable. 51 failed
outright. $0 paid for placement. Of the 232 live, 174 are dofollow and 58 are not.
Read those numbers as what they are: my history with the method, written by the ledger
the pipeline keeps. They are not a promise of your results, and they are not the output of
the exact build you are buying. v1.0.5 is a rewrite, younger than the track record that
shaped it. What transfers to you is the machine, the registry, and the gates. What does not
transfer is my ledger.
Twelve live rows are linked below this section. Open any one, look at the anchor, then
paste the URL into Ahrefs, and the numbers reach you from Ahrefs instead of from me. Buyers
in this market check everything, as they should. This page was built for the ones who do.
Now the number that should stop some of you. Across those 16 sites the average was 7.25
referring domains per site; the best site reached 9, and then it flattened. Ten distinct
referring hosts across the whole portfolio. That is a ceiling I ran into, not a throttle I
chose. It gets worse before it gets better: 130 of my links published fine and then failed the
discoverability check, so nothing could find them. 99 had no crawl path to them at all, 28 sat
on a sandbox host that serves a crawler nothing, and 3 were never in the served HTML.
Telegraph kept
13 of 114 discoverable. Arweave kept 0 of 28. Both are dead in the registry now, reasons
recorded. If your plan requires a big referring-domain number, this machine will not
produce it.
So here is the exit, ahead of the pitch. If you want volume in Ahrefs by Friday, we are
the wrong vendor, and a retainer agency will gladly take that brief at $5,000 to $25,000 a
month with a three-month commitment. If your whole program is one money page, a human
writing five sincere emails may still beat this machine. The fit check below is written to
turn those readers away, and it means it.
If you run a portfolio and would rather own the machine than rent its output, the trade
changes. Every paid alternative I audited monetises recurrence, per link, per month, per
hour; none sells the production system itself. This one does, and everything it publishes
sits under accounts you create and hold, so there is no inventory to vanish and no balance
to trap: walking away costs a revoked token, and the assets stay up. One payment, unlimited
sites. You run it once a week. The pace is the product, and the ledger it writes next is
yours.
Channels
Eight publishers ship. Eight more channels produced nothing at all.
Ten referring domains produced the 232 live links below. Seven of those domains have a
shipping publisher, which is eight ACTIVE entries in the registry because GitHub Pages ships
as two, and the last two rows of the first table are marked so you can tell which.
Both tables are below, at the same size. The second one cost 169 links and returned 13, and it
is the more useful of the two, because those channels all looked like they were working.
Several of these are nofollow on purpose. A profile that is 100% dofollow is
itself a footprint; the shipped policy targets 78% ±7 and recomputes the real ratio from
the ledger on every run, then prints it. It reports the number, it does not refuse a job
over it. Live mix across the 232 is 75.0%: 174 dofollow, 58 nofollow.
Which rel value a registry actually serves depends on where the link sits on its page, and
I read all five off the served HTML in
what rel attribute developer registries actually serve.
To read the rel values on any markup you have in front of you, paste it into the
dofollow nofollow checker, which reports ugc and
sponsored separately instead of collapsing everything into one verdict.
Whether a published link ever gets crawled at all is the other half of the problem, and
the measured odds per platform, 413 links across 22 host families, are published in
how to index backlinks.
What works
Referring domain
Asset type
Method
Link rel
Live / built
mataroa.blog
Long-form article
REST API
dofollow
69 / 70
github.com
Repo + profile README
REST API
nofollow
43 / 44
*.github.io
Technical article / hub
Git push
dofollow
33 / 45
pkg.go.dev
Go module docs
Git push + proxy fetch
dofollow
25 / 27
docs.rs
Rust crate docs
cargo publish
dofollow
23 / 24
njump.me (Nostr)
Long-form note
Relay publish
dofollow
15 / 15
code.dlang.org
D package docs
Dub registry
dofollow
9 / 9
huggingface.co (experimental)
Profile / model card
manual
nofollow
1 / 1
jsr.io (experimental)
Package README
CLI publish
nofollow
1 / 1
219 live of 236 built, 93%. The last two rows are marked experimental: they have
one verified link each, but no publisher ships for either, so they are not something you can
run. A buyer gets the seven rows above them. In the registry that is eight entries marked
ACTIVE on day one, because GitHub Pages ships as two publishers, and every one of them has
a working publisher behind it. Statuses re-verified
1 Aug 2026 with a browser user-agent, each confirmed to still contain a live anchor to the
target and confirmed not noindex.
