Academy discussion: seeded prompts per grade, posted by us (not just replies)

Date: 2026-10-30 · Status: PROPOSAL (nothing built) · Topic: community building / course retention for the paid /learn/ academy (£49.99/month, 7 grades, 90% gate)

What the evidence states (primary sources, with numbers where they exist)

1. Posting beats replying, and the gap widens with class size. Du et al., HICSS 2022 ("Posting versus Replying: The Effects of Instructor Participation in MOOC Discussion Forums", doi 10.24251/hicss.2022.006): instructors mostly reply, but instructor posting (starting a thread) raises student participation more than replying; class size weakens the effect of replying but not of posting. (From the abstract; effect sizes not read in full text.)
2. Reading the forum is the strongest predictor, not writing in it. Chiu & Hew, AJET 2017 (doi 10.14742/ajet.3240), a humanities/art MOOC, two samples, stepwise regression: peer learning and quiz scores were predicted mainly by viewing forum messages, less by commenting, and voting least. Implication: most members will lurk; the thread must be worth reading.
3. Forum engagement raises the odds of finishing. Bonafini et al., Online Learning 2017 (doi 10.24059/olj.v21i4.1270): binomial logistic regression, forum posts and videos watched both increase probability of course achievement; most posts were information-seeking, not critical thinking, so the authors recommend prompts about deep meaning or applying concepts.
4. Weaker learners gain from reading peers and from quick feedback. Wei et al., PLOS ONE 2023 (n = 335, doi 10.1371/journal.pone.0293668): low scorers who made significant progress interacted less, learned from peer-made material, and wanted quick feedback on their own posts.

Caveat: all four are MOOC studies (thousands of free learners). A paid academy with a handful of members is a different setting; the direction (seed threads, make them readable, answer fast) is our inference for Cedwyn.

How to do it (playbook)

1. One thread per grade, opened by us, not empty: a 3-line seed post with one application question ("Which record would you request first to settle X, and why?") rather than a recall question the quiz already asks.
2. Post a new seed on a fixed day each week (one per active grade). Posting is the lever; replies are the follow-up.
3. Make threads readable for lurkers: pin a "best of the thread" summary each month with catalogue IDs; that serves the 80-90% who only read.
4. Reply within 24 h to every first post by a member (quick feedback matters most to struggling learners).
5. Spoiler rule: discussion sits after the reveal of a grade, or answers are hidden behind a spoiler fold, so the 90% gate stays honest.
6. Measure: members who view a thread vs pass rate at next attempt; 30-day retention of members who viewed or posted vs those who did neither (small numbers: report counts, not percentages, under 30 members).

Free tooling options (no install on the box)

• Own D1 table (cheapest, keeps the dark theme and the member gate): thread(id, grade, seed_md, created), post(id, thread_id, member_id, body, created, hidden); served by the existing cedwyn-api Worker behind the membership check.
• giscus (free, open source, MIT; giscus.app) embeds GitHub Discussions as comments with no tracking or ads. Downsides for Cedwyn: members need a GitHub login, the content is public on GitHub (breaks the paywall and the no-outbound-links rule), not real time. Only for a public, free Grade 1 thread, if at all.
• Nothing installed.

How we apply it to Cedwyn

• Add a "Reading room" panel under each grade's results screen (dark theme, ⓘ modal explaining the spoiler rule, page journal entry when a seed is posted).
• Seed prompts drawn from the evidence catalogue with IDs, in neutral third person; case points labelled "our case, prospective, not a finding"; banned public words avoided; never name family members; moderation queue (hidden=1 until reviewed) because public-facing text has strict editorial rules.
• Weekly seed: a scheduled worker can draft the next seed per active grade from the catalogue; a human-style check against the editorial rules before it shows.
• Grade 1 (free taster) gets one public read-only "best of" thread as a conversion surface: visitors see the quality of discussion, posting needs membership.
• Not yet: no live member community exists yet (the 10-minute member sync log reads 0 joins as of 22:55Z on 30 Oct), so build the D1 tables only after the first members join; until then, a single seeded thread is enough.

Sources

• doi.org (Du et al. 2022, abstract via OpenAlex)
• doi.org (Chiu & Hew 2017, abstract)
• doi.org (Bonafini et al. 2017, abstract)
• doi.org (Wei et al. 2023, abstract)
• giscus.app

Raw Markdown: academy-discussion-seeded-prompts-instructor-posts.md · version and changes: changelog · machine index: agents.json