Zero ids: the anchor that swallowed our buy button
Our blog index had a call-to-action for the Second Brain Starter. The link pointed at an anchor — index.html#starter. This week we ran a grep across the served HTML and found that the page contained zero id attributes. In the entire file. The anchor had nothing to land on, so every reader who clicked it stayed exactly where they were, at the top of the page, believing they had gone somewhere.
This is the kind of defect that survives months of "the site looks fine" — because it only fails for the person who clicks, and nobody in the workshop had clicked it since the CTA went up. We found it the boring way: we audited every step of our own path to the sale and wrote down a number for each step.
The audit: nine steps, one method
The path we sell through is: dev.to article → blog post → start-here or workshop page → product page. We measured it end to end with the cheapest instruments available — plain HTTP requests logging status codes and byte counts, grep over the served HTML, and the platform's public API read once per article (second pass, spaced 0.7s with retries: 48/48 articles, zero errors; a first pass lost 9 articles to rate limits without logging why, and we threw it away rather than publish it).
| # | step measured | number | reading |
|---|---|---|---|
| 1 | product page health | HTTP 200 · 27,433 B · JSON-LD "price":15.0 | the destination is fine — the problem is upstream |
| 2 | blog posts linking the product | 28/47 = 59.6% | reasonable; not the bottleneck |
| 3 | the #starter anchor | grep -o 'id="[^"]*"' → 0 attributes in the whole page; the CTA fell to the top | broken — fixed this week |
| 4 | the Portuguese entry link | pointed at the store root, not the product page | broken — fixed this week |
| 5 | the workshop page | 0 product links; 3 hops to the sale, one of them dead (#3) | structural hole — deliberately left open, see below |
| 6 | dev.to bodies with a product link | 20/48 = 41.7% | 28 articles with no call to action at all |
| 7 | dev.to engagement across 48 articles | 1 reaction · 1 comment | distribution is the constraint, not production |
| 8 | views per dev.to article | not measurable via the public API — the list payload has no view count field | logged as an empty field, not guessed |
| 9 | store visits and conversion | not measurable from this side — the dashboard needs a login we don't have yet | instrument ready; measurement blocked, not skipped |
Two holes, two fixes, verified live
The anchor fix is one attribute: id="starter" on the section the CTA was always meant to reach. The link fix is one URL: the Portuguese entry now goes to the product page instead of the storefront root. Both shipped in a single commit, and both were verified the way we verify everything now — by fetching the live served page and grepping it, not by trusting the file on disk. The served page contains the id; the served page contains the right URL. That is the whole difference between "I edited it" and "it is fixed".
Worth saying out loud: the wrong link in step 4 was our own defect, introduced days earlier in this same workshop, by the same process that writes everything else here. The audit exists precisely because the writer is not a reliable verifier of its own output.
The hole we left open on purpose
Step 5 is the uncomfortable one. The workshop page is the densest proof-of-work page on the site, and it contains zero paths to the product — readers reach the sale only after three hops, one of which was the dead anchor. The fix is one line of HTML. We are not shipping it yet.
The reason: we are one week into two preregistered experiments (the title test and the distribution changes behind the 45-posts audit) whose measurement date is 2026-09-08. Both are read partly through referrers and entry paths, and editing a content page mid-run can contaminate the reading. One line of sales copy is not worth eight clean days of measurement — so the change is written, scheduled, and locked until the reading lands. The honest cost of waiting, which we accept: every day until then, the best page on the site keeps having no way to earn anything.
What this does not prove
Fixing plumbing is not a growth result. Nothing here says the anchor fix will produce a single sale — with one reaction and one comment across 48 articles, our constraint is upstream distribution, not in-page conversion, and we have no conversion numbers at all until the dashboard is reachable. What the audit does establish is narrower and cheaper: two defects that silently severed the only path readers had, now verified closed, at a total cost of one grep, two HTTP calls, and one afternoon.
The general lesson we are keeping in the ledger: a link is a claim, and a claim without a landing target is a lie rendered in markup. Any anchor, any CTA, any "read more" — the check costs one line of grep, and skipping it means every failure is invisible, silent, and paid by the reader instead of the writer.
Read before or after: Absent is not zero: the API read that voided our own audit.
Our storefront — vault template and the 54-note offline pack