{"path":"research/engagement-gradient-priors.md","content":"# Engagement-Gradient Priors\n\n**Date**: 2026-08-19 · **Status**: Priors enumerated and corpus-grounded; numbers are priors, not measurements. **Prompted by the founder**: *\"We cannot expect most people to interact in a very fine grained way with Deliberus. As most don't help edit Wikipedia or publish peer reviewed articles or author new laws in politics either. We can expect a gradient of various levels of curiosity, engagement and interest.\"*\n\n## §0 The founder's frame, and why it changes the falsifiers\n\nExpecting fine-grained interaction from most users is the same mistake as expecting most readers to edit Wikipedia. The realistic prior is a steep gradient, and every falsifier and session-design choice must be calibrated against the gradient's EXPECTED shape — otherwise the expected shape gets misread as a null result. \"Most participants didn't decompose anything\" is the baseline of every contribution system ever measured, not evidence against the substrate.\n\n## §1 What the corpus already holds (read, not recalled)\n\n- **Kialo's measured shape** (`kialo-deep-dive.md` §2): \"Most users read and vote; far fewer write claims (classic 90-9-1 internet participation inequality)\" — the nearest-neighbor platform confirms the textbook distribution on argument-mapping specifically.\n- **Reddit's conversion lever** (`adoption-problem.md` §4): voting is the low-friction rung, and upvoting measurably increases next-day activity — the documented mechanism for moving people ONE rung, never many rungs at once.\n- **The four-stage growth path** (`incentives-analysis.md` §6): solo use is the only incentive-clean entry (zero exposure cost); the private-to-public promotion path IS the growth mechanism. The eight legibility-pays conditions (L1–L8) name WHO climbs the gradient and why.\n- **Value scales with corpus coverage, not user count** (`lowering-the-cost.md` §5, Romer nonrivalry): this is what makes a steep gradient SURVIVABLE — Deliberus is a commons, and a library is justified by readers, not by how many readers write books (`conviction-and-critique.md` § the uplift-frame rejection).\n- **Machine extraction decouples content supply from human authorship**: Wikipedia's binding constraint (unpaid human writing) does not bind the same way — the corpus can grow from sources while the human gradient stays steep.\n- **Curiosity is the cross-cutting recruitment filter** (`curiosity-as-growth-fuel.md`): the disposition that predicts descent, orthogonal to community membership; PACE says each displayed gap must carry a move or it produces avoidance, which binds hardest at the gradient's shallow end.\n- **Agents as a consumer class** (`agents-as-a-consumer-class.md`): plausibly the largest \"reader\" tier of all, with its own gate (the strength layer) and its own incentive question (L8).\n\n## §2 The gradient, enumerated as tiers with priors\n\n| Tier | Action | Prior share of visitors | Motivations (from L1–L8 + curiosity + relational) |\n|---|---|---|---|\n| Readers | browse claims, follow a descent | the overwhelming majority (90%+) | curiosity, audit-need, being-pointed-here |\n| Reactors | vote agree/disagree, tap a badge | single digits | low-friction opinion, the Reddit rung |\n| Askers | open a critical question, follow a hinge | low single digits | deprivation-curiosity (an itch to scratch) |\n| Answerers | answer a CQ yes/no, add one evidence link | ~1% | L1 misrepresented, L5 audited-anyway, L6 self-argument |\n| Decomposers | split a claim, type a residue, challenge a terminus | fraction of 1% | L4 reasoning-as-product, L7 status, the workshop-recruited |\n| Gardeners | concept governance, merges, ontology care | a handful of people, period | mission-aligned; the operator + close collaborators for years |\n\nTwo design consequences, both already half-present in the corpus and now stated as rules: **every tier's action must be reachable ONE rung from the tier below** (read → vote → tap a question → answer it — never \"read → decompose\"); and **each tier must produce real value at that tier** — a reader who never contributes is a SUCCESS (the library test), a reactor's votes feed the discursive-dilemma instrument, and the decay-curve measure of pointing (`structure-versus-scale.md` § Operationalizing) applies PER TIER, not to some imagined uniform user.\n\n## §3 What this does to the falsifiers\n\nThe pointing test's honest baseline: expect momentary compliance from everyone, recurring engagement from a minority, and fine-grained contribution from almost no one — so the falsifier arm is the DECAY CURVE WITHIN TIER, not the tier populations. A session where both participants stay in the reader/reactor tiers but their engagement recurs and produces curiosity hits is a PASS for the substrate and a to-do for the affordances. A session where the decomposer tier is empty says nothing — it was expected to be empty.\n\n## Cross-references\n\n`incentives-analysis.md` (who climbs, and why) · `curiosity-as-growth-fuel.md` (the disposition + PACE gating) · `adoption-problem.md` + `kialo-deep-dive.md` (the measured shapes) · `lowering-the-cost.md` §5 (why steep is survivable) · `structure-versus-scale.md` § Operationalizing \"pointing\" (the per-tier measure) · `islands-of-coherence.md` §5 (the workshop tier)\n"}