When buildlogs named the place instead of the verdict (Sep 21–28, 2026)
Eleven high-engagement buildlogs from X and Indie Hackers for Sep 21–28, 2026, each traced to the specific place its result was decided, with the week's introduction posts separated out.
The three biggest posts under the #buildinpublic tag this week were introductions, and the biggest of them came from an account with 426 followers.
Luis Rieke, a founder in Berlin, opened with a plan: quit the job, build twelve startups in twelve weeks. That post drew 303 likes, 223 replies, 50 bookmarks and 22,506 views. 1 Alex, building a voice product in Milan, posted the same shape and drew 238 likes, 202 replies and 8,067 views from an audience of 2,063. 2 Natalie Zingg, sitting at 85 followers, asked for 100 and drew 117 likes and 91 replies from 251 followers. 3
Every buildlog in this issue's selection carried a smaller number than those three, and every one of them named a place — the screen where the signup died, the setting nobody had opened, the channel that produced no buyer, the line in the checklist nobody asked for. A reach number tells you how many people paused. A located number tells you which step to change. The second kind is what this issue collects, and the introductions sit at the end under their own heading, because they took the week's reach and the selection rule excludes them.
This issue covers eleven buildlogs published from September 21, 2026 at 08:00 through September 28, 2026 at 08:00, UTC-05:00, drawn from the #buildinpublic tag and developer timelines on X and from Indie Hackers. A post qualifies as a buildlog when a builder reports on their own product, numbers or process; introduction and connection posts, and launch-directory promotion, sit outside that selection. X figures are detail-page snapshots taken during this run, so likes, reposts, replies, bookmarks and views are capture-time values. Indie Hackers figures are the Likes and Comments visible on each post page at capture time.
The cases at a glance
| Case | Platform | Engagement snapshot | The place it named |
|---|---|---|---|
| axlura | Indie Hackers | 14 Likes, 50 Comments | The register screen: 152 ad visitors, 31 scan runs, zero signups 4 |
| Tania Bell | Indie Hackers | 19 Likes, 67 Comments | The payout setting holding about 2% of a 6% fee 5 |
| loveoftheai | Indie Hackers | 20 Likes, 40 Comments | The channel: four marketplaces, zero buyer contact 6 |
| Hiren Kalariya | X | 16 likes, 13 replies, 2 reposts, 309 views | Launch day: a Product Hunt rank and the first sale on the same date 7 |
| Harry Ashton | X | 15 likes, 4 replies, 2 reposts, 400 views | The handoff: SEO given to an agent, domain rating 4 to 10 8 |
| Vuk Andric | X | 31 likes, 8 reposts, 8 replies, 12,642 views | The release count: 255 releases and 1,868 commits in 59 days 9 |
| M4X | X | 30 likes, 8 bookmarks, 7 replies, 2,057 views | The pre-deadline checklist, including the build nobody asked for 10 |
| john santhosh | X | 46 likes, 48 bookmarks, 6 replies, 3,485 views | The shipped capability: drawing geometry straight onto a 3D canvas 11 |
| prasibv | Indie Hackers | 20 Likes, 117 Comments | The report's length: coverage added where the reader needed a decision 12 |
| Kelon | Indie Hackers | 11 Likes, 37 Comments | The uncited claim a reviewer refused to put her name behind 13 |
| Kaia_Spec | Indie Hackers | 21 Likes, 48 Comments | The boundary that lets a model answer "should I" with "stop" 14 |
Where the result was decided
axlura: count the visitors who reached the wall, then move the wall
Axlura is a free accessibility scanner for websites, and its founder ran a small Google Ads test. The result was $211 spent, 113 clicks, zero signups, and the first instinct was to blame the keywords or the landing page. 4
Then he pulled two weeks of raw server logs and counted the steps. 152 unique visitors arrived from ad clicks. About 31 of them, one in five, ran the free scan successfully with no errors, and zero of them ever reached the register page. The visitors engaged with the ad and the tool; the traffic stopped at one screen. To read your full report you had to create an account — name, email, password, captcha — which is a large ask from someone who only wanted to know whether their site was fine. 4
The fix replaced the account wall with an email box: enter your email to unlock the report, or have it emailed to you. Same lead captured, no password to invent. He also shipped a companion WordPress plugin so an agency could run scans from its own admin panel. 4
The Indie Hackers page showed 14 Likes and 50 Comments on September 28, 2026. 4
What to copy: Walk your own funnel one completed action at a time and write the count at each step; the step where the count reaches zero is your post. Then move the gate rather than the message, because an email address captures the same lead as a full account and asks for a fifth of the commitment.
