Reputation and gamification mechanics

Part of the «Waze для рыбалки» wiki · Research stream · 2026-07-02 · Status: complete
Related: Design principles · Secret spots

==== TOPIC: reputation ====
SUMMARY: Across Waze, eBird, iNaturalist, Untappd, Google Local Guides and Stack Overflow, ongoing reporting is driven far more by status, reciprocity ("thank-you economy"), collection/completion instinct, and personal utility than by any material reward — and the strongest systems make the contributor's own record (logbook, life list, badge shelf) the reward. Reliability is handled two ways: consensus/peer-confirmation (Waze thumbs-up, iNaturalist 2/3 agreement, SO votes) and expert-curated filters + human reviewers (eBird), with the near-universal finding that crowd confirmation validates plausibility/persistence, not truth — Waze reliability scores did not correlate with actual accuracy. Anti-gaming leans on account-trust thresholds, velocity/pattern anomaly detection, location verification, and separating "fun" leaderboards from "trusted" data. The fishing domain is uniquely adversarial: unlike birders or beer drinkers, anglers have a direct incentive to lie — hide, downgrade, or actively misdirect on spot locations ("send them on a goose chase") — so honest location data is the scarce good and any design must reward truthful disclosure against that grain. The Latvian/Russian-speaking market already has spot-map incumbents (fish-search.ru, placefish.ru, copeslietas.lv, mansloms.lv), so differentiation must come from real-time freshness + trust mechanics, not just another catch map.

KEY FACTS:
- MOTIVATION — Waze: no cash reward; drivers report for points/ranks (Baby Wazer up to Royalty), avatar/mood customization, and especially a 'thank-you economy' — receiving thanks from other drivers — plus 'Epic Meaning' (small acts create collective safety). Curiosity loop: users re-check whether a hazard they reported is still there. Yu-kai Chou Octalysis analysis; waze.com rank/points docs.
- MOTIVATION — eBird: the personal life list / county-state-country Top 100 leaderboards and streaks are the hook. 8,280+ birders submitted 365+ checklists in 2024 for the Checklist-a-day Challenge. Crucially, eBird does NOT police these list totals and data-quality review does not affect leaderboard totals — the 'fun' metric is deliberately decoupled from the 'trusted' scientific data.
- MOTIVATION — Untappd: collection instinct is explicit. Badges for unique beers, styles, regions, venues, and streaks (same beer repeatedly) drive users to try things they'd otherwise skip; friends compete on badge counts. Pure gamified collection + social comparison, minimal utility.
- MOTIVATION — Google Local Guides: 10 levels, points per contribution (review up to 10, photo up to 7, new place 15, edit/answer 1), badge unlocks at Level 4 (250 pts). No money — only recognition, a visible trust badge next to your name, early feature access, occasional perks. Intrinsic/community motivation.
- MOTIVATION — anglers are the OUTLIER: strong cultural norm to keep 'honey holes' secret; 'unwritten code' that you don't ask another angler their spot; some deliberately post misleading photos/name nonexistent lakes to 'send inquirers on a wild goose chase.' Fishbrain's answer: 3 privacy levels per catch (exact position / water name only / hide all location) — and even so 'not too many users post their actual catch locations.'
- RELIABILITY COMPUTATION — consensus models: Waze uses drive-by 'thumbs up' (still there) / 'not there' (gone); thanks/upvotes reset the timestamp and extend a report's life, absence of confirmation lets it expire. iNaturalist 'Research Grade' = >2/3 of identifiers agree at species level AND community taxon matches observation taxon; IDs weighted equally toward the 2/3 threshold. Stack Overflow: net upvotes; can't upvote until 15 rep, downvote ratio capped after 300 votes.
- RELIABILITY COMPUTATION — expert-curated model (eBird): automated per-region filters (built/maintained by local expert volunteers, updated as often as weekly) flag any observation whose species/count/date/location exceeds expected values; flagged records enter a regional reviewer queue. Records become 'Accepted' (public, on maps) or 'Unconfirmed' (visible only to the submitter). Reviewers can't edit your data — only set visibility. Well-documented records (photos/notes/audio) clear in seconds.
- RELIABILITY LIMIT — crowd confirmation ≠ truth: peer-reviewed study of Waze vs. video ground truth found for disabled vehicles NEITHER the reliability score NOR incident duration correlated with actual accuracy; 23% of disabled-vehicle reports were false alarms. Crowd votes measure persistence/plausibility and social agreement, not ground truth.
- RELIABILITY LIMIT — iNaturalist accuracy: ~25% of 'Research Grade' observations found inaccurate/uncertain in some studies; for hard taxa ~70% of species-level IDs unique to iNat were wrong. Degraded by 'clickers' who mass-agree without expertise and by blind 'Agree' clicks inflating consensus falsely. Identifier experience is the main determinant of whether/when a record reaches Research Grade.
- ANTI-GAMING — account trust & thresholds: Stack Overflow blocks upvoting below 15 rep, caps downvote ratio after 300 votes, and silently deletes votes from users with anomalous voting patterns (restoring lost rep to the target). Voting rings / sockpuppets (alt accounts upvoting the main) are the canonical attack; detected via communities with abnormally high mutual interaction, fast reciprocal answers, and mutual accept patterns.
- ANTI-GAMING — velocity & pattern ML (Google Maps/Local Guides): flags review-velocity spikes (e.g., 50 reviews/24h), checks whether reviewer was ever physically near the place (IP/location), analyzes linguistic similarity (shared phrases/emoji), scores account history/trust. Removed 170M+ policy-violating reviews in 2023 (+45% YoY); ~75% of fakes caught before publishing; humans handle slang/edge cases.
- ANTI-GAMING — fabrication detection (iNaturalist): reverse-image search (Google Images/TinEye) catches images lifted from the internet; 3+ fabricated observations + unresponsiveness = suspension within 24h. Community flagging + Data Quality Assessment downvotes can demote records out of Research Grade without needing an ID war.
- ANTI-GAMING — location spoofing (core risk for a geo app): GPS is trivially spoofable (mock-location apps). Industry defenses: read the OS mock-location flag, sensor fusion (accelerometer/gyro consistency), cross-check GPS vs Wi-Fi vs cell-tower, and detect impossible travel speeds/teleports. A fishing app claiming 'caught here now' must treat raw GPS as untrusted.
- THEORY — truthful-reporting mechanism design exists: peer-prediction / Bayesian Truth Serum / Peer Truth Serum reward a report by how it agrees with a peer's independent report (and with the 'surprisingly common' answer), making honesty the strategic equilibrium WITHOUT ground truth. Directly relevant when reporters are self-interested and truth can't be directly verified.
- MARKET — Latvia/Russian-speaking incumbents already exist: fish-search.ru/latviya (22,000+ points, reports w/ photo/video, paid-pond bite status), placefish.ru ('where's biting now' community map + user-added points), Navirit 'Fishing: прогноз клева' (diary + spots + reports), oRybe.com (7-day Latvia bite forecast). Latvian-language communities: copeslietas.lv (Copes vietas forum), mansloms.lv, lielaisloms.eu, makskeresanaskarte.lv. Bite-forecast + static spot maps are commoditized; real-time trustworthy 'biting NOW' is the open gap.