What did not
Referring domain
Live / built
What went wrong
telegra.ph
13 / 114
Published, HTTP 200, index-follow, and almost nothing links to it. 101 of 114 have no crawl path.
arweave.net
0 / 28
Same failure, total. Permanent storage is not the same as a discoverable page.
zenodo.org
0 / 14
Record pages carry noindex. Nine reporting cycles missed it.
stackexchange.com
0 / 4
Reputation-gated before a link renders. Not automatable honestly.
crates.io
0 / 3
SPA shell. The link is not in the server-rendered HTML.
r-universe.dev
0 / 2
404 at the published URL.
surge.sh
0 / 2
robots.txt blocked.
gitlab.com
0 / 1
The account behind it never existed. A logging error, mine.
hub.docker.com
0 / 1
No outbound link rendered at all.
169 links, 13 survivors. Telegraph and Arweave alone account for 142 of them, a third
of everything ever built, and returned 13 live links between them: Arweave 0 of 28,
Telegraph 13 of 114. Nearly every one returned HTTP 200 and passed every rel, robots
and noindex check in the suite. They failed the check nobody was running: is there any path a
crawler can actually follow to this page? That check now runs before a row is allowed to say
live, which is why the working table above is worth anything.
Eight channels are ACTIVE in the shipped registry today (code.dlang, docs.rs, the two
GitHub Pages publishers, GitHub repo homepage, Mataroa, Nostr, pkg.go.dev), out of 116
catalogued. The rest are 58 candidates, 48 rejected with the disqualifying evidence written
next to them, and 2 paused. Run ./hol channels on day one and count them
yourself rather than taking this page's word for it.
Verifiable
Twelve of them. Click any one.
Twelve rows drawn from the 232 that passed audit, one or two per referring domain, covering nine
of the ten. Every one was re-fetched on 1 Aug 2026: HTTP 200, with the destination link present in the
served HTML. Not screenshots. Screenshots are what the vendors who burned you used.
All twelve are live, because these twelve were chosen to be checkable. The failures are
not hidden, they have their own table in Channels and their own bar on
the chart. Nine of the twelve are dofollow and three nofollow: 75%, the same dofollow share the
full ledger runs at, 174 of 232. This dozen is spread one-per-domain rather than weighted
by volume.
My own properties
Three live sites the pipeline runs against. Check them yourself.
Real businesses, not demo domains. Paste any of them into
Ahrefs'
free website authority checker and you get these numbers from Ahrefs rather than from me.
That is the point of listing them.
What these numbers are, and are not. These are the whole
link profiles of three sites that have existed for years, built by every channel their owner
has ever used. Across these three the pipeline built 94 links, of which 58 passed audit. It
did not produce a Domain Rating of 52 on its own and I am not going to imply it did.
You could disprove that in one Ahrefs lookup, and a claim you can disprove poisons every other
number on this page. What they do show is that a real operator runs this against real
properties with real authority, and that those properties have not been penalised.
Running it
Set it up once. Then it is one run a week.
Agents find the APIs and walk you through the accounts on day one. After that the pipeline wants
one run a week. That cadence is what keeps velocity inside the gates, and it is where the
safety actually comes from. Add sites whenever you like; the marginal cost of site number twelve
is a JSON entry.
Day one, once
./hol doctor reports what is missing. Agents locate each platform's API, you
create the accounts and paste a scoped token per channel. Nothing is shared, nothing is
ours. Then ./hol run without --execute produces the real sprint
plan and a full draft asset and publishes nothing.
