Technical History
Build Log
22 entries · Updated 28 September 2026
How BeatlesAnswers.org is researched, engineered, verified, and shipped. Entries reconstructed from the project’s dated planning and release history; nothing is invented. The record now covers nineteen distinct decisions, newest first. New entries are weekly.
28 September 2026
Tightening the Rebuild
Six fixes to how the site is rebuilt. The review badge now rests on fact checks, with house style tracked separately; three equipment defects become one ordered lane; release entries get stricter sourcing; internal workings come off public pages; and every session now ends with the same four steps.
22 September 2026
A Miss Is Not a Fabrication
Two quotations on a rebuilt page reported as absent from the book they were credited to. Both were printed there — one of them split by a hyphen across a column break, which no text extraction reassembles. Why the checker stays strict anyway, and what loosening it would let through.
16 September 2026
The Fix That Isn’t in the Release Notes
A Windows security update broke the build machine’s shell for six days and three build sessions produced nothing. The repair shipped on 14 September in an out-of-band update whose release notes never mention it — so the evidence that it is fixed is behavioural, not documentary.
8 September 2026
The Page That Disagreed With Itself
Two review-badged pages were shipping a number in one section that their own infobox contradicted four lines above. Every automated check passed them, because nothing in the toolchain reads a page against itself.
2 September 2026
A Badge Is a Claim
A page can pass eleven structural checks and still be unfinished. The depth gate measured everything about a song page except how much it said, and fixing that exposed something larger: 124 pages that publicly claim to be done are carrying an equipment list nobody sourced.
29 August 2026
Twelve Words at a Time
Two enrichment sessions died mid-step with an API error that named no cause. The verification method was the problem: whole pages of two in-copyright books were being pushed through the model on the way to checking a single citation.
26 August 2026
Three Alarms, One Fire
Revolver finished, and a review of one automated session raised four problems with it: three turned out not to be problems, and the fourth was a credential leaked through the one field no scanner was inspecting.
19 August 2026
The Organ That Cannot Be Heard
Two Hammond organs credited on sleeves that cannot be heard on the records, eight take counts corrected downward, three source conflicts recorded rather than resolved, and the enrichment campaign crossed 100 of 214 songs.
12 August 2026
The Event That Never Fired
A GA4 event bound to a page element that exists on zero pages had been silently reporting nothing for months; nine JSON-LD blocks turned out unparseable, not the two Search Console found; and enrichment crossed 90 of 214 songs.
3 August 2026
Forty-Four Takes That Never Happened
Rock and Roll Music was a single take, not nineteen; The Night Before claimed forty-four takes when two were recorded; the enrichment campaign crossed from Beatles for Sale into Help! and the re-verify streak hit twenty-three.
27 July 2026
Nineteen Was Never the Answer
A recurring wrong take count ran through every Beatles for Sale page; a scope ruling on Releases sections clarified what enrichment is and is not; and the re-verify streak hit twenty.
23 July 2026
Measuring Before Doubling
A three-session controlled trial measured whether two enriched pages per session could match one-page quality; the data said yes, and the 1 June 1964 cluster was the proving ground.
16 July 2026
The Rhythm Is Weekly Now
Scheduled sessions carry the enrichment beat; take-counts and consoles corrected against the session record; and the deploy pipe gains an integrity gate that refuses to push a clipped file.
9 July 2026
Publish Was the Dangerous Part
A pre-mirror scan catches a silently clipped page; the push pipeline is automated and immediately earns its guards; and 999 pieces of placeholder ad code leave the site before review.
2 July 2026
Why the Second Source Matters
The dual-source remediation queue drained through the Let It Be sessions’ equipment story, release-history language corrected for the 2021 boxset era; and the build log goes live.
24 June 2026
A Quality System Has to Tighten
All 47 enriched pages on the locked structure, inherited take-counts corrected against Lewisohn, a content-hash sitemap, standing deploy authority; and the two-source rule hardened.
17 June 2026
When the Checks Lied
Inline citations became book-level footnotes; a byte-level integrity scan caught 16 silently clipped pages; the mobile nav got a burger; and the CSS skin standard was codified.
10 June 2026
The Process Is the Product
A ~92% correction rate proved in-session verification is provisional; a pre-commit footer check fixed it; and a thin-page incident produced a model-agnostic session contract.
3 June 2026
Off by Exactly Four Pages
The deep-enrichment campaign opened; and cross-session verification immediately caught 11 citations off by exactly 4 pages, a 60-page systemic error, and a source that had to die.
27 May 2026
A Reference Site, Not a Referral Funnel
Person schema and sameAs entity links on 279 pages, a 6,420-edge internal link graph, extensionless canonicals; and a citation policy with zero outbound to competitors.
20 May 2026
Boring Fixes That Matter
v9 brought photographic backgrounds, v10 an SEO pass; then three independent external audits fed a foundation-hygiene phase and a resumable project plan.
13 May 2026
Static HTML, On Purpose
How BeatlesAnswers.org went from an empty folder to a deployed static site in its first week; generator chain, versioned deployments, and a UK-canon mission.