FantasyWiki, Definitive Scoring & Economy System
This document supersedes the scoring portions of docs/domain/fantawiki-requirements.md §2, §3, and §6. It was derived by stress-testing the original design against three principles and validating every number against real Wikipedia pageview data (en.wikipedia top-1000, snapshot 2026-06-07).
Rationale for the load-bearing choices lives in:
- ADR 0001: base scoring model
- ADR 0002: language scale factor
- ADR 0003: closed trading economy (mark-to-market settlement, no income floor, no fee)
1. Design principles
Every decision below was judged against three principles, in priority order:
- Explainability, a casual-core player can predict their score in their head. The audience is 18–45 casual-core; legibility beats mathematical elegance.
- Contract balance, price, points, and budget compose so that no single strategy (buy-the-biggest, hoard-one-hub, churn-spikes) is strictly dominant.
- Diversity in the META, high-traffic, mid-traffic+synergy, and niche+event strategies are all viable. Validated by team-vs-team simulation, not assertion.
2. Language normalization
Competition is always same-language, but the scoring constants must be calibrated per language or the model breaks on low-volume Wikipedias (Italian articles bunch into one score bucket; flat synergy points become dominant). See ADR 0002.
- Language Scale Factor
L: a static per-language constant; referenceL_en = 1.0, lower-volume languages getL > 1. Measured values live in thelanguage_scalesregistry (seededen = 1.0,it = 13.9), and each league carries the factor it was founded on inleagues.languageScale, that frozen copy, not a lookup, is what every price and score in the league uses. An edition with no measurement cannot host a league: creation measures it first and refuses it if it fails the floor below. - Normalized Views =
raw_pageviews × L. All scoring below operates on Normalized Views, never raw views. L(domain) = median(en_views[i] / domain_views[i])for rank-matchedi = 1..500, using each domain's 30-day-average views (not a single day, which is measurably noisy) and content-article ranks only (namespace pages excluded via that domain's ownsiteinfo, not a hardcoded prefix list). Recalibrated ~annually (no formal season). A domain is only accepted (league creation allowed) if it has ≥300 ranks with ≥50 daily views, below that, the shape-similarity assumption behind a single scale factor stops holding, counted per day and medianed across the window. Full detail and real-data validation: ADR 0002 and Wikipedia Language Editions.
3. Base Points (view-driven daily score)
A rule-based geometric (continuous log) model with a convex top tail. One sentence for players:
"Every time an article's views double, it earns one more point, starting at 1 point for 4,000 views. Above 150,000 views, each extra 50,000 views adds a point."
┌ max(0, log₂(NormalizedViews / 2000) ) if views ≤ 150,000
BasePoints =┤
└ 6.23 + (NormalizedViews − 150,000)/50,000 if views > 150,000 (linear tail)log₂(views / 2000) is identical to the earlier log₂(views / 4000) + 1 form, the +1 is folded into halving the divisor. It is a continuous function and returns decimals, not integer tiers: the "+1 per doubling" sentence is only the headline at the doubling rungs. The max(0, …) clamp is a floor at zero score, not integer rounding, and it crosses zero at 2,000 views (one doubling below the 4,000 anchor), so the 2,000–4,000 band earns fractional credit (e.g. 2,828 views → 0.5).
| Doubling rung | Normalized views | Base points |
|---|---|---|
| zero | 2,000 | 0.0 |
| (sub-anchor) | 2,828 | 0.5 |
| anchor | 4,000 | 1.0 |
| 8,000 | 2.0 | |
| 16,000 | 3.0 | |
| 32,000 | 4.0 | |
| 64,000 | 5.0 | |
| 128,000 | 6.0 | |
| kink | 150,000 | 6.23 |
| tail | 300,000 | 9.23 |
| tail | 476,000 | 12.75 |
Validated against real en.wp data (2026-06-07): rank ~1000 (9k views) → 2.24, giants (80–130k) → 4.3–5.0, top-3 (243–476k) → 8.1–12.8. Top-to-bottom spread ≈ 5.7×.
Why this shape:
- Concave mid-field: diminishing returns so virality doesn't run away; log-binning is the standard treatment of power-law pageview data.