Every week after
# the weekly loop./hol planread the sites, build the sprint./hol buildgenerate the assets./hol rundry run, publishes nothing./hol run --executepublish./hol verifyre-fetch every URL, write the failures down
Gate
Limit
Value
Why
G01
Links per site
12 / week
A link ceiling, not a target. In practice the pipeline never came near it. The binding constraint is how many good channels exist, not the gate.
G07
New referring domains per site
6 / week
The actual KPI. Net-new discovery is what moves a profile; a second link from a domain you already have does not.
G02
Links per channel per site
3 / week
Host concentration. One channel reached 114 posts pointing at 16 domains; that is when value falls and risk rises.
G03
Dofollow ratio
0.78 ± 0.07
An all-dofollow profile is itself a spam tell. Measured from the ledger every run and printed with the plan, not assumed. This one reports, it does not block.
Eleven gates are declared in config/policy.json. Seven of them are enforced in code
and will block or reshape a job (G01, G02, G05, G06, G07, G09, G11); two are measured and
reported on every run without ever refusing a job (G03 and G04, the dofollow band); and two are
declared but not yet evaluated by the shipped code (G08 unique body per link, G10 body reuse
similarity). Values above are read from config/policy.json in the package you get,
not from this page, and you can grep the code for each gate ID yourself.
Before you buy
Wrong for a lot of sites. I would rather you find that out here.
The channels this publishes to are technical, data-hungry and developer-adjacent. That
suits some sites unusually well and most sites not at all. Read the left column first. It
is the cheapest audit you will run this year.
Do not buy this if
You run a local service business. Plumbers, dentists, trades. These channels have
nothing for you and the assets will not land.
Your site has no data, no tool and no code of its own. The pipeline builds assets
out of something. If there is nothing to build from, it refuses to invent it.
You need volume. Measured on my own sites: 7.25 referring domains per site on
average, nine the most any single site reached, then it flattens. That is a ceiling I
ran into, not a throttle I chose.
You wanted someone else to run it. You run this. If you never run it, it never
runs.
You will not touch a terminal. It is five commands, but it is five commands.
You want a big number in Ahrefs by Friday. We are the wrong vendor, and I would
rather you learn that from this paragraph than from a refund email.
Buy this if
You run a SaaS, a tool, a calculator, or anything with data or code behind it.
You have more sites than link budget. The licence is unlimited sites, and the
marginal cost of site twelve is a config entry.
You are an agency or freelancer and per-client link spend is the constraint.
Clients hold their own accounts and tokens, and offboarding is revoking a token.
You audit vendors with Ahrefs or Semrush before you pay them. Good. This page was
written for people who check.
You would rather own the machine than rent the output, and you accept that owning
it means running it, about an hour a week.
What it requires. Read this like an equipment plate.
A Mac. macOS is the only system I have verified. Linux is untested, and I
will not sell it to you as supported.
Node 18 or newer, and comfort in a terminal.
About 30 minutes of real work. A CSV of your own numbers, with a provenance
header. The engine publishes nothing it cannot trace to that file. It refuses to invent
figures.
One free account on the first publishing channel, created by you. Roughly
ten minutes, most of it waiting for a confirmation email. The tool never creates
accounts for you.
No AI subscription for manual operation. It runs from a plain terminal.
Hands-off autonomous operation also needs an agentic CLI such as Claude Code,
which is a separate recurring cost if you do not already pay for one.
If you are on the left, or the plate above lists hardware you do not have, do not buy
it. A refund costs us both more than the sale is worth, and you will have spent a month
finding out what this section could have told you in two minutes.
Pricing
What renting links costs. Then what the machine costs.
Category spend. Renting one link from
the
productised market leader costs $175 to $450 as of August 2026. The
listicles
that win the buy-backlinks search results quote $150 to $1,500 per link, and monthly
plans from $1,250 to $5,100. That is the going rate for a single placement you did not
choose, on a site whose numbers you are expected to take on faith.
What recurs. The structure costs more than the sticker.
Retainer agencies in
this category run $5,000 to $25,000 a month with a three-month minimum, as of August
2026, and one
marketplace rents links by the month, a DA40 placement from $6.72,
gone when you stop paying. The one and two star reviews I read in this
category describe the exits: contract buyouts costing thousands, credits owed back on
cancellation, refunds trapped as balance you can only spend on more of the same. Every
alternative in that list is priced so that you never stop.