Where the inference stops: This is one ad test and one product, and 113 clicks is a small sample. The post reports the diagnosis and the change, and it carries no signup figure for the weeks after the fix, so the improvement stays unmeasured here.
Tania Bell: put a number on the cost nobody has looked at
A founder had lost about 6% of an $899 sale to payment fees and was asking publicly which provider to switch to. Tania Bell answered with a different part of the bill: roughly 2% of those fees came from automatic currency conversion, quietly turning his dollar sales into his local currency. Her fix — hold an account in the currency you are paid in, route payouts there, convert when you choose — came from an account with about 100 followers at the time. A week later he posted that his fees had dropped. 5
She had met the pattern before, as a finance manager watching a team lose thousands to currency fees that nobody had flagged. Her estimate for founders who sell or spend across currencies is 2-3% of revenue, lost by routing money carelessly. Her first plan was a guide, and she dropped it once she realised the answer depends on each founder's own mix of currencies, revenue and spending. The replacement is a free, no-sign-up calculator that runs from a single HTML file, built while she learned to code with Claude and Gemini helping. 5
The Indie Hackers page showed 19 Likes and 67 Comments on September 28, 2026. 5
What to copy: Answer one specific question in public with a number attached, and treat your own answer as the market test — where the answer depends on the reader's inputs, a calculator beats a guide. A reply that changed someone's bill is worth watching even when it comes from a small account, because reach had nothing to do with it.
Where the inference stops: The 6% and the 2% are the founder's own figures from one sale, and the 2-3% range is her estimate from her own experience. One person changed one setting, and the total across other sellers stays outside the picture.
loveoftheai: rule out a channel by counting what it returned
loveoftheai bought an NVIDIA DGX Spark and set it a deadline: ten days, $1,000, with a coding agent doing the work and every outbound action gated by a rule. At the end of week one, revenue was $0. 6
What the week produced was a list with the demand caveat attached: 7 contract-versus-feature audit reports on real products; 26 unsolicited 60-second spec trailers posted within about 48 hours of other people's launches, which drew one thank-you and one price question and no sales; and 8 articles documenting the method on a developer site, with single-digit reactions. Three bounty submissions and one hackathon entry went out with no results announced. 6
The failures come with their own channels named: 40-plus one-to-one outreach messages, zero clients, and four proposals on one freelance marketplace, all four rejected. The agent also probed a submission API with junk and got the account docked as spam, after which endpoint probing became a hard rule and every submission field gets checked before it is sent. 6
His own reading after seven days: no marketplace produced a single buyer contact, and every promising conversation began with a post published at a moment — a launch day, a funding day, an approval day. The one artifact anybody picked up was also the one where every row carried a retrieval date and a raw quotation from the source document. 6
The Indie Hackers page showed 20 Likes and 40 Comments on September 28, 2026. 6
What to copy: Give an experiment a budget, a deadline and a channel list, then report the return per channel rather than the effort per channel. Put a retrieval date and a raw quote in every row of a document, because that is the version of the artifact someone can distrust and check. Gate an agent's outbound actions by category — submission, outreach, purchase — instead of adding another rule after each incident.
Where the inference stops: Seven days of a ten-day run, with the readings and hypotheses all the author's own. The spam penalty is his account of his own mistake, and the $149 offer he quotes had no buyer at the time of posting.
Hiren Kalariya: publish the rank beside the first sale
Hiren Kalariya launched on Product Hunt, finished fifth, and sold on the same day. His post reports both facts in one breath and closes with the question a rank cannot answer: did your launch day produce sales. 7
The detail page recorded 16 likes, 13 replies, 2 reposts and 309 views, against an account of 129 followers at capture. 7
A placement is a contest result for one day, which on its own gives a reader nothing to act on. A sale on the same day turns it into a claim about the traffic. Putting the two side by side, and then asking whether anyone else's launch converted, is what made the post a conversation rather than a notice.
