Mandate 2038: Design decisions

Rules reference: 0.8.0-rc.17-test Status: current rationale and implementation-boundary ledger

This document explains why the selected game has its present shape. It does not repeat card text or resolve play: core-rules.md is the only authority for the physical game.

The former complexity-proposal document was folded into this ledger after its five selected proposals were implemented. Earlier 19-tile, every-Era Realignment, Transmission, Tactic, secret-objective, and endgame-scoring assumptions are historical only.

Product identity

Mandate 2038 is a competitive institutional strategy game that also produces a shared future. Its durable pillars are:

New systems are frozen until the controlled four-player test supplies evidence.

Supported player counts

The selected product permits two through six players and suggests three, four, and five. Four players is the authoritative balance configuration because it preserves negotiation, spatial competition, and clean parity while keeping sequential resolution bounded. Three and five players remain complete games, not variants: every selected four-player change must be rerun at both counts and rejected if it creates a credible integrity, faction, seat, strategy, negotiation, or congestion regression. Two and six players are playable exploratory configurations, but their reports are non-promotional diagnostics until they receive their own evidence contract.

Selected complexity contract

Select → Move → Act

Every ordinary turn uses one visible grammar. Cost, destination, mode, risk, and exhaustion belong on the Action surface. Readying changes a future choice; it does not create an immediate action. Agent Swarm is the only compound-action exception.

This preserves difficult opportunity costs while reducing procedural lookup. The challenge should come from choosing among scarce capabilities, not from discovering a new resolution sequence on every card.

One visible exception per authority layer

An ordinary action may combine its printed rule, one destination bonus, one faction modifier, and applicable global state. Persistent Headlines use physical reminders. Every exception has a timing window. A current Headline overrides an older global effect only when both alter the same field.

Tactics remain a deferred module because the baseline already contains enough variation to test the central engine.

Eras teach the game

Later systems remain visible from setup, but the Current Era panel is the sole unlock authority.

Presentation cannot create a new phase

The current player board may group End-of-Era Resolution into four visual bands—Power for Generate, Trade, and Allocate; Economy for Produce and Partner; Dossier for the simultaneous secret filing; and Consequences for Audit and Mandate. This is a presentation hypothesis, not a rules change. Production remains the five numbered boxes, followed by the separate Audit and Mandate phases. Human testing must justify the visual grouping before it becomes a selected component layout.

Mandate stays public

Customers, Capability thresholds, Trust thresholds, Era Mandates, Fusion, and faction awards score when they occur. Facilities and control create position and production rather than receiving automatic endgame points. Final calculation applies the offline penalty, names the provisional Mandate winner, and then resolves supported Dossier claims deterministically.

This makes negotiation legible and lets every simulated score change retain a specific source.

Selected spatial contract

The game uses nineteen tiles in a complete radius-two hexagon:

The inner ring contains Research, Cloud, Foundry, Capital, Talent, and Grid. The outer ring contains two each of Research, Cloud, Consumer, Media, Government, and Renewable. Every non-Frontier tile has two Facility spaces. The complete ring preserves exact sixfold geometry, removes empty perimeter positions, and keeps every action category reachable.

Power delivery is profile-specific. Default Game uses local Power: a Generator can power its owner's Facilities on its own hex or an adjacent hex. Advanced Play gives each player one Infrastructure Network; adjacency and two Link tokens govern pooled Power delivery. There is no Network production bonus or separate Transmission graph.

The integrated starting-grid identity exists from setup. Generators and Mega-Clusters unlock in Era II. Links and binary Network connectivity are Advanced Play additions that also unlock in Era II.

Default Game keeps the randomized jurisdiction fixed for all four Eras. Jurisdictional Realignment is preserved as part of the single bundled Advanced Play profile: its blind vote rotates a physical ring while every site-bound component travels with its district. Players then receive all three Era IV actions to respond. There is no postgame rotation.