The honest workaround. Doing it yourself is free and costs time
instead. Community
consensus puts one quality DIY link at 10 to 20 hours of prospecting and pitching.
Freelancers
run $10 to $150 an hour with per-link outcomes of $100 to $500, and the average guest post
ran $459 in 2026, a
figure published by adsy, a marketplace that sells guest posts, so read it as a seller's
number. Value your own time above roughly $7 an hour and a single DIY link costs more than
this entire pipeline. That is an hours arithmetic, not a claim that the links are alike.
One more number. As of August 2026,
$99
at that same market leader buys exactly one content-syndication campaign. One
disposable order. Here, it buys the machine.
The ledger
$0 nothing to enter
Not a product. It is the evidence, published before you decide anything,
with the failures at the same size as the wins.
413 rows of my own operating history. 232 passed audit, 181 did not, and both sit
in the same table.
Twelve named live links you can click. Fetch one yourself and read the rel attribute
off the served page, the way the pipeline reads it.
The requirements plate. A Mac, Node 18 or newer, comfort in a terminal. It is on this
page so you can rule yourself out before you spend anything.
Every site you own or operate. No per-site fee, because there is no
per-site cost to me.
Unlimited sites, unlimited runs. One buyer, client work allowed.
All 8 active channels, plus the 116-channel registry with its reject reasons
recorded, full source, 94 tests.
Every published link is re-fetched, and the served rel attribute is read off the
page itself before a row may say LIVE. That is
the audit you already run with Semrush, performed before the row
exists, not after.
Runs on your machine, publishes under your own accounts and your own scoped
tokens. Nothing is hosted here and nothing phones home.
You receive handsofflinks-v1.0.5.zip the moment you pay, 415,822
bytes, SHA-256 published beside it.
30 days, unconditional refund. It is source code, so a refund cannot claw it
back. You keep the zip either way, and I would rather say that out loud than pretend
otherwise.
Before you pay, the operating requirements, all
of them. A Mac. Node 18 or newer. Comfort in a terminal. It runs from a plain
terminal with no AI subscription; hands-off autonomous operation also needs
an agentic CLI such as Claude Code, which is a separate recurring cost if you do not
already pay for one. If any line of that is wrong for you, do not buy it.
$99, one payment, via Stripe. The zip is on the next page
the moment you pay: handsofflinks-v1.0.5.zip, 415,822 bytes, SHA-256
printed beside it.
About 15 minutes to watch it run dry, no account and no
real key: one placeholder line in the config, then plan, build, run, with zero remote
writes. A first verified LIVE ledger row typically takes one sitting: 8 steps, about
56 minutes measured.
What does not happen: no subscription, no card kept on file, no
account to create here or at checkout, no email list, and nothing publishes until you
type --execute. The refund is 30 days, unconditional, one email, no
reason required. It is source code; a refund cannot take it back. You keep it either
way.
The same licence and the same unlimited sites. What it adds is my own
time, twice, by hand.
Everything in the pipeline, plus
A written review of your first ten publishing targets, before you run them. Cheaper
to strike a bad target off a list than to read about it in your own ledger.
A written review of your first live placement, read off the served HTML.
Three business days for each of the two reviews, from the day you send it.
The download link arrives by email within one business day rather than instantly.
The automatic download gate is wired to the $99 payment link only, so this one I send
by hand.
The buy buttons are hidden with JavaScript off, because the check that decides whether a
checkout link is safe to click cannot run without it. Nothing here will track you for turning
JavaScript back on. If you would rather leave it off,
email hello@handsofflinks.com and I will
send a direct invoice.
Checkout is opening shortly. The payment processor is still in verification,
so the buy button is deliberately inert rather than pointing somewhere that does not
work. Lock this price in the meantime, or
email me for a direct invoice today.
Owning it, against renting it
What recurs
Own the pipelineOne payment. No renewal, no update window, nothing on file.