What to copy: Pair the ranking you earned with the revenue it produced on the same date, and ask the question your own number cannot settle, so that other founders' answers become the comparison. Keep the reply count in view when you judge a launch post, since a placement plus a sale is a claim other builders can test against their own.
Where the inference stops: 309 views on an account of 129 followers makes this a small-audience post. "Actual sales" carries no count, and a rank describes one day's field.
Where the work went
Harry Ashton: hand over one recurring job and count what it moves
Harry Ashton's introduction post, four days earlier, drew 207 likes, 107 replies and 8,476 views from the same 2,601 followers, while the buildlog itself reached 400 views. 15
The buildlog is the more useful post. He handed his search optimisation to an agent that works around the clock and leaves pull requests waiting for review every morning. Two numbers came back: domain rating moved from 4 to 10, and signups arrive nightly. 8
The detail page recorded 15 likes, 4 replies, 2 reposts and 400 views. 8
Loading content card…
His background explains the shape of the post. He left a job in early 2025, reached $100,000 a month, lost the ranking that produced it overnight, and is now rebuilding a different product in public. The job he delegated is the one his last business already proved matters. 8
What to copy: Pick the recurring job you already know matters, hand it to an agent with a review queue rather than full authority, and publish the two numbers it moves instead of a description of the setup. A domain rating and a nightly signup count are both checkable later, which turns the post into a baseline you can be held to.
Where the inference stops: Domain rating is a third-party estimate, and "signups every night" arrives without a count. Four hundred views is the smallest audience in this issue's set, so the post is here for the decision it records, with reach and relative engagement both low.
Vuk Andric: count the days and the releases, then hand over the arithmetic
Vuk Andric has been building One4Home in public for 59 days, and he published the pace as four lines: 255 Play releases shipped, 1,868 commits on GitHub, 4.3 releases per day, 31.5 commits per day, closing with the claim that winning the shipping competition he entered takes volume. 9
The detail page recorded 31 likes, 8 reposts, 8 replies, 6 quotes and 12,642 views, against an account of 378,142 followers. A second post the next day reported growing users and free trials without a figure. 916
Loading content card…
The numbers invite a reader to check the work: 255 releases across 59 days, 1,868 commits across the same span, and both rates reproducible from the totals. That is the difference an effort post makes, since it lets someone else verify the pace instead of accepting a claim about it.
What to copy: Report the period and the totals together, then divide them in public. Four numbers that agree with each other are harder to argue with than one adjective, and a reader can run the same arithmetic on your repository.
Where the inference stops: Release count and commit count measure output, with nothing about retention, revenue or churn beside them. Twelve thousand views on an account of 378,142 followers is a low relative engagement, and the account's size explains the raw reach.
M4X: publish the checklist, and keep the line nobody asked for
Seven days from the deadline of a shipping competition, M4X published what was left: shipped to the store on September 9, submission filed, a browser demo running, an Apple Watch build, an Arcade submission in review, and PlayStation controller support pencilled in with a question mark. 10
The line that carried the post describes the Apple Watch build as one nobody asked for, and notes that it is a new track that is still unrequested. 10
The detail page recorded 30 likes, 7 replies, 8 bookmarks, 1 repost and 2,057 views, against an account of 223 followers, which makes it one of the highest like-to-follower ratios in this week's X set. 10
Loading content card…
A checklist with a status on every line scans faster than a paragraph update, and the self-aware line is the part a reader remembers, because most builders have shipped something nobody requested. Ending on one open question turns the list into an invitation to answer.
What to copy: When a stretch of work is nearly done, publish the remaining list with a status on each item, include the item you cannot justify, and close with the one decision you have still to make so readers answer it. Read bookmarks above replies as a sign that the format travelled further than the question did.
