A template that can only be applied whole, with its structure invisible, teaches nothing

What it is

The button says “smart weekly.” Behind it sits a system prompt the user has never seen: role, format, bans, few-shot samples. They only see the three words they typed. Next week the button is gone, or the thing to write is a monthly, and they take nothing with them. Opaque template non-learning means that if structure cannot be seen and taken apart, no number of applies turns into “I will know how to ask next time.” The template consumes a one-shot result. It does not produce transferable prompt skill.

Editability is control. Visibility is learnability. Locked but visible, people can at least copy the slots. Invisible, there is nothing to copy.

Why it happens

Skill extracts invariants from examples. In a template the invariants are slots and constraints, not the three filled words. If the UI only exposes the fill end, what gets encoded is the button’s name and the look of the result, not “a weekly needs a range, an object, highlights/blockers, a reader.” On a new task the button name does not match and the skill goes to zero.

Few-shot samples and system roles are often hidden on the side the user cannot see. The deeper the part that actually does the work is buried, the thinner the surface available for induction. The product gets stable output; the user gets a dependency on that one button. This is skill sealing on purpose: not a failure to teach, a failure to hand over the textbook.

Studying it

Training uses whole-template apply (structure invisible). Test removes the button and switches to a neighbor task (weekly → monthly, or a swapped required-slot set). Watch whether people spontaneously write a prompt close to the original structure. Independent variables: whether training showed slot names and constraints, whether “the prompt that will actually be sent” can be expanded, distance of the neighbor task. Dependent variables: slot coverage without the button, number of transfer attempts from train to test, subjective “I know what it was doing.”

A control group should see the structure and never apply. If that group beats the apply-only group at test, learning came from seeing, not from repeated clicks.

Where it stops holding

One-shot consumer tools (remove background, outpaint) do not need transferable skill; opacity is reasonable packaging. In enterprises where experts maintain templates and operators should not edit prompts, not learning is a division of labor. For children or laypeople facing dangerous capabilities, not teaching them “how to ask in order to get around the limit” can even be the goal. In knowledge work, where the product hopes people will gradually leave the scaffold, opacity directly contradicts the teaching goal. This entry does not treat a wrong designer process assumption — a wrong textbook is a different failure; here the textbook was never issued.

Applying it

• Offer “view the structure that will actually be sent”: slot names, constraints, hidden system sentences. Collapsed by default is fine; absent is not.
• Near the result, restate in one human sentence which slots were used (“written on range, object, reader”). Put the invariants in the same visual field as judging the artifact.
• On a schedule, have people try a neighbor task without that button, as a probe of whether the template is teaching — not just the button’s click rate.
• Check: people who have used “smart weekly” ten times, button off, asked to write a prompt for a monthly. If the prompt lacks range, reader, highlights/blockers — the original structure — the template left no learning. Put “transfer without the button” in template acceptance, scored apart from “tidy output with the button.”

Related

• Same group: L2.03.1 Templates lower the cost of getting started · L2.03.2 Templates constrain the range of what users will express · L2.03.3 Templates must be editable, not only fillable · L2.03.4 Templates turn open input back into structured blanks, at the cost of openness · L2.03.5 Placeholder content in templates is kept verbatim; empty slots must be unskippable · L2.03.6 Templates encode the designer’s task assumptions and actively mislead when those do not match real work
• Nearby: L2.07 Prompt History and Reuse · L4.09 Skill Degradation · L2.02 Discoverability of What Can Be Said
• Search terms: opaque template non-learning · prompt skill transfer · visible scaffold