Renting linksPer-link reorders forever, $6.72 to $31.47 a month per rented link, $5k to $25k a month retainers with 3-month minimums, credit-clawback cancellation traps.
Who owns the assets when you stop
Own the pipelineYou. Everything publishes under your own accounts; offboarding is revoking a token. Nothing can be repossessed.
Renting linksRented links vanish or get “replaced” for a fee when payment stops; guarantees are link-replacement clauses that admit impermanence.
Marginal cost of the next link or site
Own the pipeline$0. The ledger's cost column sums to zero across all 413 rows; the licence is unlimited sites and runs.
Renting links$175 to $450 per additional link at the productised leader; every additional site multiplies the retainer.
Whether failures are visible
Own the pipelineThe ledger publishes them at full size: 413 built, 232 passed, 181 did not, decomposed to the row.
Renting linksNo competitor page I fetched publishes any failure rate; the category's one and two star reviews are precisely about hidden failure.
How LIVE is verified
Own the pipelinePost-publish re-fetch reads the served rel attribute; a link cannot count as LIVE without a crawl path. 130 of my own links failed that check and were counted against me.
Renting linksA vendor screenshot and a DA number; buyers report discovering thousands of spam tag links only after Search Console picks them up.
Renting is still the defensible choice when you need one strong placement, once, and
never again. If that is you, rent it. This one is for the other case: the operator who will
still need links next quarter, on more than one site.
One number on this page is small
The 413-row ledger above is my own operating history with this
method: 413 built, 232 passed audit, 181 did not, 16 sites, 37 days. The zip you receive
is the rewrite of the code that built that history, carrying the same gates and 94 tests,
and the rewrite's own lineage has produced 2 independently verified live links so far.
Its README says so in those words. If you need the shipped build to have a long public
history of its own before you spend $99, do not buy yet. The refund window exists so you
can weigh the machine on your own hardware instead of on my word.
The refund, written against this category's habits
In this category, buyers report refunds that arrive as non-withdrawable platform
balance, credits that must be paid back before you may cancel, buyout clauses costing
thousands, and charges that continue on paused campaigns.
Here is mine, in full. Email hello@handsofflinks.com within 30
days of purchase and you are refunded in full, to the card you paid with. No reason
required, no form, no call. No balance, because there is no account to trap it
in. No buyout, because there is no contract. No card on file to forget
about. And you keep the source, because clawing back a zip is not possible and I will
not pretend it is.
If you followed the first-link guide and no ledger row reached LIVE, attach the run
log to your refund email. Same refund, no argument, and the log fixes it for the next
buyer. It is self-guided: no call, no calendar, nobody to schedule.
What I guarantee is that the software runs and that I repair channels as they break.
Never rankings.
One payment, not a subscription. Nothing renews and no card stays on file.
One honest footnote: hands-off autonomous operation needs an agentic CLI such as
Claude Code, a separate recurring cost if you do not already pay for one. Manual
operation from a plain terminal needs no AI subscription at all.
Every future fix and channel included, no update window, no renewal fee.
That is my stated commitment. I phrase it as a commitment because a promise about the
future cannot carry proof.
One setup step before the first link. A free account on the first channel
and its API key pasted into your config. Every other channel is optional and can
wait.
FAQ
The hard questions first.
Why is the whole pipeline $99 when agencies charge $5,000 a month?
Because they sell labor and placements every month, and this sells software once. A retainer buys people who must be paid again in March. But this is a tool: my marginal cost per buyer is near zero, placement cost inside the pipeline is $0 across all 413 ledger rows, and you supply the labor, about an hour a week. The price is low because nothing in it recurs to me. It signals what the product is, not what it is worth to the right operator.
Why did 130 of your own links count for nothing?
Because discoverability is a separate check from validity, and almost nobody runs it. 130 of my own 413 links published fine and then failed that check, and were worth nothing because no crawler could reach them: 99 with no crawl path to the page carrying the link, 28 on a sandbox host that serves a crawler nothing, and 3 that never appeared in the server-rendered HTML. Telegraph produced 13 live links from 114, Arweave 0 from 28. The pipeline now tests for a crawl path before a row is allowed to say LIVE, and the failures keep their rows. A dead-link rate that only ever improves is a dead-link rate nobody is measuring.