- Convex tail above 150k: deliberately rewards the volatile daily top ~10 (catching a breakout is a large point swing). This is the only place viral strength is reintroduced; it is contained by price, not by the curve (see §6).
- 2,000 zero point: articles below 2,000 views score 0 on base and live on synergy (the niche archetype); the 2,000–4,000 band earns a fractional 0–1 (continuous, not a step), so a sub-4k article is not flatly zeroed.
4. Synergy (additive chemistry)
Chemistry is additive flat points on the schema-adjacency topology, not a multiplier and not computed over every owned pair. It is evaluated only between articles placed on positions joined by a Chemistry Link in the formation schema (16–21 links depending on schema; see dto/enums.ts CHEMISTRY_LINKS).
Per-link value, summed to a team total:
| Level | Condition (two placed articles) | Points |
|---|---|---|
| Excellent | mutual Wikipedia link (A→B and B→A) | +1.5 |
| Good | one-way link (A→B xor B→A) | +0.5 |
| Weak | both placed, no link | 0 |
| Empty | a slot is unfilled | 0 |
TeamSynergy = Σ over schema links ( value(level) ) # mutual = 3 × one-way- Mutual links are worth 3× one-way (preserves the original "reward two-way relationships" intent).
- No per-article cap (the ~16–21 link count is the natural ceiling); an optional team-level cap ≈ 20 guards degenerate all-mutual builds.
- For per-article display, split each link 50/50 between its two endpoint articles; the team total is identical.
Calibration (E = 1.5). Simulated against real data at a fixed 1,000-credit budget, a fully-synergized thematic cluster (which pays a "base tax" because you can't field 11 same-theme giants) draws level with the budget-optimal homogeneous team at E ≈ 1.5, and a cluster that also spends spare budget on 1–2 themed anchors tips ahead. Below 1.5 synergy is a minor tiebreaker; above it, themes dominate.
5. Daily and cumulative scoring
ArticleScore(day) = BasePoints(NormalizedViews_today) # §3
TeamScore(day) = Σ ArticleScore + TeamSynergy + EventBonus # §4, §7
Cumulative = previous cumulative + TeamScore(day)- Scoring uses today's (daily) Normalized Views: volatile and exciting.
- Pricing (§6) uses the 30-day average: smoothed and stable. This decoupling is the heart of the meta: a spiking article scores big today but costs almost nothing, while a sustained performer is expensive.
6. Contract economy
A closed trading economy with mark-to-market settlement (ADR 0003). Players build wealth by trading, holding a contract to its full committed term settles as a real gain or loss against what they paid, not just a spend. There is no flat income floor; a broke player's comeback path is always-available free (0-credit) articles, not a guaranteed stipend. Each league resets.
6.1 Pricing
Canonical: ADR 0005 (single consolidated ADR, supersedes both the original /1000 × weeks linear formula and an intermediate convex-in-raw-views formula that was tried and replaced after live data showed it over-concentrated price differentiation at the extreme head of the distribution):
ContractPrice = D × BasePoints(Normalized_30dAvg_Views)^k × contract_daysBasePoints here is the exact same curve as §3 above (log-compressed, same 2000 zero-point and 150k kink), just fed the 30-day average instead of daily views, one curve, one place to tune the shape. k (locked at 1.7) and D are pricing-only constants, derived from a full-11-giant-team budget target rather than from a legacy low-band anchor (see ADR 0005 for why anchoring at the low end no longer works once price is points-based). Price is convex in points, not in raw views, giants and top-tier articles still cost progressively more per marginal point than mid-tier articles do, which is what stops a day-one player from affording a top-tier team, but without the price curve diverging from the value it buys the way a views-based exponent did.
- Priced on the 30-day average, never daily views: the anti-jackpot guard. A 1-day spike barely moves the 30-day average, so you cannot buy-at-10k / sell-at-400k overnight; trades only pay off on sustained multi-week momentum.
purchasePriceis locked at signing;currentPricefloats with the live 30-day average, evaluated at the contract's tier duration. It's the basis for both exit paths, prorated (unused time) for an early sell, or a full-currentPricemark-to-market settlement at natural expiry (buy debitedpurchasePrice, so expiry returns stake + P&L; §6.3, ADR 0003).- Contract duration tiers are locked in days (ADR 0005): SHORT = 3 days, MEDIUM = 7 days, LONG = 14 days. These are the actual purchasable durations, the
contractDTO.tiergetter derives the display tier from an existing contract's duration; this is the forward (tier → days) mapping the buy flow uses.
