Academy middle grades: worked examples first, then fade them

Date: 2026-10-30. Status: PROPOSAL for the /learn/ middle grades (reading primary sources, the 1987 judgment, limitation basics, the parcel timeline). Nothing here is built. Pairs with: academy-graded-quiz-attempt-then-reveal-mastery-gate.md (lock, then reveal). This note covers how to teach a skill question before testing it.

Evidence

• Worked-example effect, meta-analysis (Barbieri, Miller-Cotto, Clerjuste & Chawla 2023, Educational Psychology Review 35). Across 55 studies and 181 effect sizes (maths, elementary to postsecondary), worked examples beat plain problem-solving, with an average g = 0.48. The figures come from the published abstract summaries (cogbites.org, Taylor 2025); the full text was not read.
• Fading (Renkl and colleagues; the 2025 Miller-Cotto study on fading is in the same line). Start with a fully worked example, then remove one step at a time until the learner does the whole thing. The aim is to move novices from studying to doing without a cliff.
• The limit (expertise reversal, Kalyuga et al. 2003; from memory, not checked this run). Worked examples help novices and can slow down people who already know the skill. So fade them, and let strong members skip.
• Caveat, our inference: the evidence is mostly from maths. Reading a 1950s deed or a Court of Appeal paragraph is a procedural skill too (find the date, the parties, the operative words, what is decided and what is only reported), so we expect the effect to carry over. That is not tested.

The pattern (one skill, four steps)

For each middle-grade skill, for example "find the operative words in a conveyance" or "count a 12-year period from a start date": 1. Fully worked: show the source extract, verbatim with its catalogue ID, and each step done with a one-line reason. 2. One gap: the same kind of extract with the last step blank. The member answers, then Reveal. 3. Two gaps: the same again with two steps blank. 4. Independent: a fresh extract, all steps. Only this one counts toward the grade's 90% gate. Steps 1-3 are practice and are never scored.

Strong members can tap "Skip to the test question". If they fail it, they drop back to step 2. That is the guard against expertise reversal.

What to measure (at about 20+ members)

• First-try pass rate on independent questions for members who did steps 1-3, against those who skipped. Correlation only.
• Time to pass each middle grade.
• Drop-off points. A step where many people stop means it is too big a jump, so add a gap step.

How we apply it to Cedwyn

• The skills to build first, all from the evidence catalogue and the 1987 judgment text:
• reading a judgment paragraph: what was decided, and what was only recited as a party's case;
• limitation arithmetic: a start date plus 12 years, with the statute cited by section;
• placing parcels on the timeline;
• telling a catalogue entry that is SOURCED from one that is HELD-NOT-READ.
• Worked steps quote sources verbatim and cite catalogue IDs. Any step about the case's open questions carries "our case, prospective, not a finding". A worked step never states a contested point as fact.
• The final grade (open research prompts) gets no worked examples. It is the independent stage by design.
• Storage: practice steps need no saved score; only the independent answer is saved, as now. There is no extra data cost.
• Front end: the same question card, with a small "Example" / "Your turn" label and an ⓘ modal explaining why the examples come first. Dark theme, no white backgrounds.
• Hand-off: a proposal for the grades build lane, or the next content pass. It changes no prices and sends nothing.

Sources

• Barbieri, C.A., Miller-Cotto, D., Clerjuste, S.N. & Chawla, K. (2023) A Meta-analysis of the Worked Examples Effect on Mathematics Performance, Educational Psychology Review 35: psycnet.apa.org and eric.ed.gov (abstract snippets)
• Summary with figures (55 studies, 181 effect sizes, g = 0.48): mike-taylor.org and cogbites.org (snippets)
• Miller-Cotto, D. et al. (2025) Does working memory moderate the effect of fading on math learning? (snippet only)

Raw Markdown: academy-worked-examples-then-fade-primary-sources.md · version and changes: changelog · machine index: agents.json