IMPLICATIONS:
- Make the angler's OWN logbook the primary reward, not public spot-sharing. Every proven system (eBird life list, Untappd badge shelf, Fishbrain logbook) shows people log obsessively for private collection/record-keeping even when they won't broadcast. Capture the catch data first via private/personal value; treat public disclosure as a separate, incentivized opt-in — exactly Fishbrain's 3-tier privacy (exact pin / water-name only / hidden).
- Decouple the 'fun' leaderboard from the 'trusted' data layer, like eBird. Rank/badges/streaks can run on volume and engagement (checklists logged, species diversity, consecutive days) which are cheap to verify and low-stakes if gamed; keep a separate, harder-won reliability/trust score that governs whether a report influences the live 'biting now' map. Never let leaderboard hunger contaminate the map anglers actually rely on.
- Reward truthful disclosure against the anglers'-secrecy grain — this is the central design problem and where generic gamification fails. Options: reciprocity gating (see others' fresh spots only if you contribute verifiable ones — a 'give to get' unlock), time-delayed or fuzzed geolocation (share a 500m hex now, exact pin after 48–72h so the honey hole isn't instantly burned), and peer-prediction-style scoring that pays out when your report is later corroborated by an independent angler. Assume a meaningful fraction will deliberately misdirect.
- Treat raw GPS and self-reported catches as untrusted by default. Require or heavily bonus corroborating evidence — photo with EXIF time/geo, on-device capture (not gallery upload), reverse-image-search against known internet photos (iNaturalist's fabrication defense), mock-location-flag checks, and impossible-travel detection. A 'verified on-site' badge on a report should mean the app captured the evidence, not that the user typed a pin.
- Use crowd-confirmation for FRESHNESS/DECAY, not for truth. Copy Waze's mechanic: a 'biting here' report decays fast and is kept alive only by independent confirmations from other anglers passing through ('still biting' / 'nothing here'), with timestamp resets on confirmation. But remember the Waze finding — confirmation proves persistence/agreement, not accuracy; surface report age and confirmation count so users self-calibrate.
- Build a graduated trust/reputation score (à la Local Guides levels + SO thresholds) that gates power: new accounts can log privately but their reports carry low map-weight until they accumulate corroborated reports; unlock the ability to confirm/flag others' reports only above a reputation floor (SO's 15-rep gate). Weight a report's map influence by the reporter's historical corroboration rate, not just raw count.
- Instrument anti-gaming from day one, because a competitive-angling audience WILL sockpuppet and spoof: detect voting/confirmation rings (clusters of accounts always corroborating each other), report-velocity spikes, location-teleport patterns, and duplicated photos. Silently down-weight rather than publicly punish (SO model) to avoid arms-race feedback. Plan for regional human moderators — eBird's expert-reviewer network shows curation doesn't scale on volunteers alone but is essential for the trusted tier.
- Localize trust and content, not just language. Serve LV and RU audiences (the Russian-speaking angler segment is large and already served by fish-search.ru/placefish.ru); seed with regional 'filters' of what's plausible (species/season/size expectations per Latvian water body) so absurd reports (a marlin in Ķīšezers) auto-flag — eBird's per-region filter model adapts directly to fishing. Recruit respected local anglers as reviewers/ambassadors for legitimacy against entrenched forums (copeslietas.lv, mansloms.lv).
- Lean on the 'thank-you economy' and reciprocity as the emotional engine. Waze's most durable motivator is social gratitude, not points. Let anglers thank a report that led to a catch ('this tip worked'), and make that gratitude the primary input to reputation — it aligns status with genuinely useful, honest reports and is far harder to fake convincingly than raw check-in counts. Collection badges (species caught, waters fished, seasons completed) satisfy the completion instinct that keeps eBird/Untappd users logging for years.