6.2 Budget, slots, and the wealth ceiling
- Starting budget: 1,000 credits. The squad cap is stated once, in Lineup Rules: 22 active contracts, of which 11 can be placed. Both §3.1's "10" and the "11 contracts" this section used to state are superseded.
- Wealth ceiling: previously ~2,400 credits (cost of the top-11 team for one week) under the old income-floor economy. Stale, pending re-derivation (ADR 0003): under mark-to-market settlement, wealth isn't just credits spent so far, it includes unrealized settlement gains/losses on every held contract, so the ceiling has to be redefined in terms of position value, not spend. Not yet re-derived.
6.3 Income, sinks, and frictions
Canonical: ADR 0003 (single consolidated ADR, supersedes the original income-floor model: flat stipend, 8% transaction fee, and an expiry with no defined payout are all removed/replaced).
| Lever | Value | Role |
|---|---|---|
| Renewal premium | +10% per consecutive renewal (resets after dropping ≥1 cycle) | anti-hoard (sink) |
| Early sell | currentPrice(liveViews, tierDays) × (remainingDays / tierDays) credited | pays only for the unused time at today's rate, proration is the sole anti-exploit guard (no minimum hold): holding 3 of 14 days recovers only 11/14, so a partial hold can never return the full price |
| Expiry settlement ("sold to system") | credit currentPrice(liveViews, tierDays), i.e. purchasePrice + (currentPrice − purchasePrice) | full-term mark-to-market: the buy already debited purchasePrice, so expiry returns the whole stake plus the P&L (currentPrice − purchasePrice). Net gain if views rose, net loss if they fell. Only reachable by holding the entire committed term |
| Renewal decision | owner elects Renew / let expire during the final 24h of the term; the choice locks for expiry (default = let expire) | right-of-first-refusal without midnight sniping or a permanent lock; a renewal rolls the window forward (purchaseDate ← old expireDate, expireDate += tierDays) at currentPrice + renewal premium |
| Settlement trigger | daily Cloudflare Cron on the backend Worker (07:00 UTC, two hours after the scoring collector), which starts the durable ContractSettlementWorkflow; 30-day-average views fetched per contract via the Wikimedia client | backend stays the single money-writer (ADR 0004); one retryable step per contract, made idempotent by guarded writes rather than by not retrying, Contract Settlement |
| Broke-player recovery | free (0-credit) sub-2,000-view articles always available | skill-based comeback (scout undervalued content), not a guaranteed stipend |
Removed vs. the original model:
- Base stipend (was 15 credits/day): removed: a flat income floor is passive and patience-rewarding, at odds with pricing built entirely around active-trading skill (ADR 0005's k=1.7). Recovery now comes from the pricing curve's own zero floor instead. Trades a guaranteed no-death-spiral floor for a probabilistic one, an explicitly acknowledged, not-yet-fully-mitigated risk (see ADR 0003's "Open risk").
- Transaction fee (was 8% of sale proceeds): removed: redundant with the 30-day-average smoothing plus early-sell proration, which already do the anti-churn/anti-spike-arbitrage work the fee existed for; kept alongside a removed stipend it would've been the one guaranteed drain in the economy.
- Minimum hold (was 3 days): removed: early-sell proration (
× remaining/tier) already makes a partial hold cost the used portion, so the separate churn block was redundant. With 30-day- average pricing a same-day round-trip pays ≈full price back but scores ~0, so churn is economically neutral without it.
Self-balancing properties:
- Even view-budget allocation beats viral concentration (Jensen): pricing convex in points, not raw views (ADR 0005), giants and top-tier articles are progressively less credit-efficient per point than mid-tier ones, so spreading budget is always favored over concentrating it.
- The system is always its own counterparty (infinite free agents), so the economy closes with as few as 3 players. Player-to-player offers are deferred post-MVP.