Fusion deliberately occupies one of the Grid tile’s three Generator slots. Filling those slots can deny Fusion; the first controlled test must measure whether that creates meaningful spatial competition or an accidental lockout.

Selected interaction contract

In Default Game, one immediate resource trade may occur during any player’s resolution: a complete named offer is accepted or rejected. Each player may make one Production Power purchase request. Advanced Play is one bundled profile, but its constituent changes are separately identified in the rule-change register below. Counteroffers and third-party claims are distinct: a counteroffer can be accepted or rejected by the original offer maker; claims let other eligible players compete for that published counteroffer. Influence is required for persistent Joint Ventures and Trust manipulation. Power is bought immediately during Production from a consenting adjacent Network; no Power contract persists.

Influence is an action, not a component supply. The acting CEO or Team must end at Media, Government, or Capital to change Trust or Scrutiny. Political control counts CEOs as two presence and Teams and Facilities as one; ties remain uncontrolled. A Joint Venture must be proposed or terminated by an acting piece at one of the player’s Facilities. Joint Ventures use numbered neutral host pairs. Their identities survive Realignment; operation still depends on printed range, Power, and connectivity.

Promises about later turns are not binding. The game supports negotiation without requiring a general contract-enforcement system.

Rule-change register

The generated rule-change register is the single status record for Default/Advanced changes. Its structured source records each change’s decision, implementation state, dependencies, and module IDs. It does not authorize players to assemble ad hoc profiles: Default Game and Advanced Play remain the only supported profiles until a new profile is deliberately selected and validated.

Future complexity reductions are governed by the complexity-reduction-protocol.md. The August 2026 simplification package is current: one location-defined Generator, presence-only politics, two programs per Faction, removal of seven stored-token families, and a stricter Default/Advanced boundary. These are baseline rules, not optional modules, so they do not appear in either profile’s module list. The register records them as accepted current decisions.

Selected risk and ending contract

Audit draws scale by player count. Scrutiny beyond a player’s ten physical cubes immediately becomes a Runway-or-Trust penalty. Era IV Audit converts risk into Runway or Mandate loss so late exposure remains consequential.

AGI is a hidden four-Era Dossier rather than an Action or deterministic declaration. After each Production, every player secretly files Commit or Hedge for that Era. In Era IV all cards are revealed: every Commit costs one Compute and adds one Scrutiny before the final Audit; Publication must be committed and the full cost paid for a claim to be eligible.

The ordinary final-scoring winner, including the printed tie breakers, is the provisional winner. The Audit bag is then rebuilt with all eighteen black no-AGI cubes plus each eligible claim's faction-coloured prediction tokens. Two cubes are drawn without replacement. Matching faction colours form AGI and make that claimant the sole winner; every other pair preserves the Mandate result. Claim tokens come from committed Dossier modules, Capability thresholds, and a bounded Mandate-rank comeback bonus. This produces zero or one declaration without a die, app, published percentage, or player-authored arithmetic. The ending is Open only when collective Trust improves beyond setup by the player count and unresolved Systemic Risk remains below the player count; otherwise it is Closed. Winning and building the preferable future are deliberately different achievements.

Selected content boundary

Dovetalis Labs, Loopfold AI, Mirevanta Works, Kestralyn, Orisonix, and Corthaven are the canonical fictional institutions. Stable institutional IDs remain internal compatibility keys for simulations, saved games, and balance evidence; they are not player-facing aliases.

The institutions and their CEOs are fictional. They must not imply endorsement or turn a Headline into a factual accusation. Commercial publication remains a separate decision requiring appropriate legal review.

The tone is solemn institutional absurdity. Early controversies must have credible benefits and harms. Later effects may become polarized and alarming, but the dystopia must emerge from defensible local decisions.

Rules titles, display titles, rules text, flavor, and art direction remain separate fields. Flavor never creates a mechanic.

Baseline exclusions

The following material exists as design inventory but is not part of the controlled 0.8.0-rc.17-test test:

Evidence from a run that includes an excluded module must name that variant.

