You don't have something to sell yet — and the gap isn't the website.

You're right that a real site is higher fidelity, and right that it's too much for what it's worth. But the framing hides the actual decision. Auth and a database are not what separate a document from a product — video is. Strip the login out and you can still deliver everything the paywall promised. Strip the video out and you can't, at any price.

This is from the live paywall copy, not a plan. Twelve women tapped the button underneath it.

The objection isn't that a document feels cheap. It's that you cannot teach a sit-to-stand to a 57-year-old with a bad back using stills and a paragraph. Form is the entire safety margin in this population, and the paywall explicitly sold “a gentler version one tap away” — which is a video affordance, not a document one.

Ship a PDF against that copy and you get refunds from the women who read carefully and injuries from the ones who don't. It's the one option that is both cheaper and riskier.

A PDF is fine as a companion — the week's schedule, what to have nearby, what “too much” feels like. It is not the deliverable.

Two reasons, and the second is the one that matters.

You'd be building for eight people. Test 2 is scoped to roughly 540 paywall views. At the 1.4–2.9% range the round-1 data implies, that's four to sixteen buyers. Accounts, password reset, session handling, a progress schema, and the support burden of a 60-year-old who can't log in — for a cohort you could name individually.

And it would measure the wrong thing. Your product thesis is that this is the programme she doesn't quit. A database tells you she stopped on day nine. It cannot tell you that she stopped because her daughter visited, or because the standing balance work frightened her, or because eighteen minutes was fine but changing clothes for it wasn't. That sentence is the product. At this scale the only instrument that captures it is you, asking.

Automated tracking is what you build once you already know what to look for. Right now you don't, which is precisely why the manual version isn't a compromise — it's the higher-fidelity option.

Note where the middle option breaks: around forty buyers. If Test 2 puts you there, it has passed emphatically and you've earned the build — with a spec written from observed behaviour instead of a constructed persona.

The four archetypes look like four programmes. They aren't — and the quiz code is unambiguous about it. The four-week structure is a single phase progression, identical for every woman:

Those four cards are static HTML. personalise.js never touches them. What the archetype actually changes is copy: which axis gets the “Start here” badge, the order of the milestone list, the headline, the research statistic, and the twelve-week projection curve. Not one exercise.

And the line that reads most like personalised programming — built around your back and your pelvic floor — comes from planAround(), which reads her pain regions and pelvic-floor answer, not her archetype. That is a guarding layer: what to be careful with. It is already served by the gentler variant you were filming anyway.

The structure was reached by not having time to diverge it. It happens to be the correct answer anyway, and it is worth being able to say why — because “everyone gets the same thing” is the kind of decision that gets second-guessed later.

You have already built the layer that matters. level() bands each axis score into four rungs — Needs attention through Going well. That is an entry-level system, and it is a better form of personalisation than four separate programmes.

A woman scoring 20 on mobility and one scoring 70 do the same movement at different rungs. The same hip work is “what loosens your mornings” to one and “what makes your ankles trustworthy” to the other — both true, one clip.

The adherence worry — that a shared week 1 makes the personalisation feel hollow — does not survive contact with how she actually experiences it. She never sees another woman's plan. She sees hips & lower back, because you mentioned stiffness, at her level, with her name on it. The exercise identity is invisible; the framing is the whole experience.

Genuine divergence belongs at week 3 and after, once she has a base — more single-leg progression for the balance primary, more load for the strength primary. That is real archetype-specific footage, and it is filmed after the validation question is answered, with revenue in hand.

The twelve-week finding collides with the pricing recommendation from the round-1 read, and the collision is worth being explicit about rather than smoothing over.

The annual price wins on acquisition maths and loses on fulfilment. $89 collected on day one turns a retention bet into cash — but if it is sold as a year, you owe a year of programming you have not filmed, to strangers, which is the obligation the August verdict already flagged as unresolved.

The monthly price is the reverse. $18/month means you never have to be more than four weeks ahead of your newest buyer. Content obligation renews with the payment. It is the safest fulfilment path and the worst CAC path.

Then make the page agree with itself: either delete the runtime 12-week overwrite and sell four weeks, or fix the static copy to say twelve and commit to filming it. Whichever you pick, one of those two strings has to change before checkout opens.

The plan she sees at purchase should span the full twelve weeks, forked by archetype, with only week 1 actually filmed. This is not a workaround for missing footage — it is where the archetype finally earns its keep. Week 1 does not fork, so if the plan view stops at week 1 the personalisation is invisible. Extend it to twelve and the divergence appears exactly where it is real.

And the resolution should fall with distance, because that is how periodisation actually works. You cannot responsibly specify week 9 before seeing how she answered weeks 1–4. A plan claiming to know week 9's sets and reps is either lying or is not adaptive.

Which leaves the middle band — eight paragraphs of block theme. An afternoon.

No new quiz work is needed for any of this. p.primary, p.secondary and the goal-versus-measurement reconciliation at CONFIRM_CEILING are already computed on every completion.

The tempting order is: film everything, then open checkout. That spends a trainer day and two weeks before knowing whether a single stranger pays — which is the exact mistake round one already made in the other direction.

Invert it. The obligation window is the build window.

The only hard commitment this creates is that week 2 must exist by day eight. Put the sequence in writing on the paywall — your first week is ready now; each week unlocks as you reach it — which is both true and a normal, well-understood way for a programme to work.

Everything the page promises becomes a contract the moment a charge lands. Three things on it cannot be honoured on the timeline above:

This will cost conversion — the page gets less impressive. Take the hit. A conversion rate measured against a promise you can keep is the only one worth having, and an inflated round-2 number would send you into a build on false evidence, which is a far more expensive mistake than a lower percentage.

Keep the week-4 retest. It's cheap — a form comparing four numbers she already gave you — and it's the proof-point that distinguishes this from a workout library.

You are about to charge women in their fifties and sixties for balance training and pelvic-floor work. Several told the quiz they are unsteady on stairs and plan their day around toilets. The failure mode here is a fall, not a refund.