Tin Ear landing page — review items
Built 2026-08-30 by an unattended autopilot session through the
single-page-sales pipeline (all eleven stages). Deployed to a preview URL, noindexed.
Publication is not approved. Nine items below need Tim. Six of them block.
← back to the page · the artifact itself
Blocking
- blockerShould Tin Ear have a landing page at all?
It shipped 2026-07-01 and has never been announced anywhere. A page is a decision to put it
in front of people. Nobody made that decision; this session assumed it as the task. If the
answer is no, everything else here is moot.
- blockerThe artifact is being served from this deploy at
/play/.
That was a call made so the CTAs would actually work. It is a byte-for-byte copy of
~/Programming/surprise-and-delight/tin-ear/index.html as of today, which means
it will drift the moment the source changes. Decide: keep serving a copy and set up a sync,
point the CTAs at the GitHub repo instead, or give the artifact its own home.
- blockerThe AI-authorship disclosure.
The page says, in its own section, that Claude built it on commission from you, and names
you. Read that section and decide whether the framing is right and whether you want your
name on it. It is deliberately volunteered rather than buried, which is the only posture
that survives someone discovering it, but the wording is a judgement call and it is yours.
- blockerThe maker's note is referenced but not linked or hosted.
The page says “the whole design story is in the maker's note, which ships next to the
file.” Right now the only copy is in the repo. Either host it, link the repo copy, or
cut the reference.
- blockerFour claims in the repo's own prose docs are wrong, and are not corrected at the source.
The page uses the code values. The docs still say the wrong thing:
· Maker's note and a code comment both say tine 7 is 4 cents flat. Replaying the seed gives −3.65, and tine 9 (−4.91) is the flattest.
· README says an overnight is “about two hundred practices.” Eight hours is 192.
· Maker's note places the relapse stop at “practice 140 to 190.” The code gates it on flags and hole count, not a practice number.
· 232 candidate lines minus 58 killed is 174, but the shipped CORPUS holds 98 lines in 20 arrays. The extra cut is undocumented.
Worth a five-minute fix in the repo so the docs and the page agree.
- blockerNobody has ever used this.
The page has zero testimonials, zero usage numbers, and zero press, because none exist, and
it is built to work without them. It never implies otherwise. If that reads as thin to you,
the fix is a real reaction from a real person, not a line of copy.
Worth a look, not blocking
- reviewThe screenshots are real, from today.
Four captures driven with Playwright against a local server: 21 real clicks punching the
first three phrases of Twinkle Twinkle, then a real first play. The “again.” note
in the hero is the actual scripted fumble. Nothing composited, nothing posed. If you would
rather ship captures of a song you punched yourself, they should be recaptured.
- reviewThe page quotes sixteen pencil lines verbatim from
index.html.
Template slots are filled with concrete values (f, g, bar 2, bar 4) and the copy says which
bar you get depends on your holes. If you ever rewrite the corpus, these quotes go stale
silently.
- reviewNo urgency device anywhere, on purpose.
No streak, no countdown, no scarcity, no “your box misses you.” Using one would
contradict the product's own rule 2 in the same scroll. If someone later asks why the page
does not push harder, this is the answer.
How it scored
Adversarial gauntlet: two cycles, ten personas tuned to the three ICPs. Cycle 1 failed the
per-section floor (the mechanism section scored 2.5 with the Complete Novice). Seven fixes
were prescribed and applied. Cycle 2 passed at 4.36 overall with no cell
below 3.5. Ten of ten would open it; eight of ten would come back the next day, which is the
metric that actually matters here.
Voice: every rhythm and AI metric passes against a corpus-recalibrated scorer, validated
by running it on Tin Ear's own README and maker's note first. Zero em dashes, zero
exclamations, zero blacklist hits, zero instances of the bolded-list AI tell.
The full paper trail is in ~/Programming/single-page-sales/pages/tin-ear/,
one folder per stage. The honest risk, from the red team's pre-mortem: people may enjoy
reading about a machine having a small crisis without ever opening the box. The mitigations
are structural and they are in place, but it is a real risk of writing this well about this
product.