Implementation status

Executable game 0.14.16 implements 0.8.0-rc.17-test under engine coverage nineteen-hex-simplified-v1. Each Production replaces the prior Power allocation; its cubes remain on the map as the authoritative powered/offline snapshot for later rules. The executable uses the complete radius-two board, six shared Programs, Research Protection, a forty-card Training deck, solo Mega-Clusters, one direct 1-for-1 trade, automatic Audit penalties, binary Advanced Networks, and deterministic evidence-backed Dossier resolution. Browser-native deterministic play, server-backed LLM play, replay, policies, and Monte Carlo share that contract. The published root now presents the playable game as its primary action while keeping the complete generated documentation and component galleries available from the same index.

Candidate 0.8.0-rc.17-test changes no physical mechanic from rc.16. Executable 0.14.16 binds the published root to the canonical browser icon so the live navigation and game journey complete without a failed asset request.

Candidate 0.8.0-rc.16-test changed no physical mechanic from rc.15. Executable 0.14.15 published the complete browser-module closure required by the selected-rules game. Physical Chrome must populate both setup selectors and begin a game without a failed module request before deployment.

Candidate 0.8.0-rc.15-test changed no physical mechanic from rc.14. Executable 0.14.14 corrected the generated decision prose for the Public Capability Covenant and Reorganization, removed unreachable joint-funding choices from the solo Mega-Cluster contract, and made the deferred cooling corridor remain explicitly wartime. Every current player-facing source is regenerated from the content graph. No balance or physical-play evidence transfers merely because the prose and published navigation changed.

Candidate 0.8.0-rc.14-test changes no physical mechanic from rc.13. Executable 0.14.13 preserves every stable component ID, rule, Era count, and Headline count while extending the canonical four-Era history. Progress now includes useful cybernetics and subscription biology. Capacity includes locally compliant robot congestion, biological utilities, and a jointly owned water bridge between states that remain at war. Authority adds licensed-organ testimony and machine-readable pollinator corridors. Continuity makes those systems a metropolitan nervous system and a living watershed seeking standing. The lore contract binds each idea to an existing component or Era panel. No balance or physical-play evidence transfers merely because the prose changed.

Candidate 0.8.0-rc.13-test changes no physical mechanic from rc.12. Executable 0.14.12 preserves every stable component ID, rule, Era count, and Headline count while adding the Billion-Instance Bloom to Authority and making its sensor license the Continuity precedent for self-replicating maintenance. The lore completeness contract now checks Era panels, actions, resources, locations, board motions, training faces, power contracts, factions, Headlines, Mandates, Programs, reference aids, world copy, endings, and every deferred card surface. No balance or physical-play evidence transfers merely because the prose changed.

Historical candidate 0.8.0-rc.12-test changed no physical mechanic from rc.11. Executable 0.14.11 made the Thematic Content Bible the sole lore authority. The former scratchpad's placement decisions, research boundaries, and retained backlog moved into the Bible; the parallel source and orphaned rendered page were removed. All twelve deferred Tactics, twelve reserve Specialists, and eighteen Secret Objectives expressed the same four-Era causal history without entering baseline play. The playable browser and documentation index adopted the current box premise.

Historical candidate 0.8.0-rc.11-test changed no physical mechanic from rc.10. Executable 0.14.10 preserved every stable Headline and Program ID, rules effect, profile boundary, Era count, card count, and unlock timing while completing the selected four-Era narrative across the world primer, endings, Programs, Mandates, faction surfaces, companions, and published galleries. The retained internal Program ID open_weights now presents as Public Capability Covenant in Authority; strategic Open Weights remains a Progress Headline. Firebase publication copies only graph-declared runtime artifacts, preventing ignored legacy projections from remaining publicly addressable. Existing balance evidence remains evidence for the unchanged mechanics; it does not validate the new prose, tone, or commercial presentation.

