Closed trading economy: mark-to-market settlement, no income floor, no transaction fee

Status: decided and implemented (#442, issue #407). model/contract.ts holds the term arithmetic this ADR fixes, Early Sell proration, the settlement delta, the Renewal Premium and the final-24h renewal window, and ContractService is the single money-writer that spends and credits against it. The daily sweep runs as the ContractSettlementWorkflow, described in Contract Settlement. The stipend and the transaction fee this ADR removes are absent from the code: there is no stipend left in team.ts to delete.

Two things it decided are still unbuilt, and are marked as such where they are stated: the Wealth Ceiling has not been re-derived for mark-to-market, and the economy's flow figures have not been re-simulated (docs/domain/scoring-system.md §6.2, §6.3).

Like ADR 0005, this is a single consolidated ADR, later revisions are folded in here rather than left as a chain of separate superseding documents.

The constraint that shaped it

Closed economy (system is always its own counterparty, no player-to-player market, deferred post-MVP) + real credit losses must be possible (otherwise nothing is actually at stake) + no permanent unrecoverable death spiral. The mechanics below are the current answer to reconciling those three.

Current model

Contracts are positions, not consumables. Buying an article isn't spending credits on a disposable good, it's allocating capital into a tradeable position that settles later, based on whether the article's value moved in your favor.

• purchasePrice: locked at signing, per ADR 0005.
• currentPrice: the same ADR 0005 formula re-evaluated with live 30-day-average views, at the contract's tier duration (D × BasePoints(liveViews)^k × tierDays, where tierDays = expireDate − purchaseDate for the current window). This is a "replacement cost" number: what this exact contract would cost fresh, today.
• Early sell (exiting before the committed term ends): payout = currentPrice × (remainingDays / tierDays), credited. You're paid only for the unused time, at today's rate. Proration is the sole guard (there is no minimum hold, removed below): holding 3 of 14 days recovers only 11/14, so a partial hold can never return the full price plus free bonus days of points, the exploit a naive full-currentPrice-on-any-sell payout would open.
• Natural term completion, not renewed (the "sold to system" case): you've already banked every day's points across the full committed term. The buy debited the full purchasePrice, so settlement returns the whole stake plus the mark-to-market P&L, i.e. the team is credited currentPrice(liveViews, tierDays), equivalently purchasePrice + (currentPrice − purchasePrice). Views rose over your whole hold → net profit on top of the points already scored; views fell → net loss. (Crediting only the currentPrice − purchasePrice delta, as an earlier draft of this ADR did, was a bug: it silently forfeited the stake, so a flat-views hold lost the entire purchasePrice.) Reachable only by holding the entire committed term, so the early-sell proration above can't reach it.
• Settlement runs on a daily Cloudflare Cron sweep on the backend Worker (~06:00 UTC, after the scoring engine has written the day's views): it finds ACTIVE contracts past expireDate with no renewal elected, fetches each article's 30-day-average views via the Wikimedia client, computes currentPrice, credits/debits, flips the contract to SETTLED, and writes the notification. Backend stays the single money-writer (ADR 0004); the sweep is idempotent on the status guard, so a re-run is a no-op.
• Renewal is elected in the final 24h of the term (right-of-first-refusal, reworded: no midnight sniping, no permanent lock): during the last 24h the owner picks Renew or let expire, and the choice locks for expiry (default = let expire). A renewal rolls the window forward, purchaseDate ← old expireDate, expireDate += tierDays, purchasePrice ← currentPrice + renewal premium, renewalCount++, so a contract's tier stays derivable from its own two dates and no tierDays column is needed.
• No transaction fee. The original 8% fee (churn/spike-arbitrage throttle) is redundant with two guards that already exist: pricing off the 30-day average (a 1-day spike barely moves it) and early-sell proration (a partial hold only recovers unused time). Combined with removing the income floor (next point), keeping the fee would make it the only guaranteed force in the economy, a pure drain that guarantees an average or below-average trader slowly bleeds credits even if they roughly break even on views. Removed.
• No base stipend. The flat 15/day income floor is a passive, patience-rewarding mechanic that sits at odds with this whole redesign's philosophy (grinding should mean trading skill, not just waiting, the same reasoning that chose the k=1.7 pricing exponent). Removed. Recovery for a broke player instead comes from the pricing curve's own zero floor: articles under 2,000 views genuinely price at 0 credits (ADR 0005), so a broke player can always fill all 11 slots for free and speculate on one rising in popularity, a skill-based comeback path (scouting undervalued content), not a guaranteed one. See "Open risk" below.
• Renewal premium retained (+10% per consecutive renewal, resets after dropping the article ≥1 cycle), this solves a different problem (anti-hoard) than the fee did (anti-churn), so it's unaffected by removing the fee.
• Minimum hold removed (was 3 days). Early-sell proration already makes a partial hold cost the used portion, so the separate churn block was redundant; and with 30-day-average pricing a same-day round-trip pays ≈full price back but scores ~0 points, so churn is economically neutral without it. (If spike-churn ever appears in playtesting, reintroducing a minimum hold or a small fee is the lever, see "Consequences".)

Why not just pay full currentPrice on every sell (early or at term)?

Considered and rejected: if any sell, early or at term, simply paid full currentPrice at the tier duration, buying LONG (14 days), holding only a few days, and selling at flat views would return a full refund (14-day currentPrice ≈ what you paid, if views didn't move), despite only holding a fraction of the term, free days of points for zero net cost. That makes LONG strictly dominant over SHORT/MEDIUM. The two-rule split (prorated for early exit, full settlement only on natural completion) closes this: partial holds only ever recover partial value. This is also why there is no separate minimum hold, proration, not a time gate, is what makes early exits fair.

Open risk (acknowledged, not yet mitigated)

Removing the stipend turns "no death spiral is impossible" from a guaranteed floor into a probabilistic one, a broke player recovers only if a free pick they made actually rises. A genuinely unlucky run (every free pick stays flat or falls for the rest of the league) has no guaranteed recovery under this model. Mitigating factors that exist by construction, not by explicit design: a broke player can fill all 11 slots with free articles simultaneously (diversified speculative exposure, 11 independent chances, not one), and synergy points (schema- adjacency chemistry) score regardless of an article's price or view count, so a fully-broke player's score never floors at literal zero. Whether this is sufficient, or whether some additional backstop is warranted, is a live-tuning item, call it out explicitly if playtesting shows players getting stuck.

Consequences

• Supersedes this ADR's own original income-floor-based model (flat stipend, 8% fee, no settlement mechanic, expiry previously had no defined payout at all).
• docs/domain/scoring-system.md §6.2/§6.3 (stipend, fee, wealth ceiling figures) needs updating to match: not yet done as of this ADR. The wealth ceiling concept (~2,400 credits under the old model) needs re-deriving under this economy; it's no longer just "cost of the top-11 team for a week" once settlement gains/losses are a factor.
• No backend implementation exists yet for renewal, expiry, or settlement: this is purely a design decision pending implementation.
• Grind-timeline figures (active-trader vs. passive-saver weeks-to-full-team) inherited from ADR 0005 are now doubly stale: they assumed both a stipend and a fee that no longer exist. Needs a fresh simulation once implemented.
• Player-to-player trading remains deferred post-MVP; the system stays its own sole counterparty for both buys and settlements.

Related

• Scoring & Economy System
• ADR 0005: Contract Pricing