- Superseded pending re-simulation (ADR 0003 + ADR 0005): the day-one optimal team, the "~60 base points" / "~42.5 base points" comparisons, and the "+85/week, 1,680 passive / 2,300 skilled-trader over 8 weeks" flow figures were all calibrated against the old linear-then- views-convex pricing and the now-removed stipend/fee. None of these numbers are current, re-simulate once the mark-to-market economy is implemented.
Live-tuning levers: players stuck broke too often → revisit the no-guaranteed-floor trade-off (ADR 0003's "Open risk"); spike-churn appears → reintroduce a minimum hold or a small fee (both currently removed); giants hoarded → steepen renewal escalation.
7. Weekly events, DEFERRED (post-playtest iteration)
Decision: events are intentionally held back until the core scoring loop (§3–§6) has been validated with a test group. Calibrating event bonuses on top of an untested base would mean tuning two unknowns at once; the base must be confirmed against real player behavior first.
The four §2.3 events (Momentum Surge, Low Views Protection, Double Synergy, Cluster Bonus) are not yet recalibrated to the new scales and have known issues to resolve when they are picked back up:
- Flat bonuses (+4/+3/+1/+5) were sized against the old base scale and must be re-expressed against the new Base Points (top article ≈ 12.8, giant ≈ 5).
- Double Synergy now means E 1.5 → 3.0; re-check it does not dominate.
- Thresholds in raw views (
<500 views/day, top-50 global) must move to Normalized Views to stay language-fair. hash(user_id + week_number) mod 4gives each player a different event the same week, which contradicts "1 randomized event per week" and hurts explainability, decide per-player vs global-per-week.
8. Summary of changes vs Requirements §2 (v5.4)
| Area | Was (§2 v5.4) | Now |
|---|---|---|
| Synergy mechanic | additive over all owned pairs (0.75/0.25, cap 3.0) and a position multiplier in the user story (contradictory) | additive flat points on schema-adjacency; excellent +1.5 / good +0.5; weak & empty 0 |
| Base scoring | 3 fixed tiers, knees 5k/20k | log-binned "+1 per doubling" from 4k, convex tail above 150k |
| Per-language | none | single Language Scale Factor on views (en = 1.0) |
| Contract pricing | two contradictory formulas (§3.1 vs §6.1) | C × Normalized_30dAvg^1.5 × days (§6.1, ADR 0005, convex, days-based, supersedes the earlier linear-weeks form) |
| Economy | implicit/undefined | closed trading economy + flat stipend floor + fee/renewal sinks + reset |
| Max contracts | 10 (but formations need 11) | 11 |
| Income | none | 15/day stipend; 8% fee; +10%/renewal; 3-day min hold |
9. Still open
- Weekly events recalibration (§7).
- Dynamic resale price curve detail (
currentPricemovement model) and the spike-arbitrage edge cases, the broad guard (30-day-avg pricing + fee + min hold) is locked; fine balance is a live-tuning item. - Player-to-player transfer market (offers, time-bounds): post-MVP.
Exact: resolved (ADR 0002, 2026-07-07) and implemented (#532, 2026-08-17): median rank-matched top-500 ratio on 30-day-average views, domain accepted only above a ≥300-ranks-@-≥50-views floor, measured at league creation and frozen onto the league.Lvalues per language and the recalibration procedureLis measured per edition on demand rather than tabulated by hand; see Wikipedia Language Editions.- Recalibration cadence. ADR 0002 says ~annually, but
itdrifted ~17% in six weeks (13.9 → ≈11.5 between 2026-07-06 and 2026-08-15). Frozen factors mean this changes nothing for leagues already running, which is the point, but whether a new league should be founded on a year-old measurement is now an open question with a number attached to it. Contract duration bounds: resolved (ADR 0005): SHORT = 3 days, MEDIUM = 7 days, LONG = 14 days, locked.- Re-simulate §6.3's economy flow (passive/skilled-trader credit trajectories) under the new convex pricing formula, the old linear-formula figures are stale (ADR 0005).
Related
- Chemistry Links
- FantaWiki Requirements (Game Design Document v5.5)
- ADR 0001: Base Scoring Model
- ADR 0002: Language Scale Factor
- ADR 0003: Closed Trading Economy
- ADR 0005: Contract Pricing
- Standings and Podium: what the cumulative total becomes on screen
- Contract Settlement: how §6.3 is carried out each night