Where the inference stops: The list reports submissions and builds, with no revenue, install or rating figure. The deadline belongs to a competition, which supplies an urgency the product did not generate on its own.
john santhosh: ship one capability and let the demo carry the post
john santhosh shipped an update to a browser CAD tool: you draw directly on a 3D canvas and an artificial intelligence builds parametric geometry from the drawing in real time. The post is one line of capability, a link, and a video. 11
The detail page recorded 46 likes, 48 bookmarks, 6 replies and 3,485 views, against an account of 164 followers. Around 28% of that account's followers liked the post, and bookmarks ran above likes. 11 The product's own account had posted the same demonstration four days earlier to a larger audience of 2,318 followers, drawing 78 likes, 74 bookmarks, 11 reposts and 4,101 views. 17
Loading content card…
One capability, demonstrated rather than described, is the whole post. Bookmarks outrunning likes says readers saved it to try later, which is the behaviour a buildlog wants from a tool demo.
What to copy: Ship a capability, then post the capability and the ten-second demonstration with almost no commentary. Read bookmarks above likes as saved intent, and expect few replies, since a video that shows something working leaves less to argue with than a claim does.
Where the inference stops: The post carries no trial, signup or usage figure, and the small account base means 46 likes is a rate with a matching raw count. A product account and a founder account posting the same demonstration is one distribution pattern among several.
Where the claim had to change
prasibv: rebuild the report around the question the reader opens it with
prasibv built GazeRank to scan a site and produce a health score plus issues across search, performance, accessibility and AI search visibility. He spent weeks adding coverage on the assumption that users wanted completeness: more checks, more findings, more categories. 12
What he noticed instead is that people stall when a report gets long, because it never answers what to do on Monday morning. He rebuilt the report around a sequence — scan, prioritise, fix, verify, monitor — so the top of it answers three questions: the highest-impact thing to fix now, why it matters for this site, and how you will know it is fixed. His current default is three to five prioritised items with the full findings list underneath. 12
He says the answer is still open, and names the two questions he cannot settle: how much detail a report should show before it stops helping, when competitors ship a hundred findings and users seem to like it, and where a fix ends when the item is "low content depth" rather than "missing meta description". 12
The Indie Hackers page showed 20 Likes and 117 Comments on September 28, 2026, the largest comment count in this issue's set. 12
What to copy: Put the three to five highest-impact items at the top and keep the full list below, so the report answers a question instead of presenting a catalogue. Then publish the specific questions your own product cannot answer, with the alternatives you have weighed, since a question precise enough to compare two designs is what fills a comment thread.
Where the inference stops: No user, conversion or revenue figures appear. The restructured report and the two open questions are the author's account of a product still in progress, and the comments measure the community's interest in the question.
Kelon: turn every claim into something a skeptic can open
Kelon's product scores how trustworthy a site is, and it had reached four months of building at $0 revenue when a creator who covers scams for a living agreed to try it. She declined to recommend it, and gave the reason plainly: she had no way to independently verify where a score's claims came from. A flag saying that a forum reported something was text on a screen, with no link and no source text behind it. 13
He describes that as the correct objection: a tool built to stop people trusting things blindly was asking to be trusted blindly. 13
Two changes followed. Every source a scan cites is now a real link to the actual page rather than a domain name taken on faith, and a public page shows how the score is calculated, with the three signals, their weights, and where the method can go wrong. Revenue across the four months is still $0, which he reports without dressing up, along with cold outreach that mostly went nowhere. 13
The Indie Hackers page showed 11 Likes and 37 Comments on September 28, 2026. 13
What to copy: Put your product in front of the most skeptical qualified reader you can reach, and treat a refusal as a specification, since the sentence inside it names what to build. Then make the method public on a page of its own, weights and known failure modes included, because a product that asks for trust does better work when it shows its arithmetic.
Where the inference stops: Revenue is still zero, and one reviewer's refusal is one person's judgement. How much the two changes affected conversion is not reported. The 11 Likes count is low, so the post earns its place as an unusually clear diagnosis rather than as a high-engagement item.