Is this just a PBN with extra steps?
No, and the difference is checkable. A PBN is a set of sites that exist to pass link equity, with hidden ownership and no readers. The active channels here are public platforms nobody at this company owns or controls: GitHub, docs.rs, pkg.go.dev, code.dlang.org, Mataroa, Nostr relays, Hugging Face, jsr.io. They cannot be made to accept anything, which is the point. A network you cannot coerce is not a network you can turn into a PBN. There is no self-hosted blog network in the shipped registry, and the ledger has zero rows on one.
Will automated backlinks get my site penalised by Google?
Weigh the mechanism, then my report, in that order. The mechanism: the patterns Google acts on are volume spikes, exact-match anchor stuffing, near-total dofollow, and thin networks. The system is configured against all four: exact-match anchors are gated at 15 percent, dofollow is reported against a 78 plus or minus 7 band, and the hard cap is 12 links per site per week. My report: Search Console showed no manual action across the checks I ran on my own sites. That is my account of my sites, you cannot audit it, and it is not evidence about yours. Anyone who guarantees you no penalty is claiming a relationship with Google they do not have.
Could I not build this myself, free, with an agent and a weekend?
The automation, yes. What you would then spend weeks rediscovering is the map: which of 116 channels reject and why, recorded entry by entry in the shipped registry, and which links pass every technical test and still count for nothing, which cost me 130 links to learn. The 94 tests and the gates encode the failures. If your time is free, DIY is a fair rival. Mine was not, and the ledger is the bill.
Is the first link just a post on a free blog platform?
The recommended first channel is the fastest end-to-end proof of the loop, not the strongest link. Its job is to show publish, re-fetch, verify happening on your machine, inside the refund window. The channels that carry real weight, pkg.go.dev, docs.rs, code.dlang.org, require your code to exist there and come after. Judge your first LIVE row as a working machine test. Judge the machine over two quarters.
What happens when a publishing channel changes its API and breaks?
It has happened and it will happen again. A channel that starts rejecting or silently dropping submissions shows up in ./hol verify as failed rows, and I mark it REJECTED in the registry with the reason written next to it. Two channels sit disabled that way today, each with its cause recorded. Fixes reach you free, with no update window. That is the commitment I sell under, and the registry is where you can watch me keep it or fail it.
Why trust a solo founder with this?
Price what it costs you first: no support desk, no SOC 2 binder, a real bus factor. All true. What it gets you: the person who wrote the pipeline answers the email, the ledger on this page is my own operating history published with its failures at full size, 232 passes and 181 misses in the same table, and the smaller number is the one that got published. Vendors do not volunteer their failure rates when the numbers are the pitch. None of the alternatives I audited publishes a failure rate or any discoverability claim at all.
Who owns the accounts, and what do I keep if I stop?
You own every account, on every channel. You create them, you hold the tokens, and revoking a token is the entire offboarding process, about two minutes. Published assets stay where they are because nobody here holds any means of pulling them down. That cuts both ways: I also cannot force a platform I do not control to remove something. And the licence, the source and the zip are yours outright, including after a refund.
What exactly happens in my first hour after paying?
Checkout hands you the zip immediately. ./install.sh runs with no sudo and no network. Then you can watch the machine work before anything real happens: no account, no real key, one placeholder config line, and plan, build, run print a gate-checked plan with its SKIP reasons, grounded draft bodies built from the included example CSV, and DRY-OK with zero remote writes. The drafts are grounded extracts of the data: your editor or your model makes them yours. First LIVE ledger row, following the included guide with your own account and your own CSV, typically lands in the first sitting, 30 to 60 minutes including the confirmation-email wait. Self-guided. No call, no calendar.
Not buying today
Lock this price instead.
Checkout is not connected yet. The payment processor is in
verification, and I would rather say that than fake a working button. Leave your domain
and your email and you hold the price on this page. You get one email and nothing else. No
card, nothing charged now. The buy buttons above are the faster route if
you already know you want it.