Kaia_Spec: write the boundary the model is allowed to use
Kaia_Spec spent more than 5,000 hours working with ChatGPT, and the post that came out of it starts with the model stopping: enough information, enough authority, an available next action, and an answer of "something is wrong, I think we should stop here." 14
His rules are written as things the human also has to keep. Make the boundary clear, then respect it yourself: if a decision belongs to the model, do not overrule it later because you disliked the answer, and if stopping is permitted, do not punish the stop. He sums up the trade as pretending being expensive and asking being cheap, because a false premise carries into the next five decisions while "I don't know, can you decide this" costs thirty seconds. 14
His argument against rule lists is the one an agent builder can use directly: you can build a thousand guardrails, and then the model meets hole number 1,001. What he puts above the rules are the questions that generate them — what is this job for, what is the human responsible for, what evidence is required before acting — which sorts three questions that are easy to blur into one: can the model do it, is it allowed to do it, and should it do it at all. 14
The account of how he works is concrete. In August 2026 he wanted a small digital keepsake for crowdfunding supporters, which became a WordPress plugin, and the work split across several conversations: one holding the project's direction and earlier decisions, one on implementation, one inspecting what was built, and sometimes a fourth comparing versions. He kept product and runtime authority, installed and ran the builds, and left the source alone. 14
The Indie Hackers page showed 21 Likes and 48 Comments on September 28, 2026. 14
What to copy: Define what the model may decide and then hold the line from your own side, including when the decision goes against you. Let "I don't know", "this is harder than it looks" and "stop here" count as valid outputs, and write the purpose behind each rule so an unfamiliar case has something to reason from. When a project outgrows one thread, split direction, implementation and inspection into separate conversations instead of buying more context.
Where the inference stops: This is one person's method across years, with no measurement of speed, quality or cost beside it. The plugin and its architecture are his account, and the engagement describes how strongly other builders recognised the problem.
What the introductions bought
The three posts with the week's largest numbers were introductions, so they are described here instead of in the selection above, and the difference matters: an introduction asks a reader for a reply or a follow, while a buildlog asks them to judge a result.
Luis Rieke's version carried a plan rather than a request — a job quit, twelve startups, twelve weeks — and took 22,506 views and 50 bookmarks from 426 followers. 1 His week-two update, posted seven days later in the format the plan promised, drew 4 likes and 296 views. 18
Loading content card…
Natalie Zingg's asked directly for the milestone: 85 followers, a joke about needing 100, and an invitation to say hello. It drew 117 likes, 91 replies and 4,091 views from 251 followers. 3 Alex in Milan wrote the same post with a company behind it — a voice product where buyers pick what ships next — and drew 238 likes, 202 replies and 8,067 views from 2,063 followers. 2
Read together, the three measure one thing: an introduction converts a stranger into a follower, and the audience count is the result each of them publishes, while a price, a customer and a product number stay outside their reports. 123
What to copy: Use the format if an audience is what you want, and attach one checkable line — what the product is, what it costs, what it has done — so the followers arrive with a reason to stay. Then count your growth per format, since an introduction and a buildlog move different numbers and averaging them together hides which one worked.
Where the inference stops: Follower counts, likes and views are capture-time snapshots, and the three posts report reach with nothing about what the new followers did next. The introduction Luis quoted from another founder carried no hashtag and was published on September 20, a day before this window opens, so it is named here without being counted among them. 19
Writing the place into your next update
Each of this week's cases puts a number next to a place rather than a verdict. The shape moves to any product:
- Place. The screen, setting, line item, report section or channel where the result is decided.
- Count. What happened at that place, with the denominator it was drawn from.
- Boundary. What the count covers — one ad test, one sale, one week of a ten-day run, four months of building.
- Change. What you altered at that place, named specifically enough that someone else could apply it.
- Watch. The one number that will move if the change was right, and the date you will read it.
The first line is the one worth the most effort, and this week kept showing why. A post built on an audience number — an introduction, a follower ask — describes how many people arrived. A post built on a located number describes a step, and a step is the only unit a reader can act on: axlura moved one signup gate, Tania Bell named one payout setting, loveoftheai closed four channels, and each of those is a decision someone else can copy on Monday.
References
- 1
- 2
- 3
- 4We got 152 ad clicks and zero signups, by axlura
indiehackers.com
- 5How you can save 2-3% of revenue on FX fees, by Tania Bell
indiehackers.com
- 6
- 7
- 8
- 9
- 10
- 11
- 12More findings don't mean more fixes, by prasibv
indiehackers.com
- 13A creator refused to endorse it, by Kelon
indiehackers.com
- 14
- 15
- 16
- 17
- 18
- 19
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.