Mechanics projection contract v2 also removes Headline newswire, quotation, labels, and other presentation-only fields from the mechanics fingerprint. The historical v1 fingerprint included those fields, so its digest cannot be compared directly with v2. Reprojecting the 0.14.8 and 0.14.9 rulesets through v2 produces the same mechanics digest. Ruleset, playtest-kit, content, and engine identities still change and remain separately attributable.

Historical candidate 0.8.0-rc.10-test and executable 0.14.9 preserved every stable Headline ID and mechanic while replacing the four Era panels and all twenty-four Headline presentations. That release established the selected arc but left the Era III Program name and several published narrative surfaces inconsistent with it.

Historical candidate 0.8.0-rc.9-test and executable 0.14.8 aligned the release declaration and version-independent contract tests without changing a setup value, Core Action, score, component, or Headline mechanic.

Historical candidate 0.8.0-rc.8-test and executable 0.14.7 carried exact profile-artifact paths and byte hashes into the unified balance matrix, retained the complete executed ecology, and rejected source/profile identity mismatches before a holdout started. No setup value, Core Action, score, or component changed.

Historical candidate 0.8.0-rc.7-test and executable 0.14.6 let strategy evolution inject exact frozen opponent profile artifacts. It validates profile identity, rejects duplicate and unknown overrides, executes the complete substituted profiles, and records their source paths, byte hashes, snapshots, and strategy fingerprints. This closes the gap between a training ecology's opponent IDs and the exact policy versions used by its intended holdout. No setup value, Core Action, score, or component changes.

Historical candidate 0.8.0-rc.6-test and executable 0.14.5 made strategy evolution distribute each focal-seat match budget across every circular opponent window. It can calibrate three-, four-, and five-player fields together against their separate neutral shares, ranks the largest target miss before mean miss and fitness, and records every roster window in the report. The historical fixed-window mode remains available only to reconstruct diagnostics. This engine correction changes no starting resource, Core Action, scoring rule, or physical component.

Historical candidate 0.8.0-rc.5-test and executable 0.14.4 made every simulation placement measure follow the authoritative institutional winner, including an AGI declaration that replaces the provisional Mandate result. Mandate remains a separately reported score. This corrects matchup, rank, supplier-finish, and paired-rule evidence without changing any playable number or action.

Historical candidate 0.8.0-rc.4-test and executable 0.14.3 added profile-scoped Action and Mandate-source telemetry. Their reports retain that identity; reports using raw Mandate as final placement require the disposition recorded in their tracked receipt.

Historical 0.13.7 / 0.7.0-rc.13-test evidence retains its exact old identity. It does not qualify this redesigned map, component set, action eligibility, Audit, or endgame.

Synchronization is implementation proof, not balance proof. Structured balance numbers remain hypotheses until a tracked study receipt replaces their provenance.

The 0.11.0 / 0.7.0-rc.4-test correction made eligibility explicit: select only an Action that resolves from current state or can become legal through one accepted immediate trade before Act. Simultaneous target loss and a rejected required trade still create real commitment risk. Purely impossible choices no longer masquerade as strategy. The same executable evaluates an immediate trade as one complete exchange, creates post-Act offers from the post-action resource state, and executes declared persona partner, placement, and resource preferences. The v10 Codex session motivated this correction but does not establish balance or human teachability.

The 0.13.0 / 0.7.0-rc.6-test candidate replaces the rejected fixed gate with four secret Era Dossier choices and the physical Prediction Bag. It also removes Declare AGI from the Escalation deck and removes persistent Grid-Ready state. The executable behavior is synchronized, but Dossier bluffing, final Audit pressure, perceived fairness, and AGI frequency remain unmeasured.

The 0.12.0 / 0.7.0-rc.5-test fixed-gate candidate is retained only as historical evidence. It used a five-percent gate and fourth-power Mandate selection; those rules are not part of the current game.

The 0.10.2 / 0.7.0-rc.3-test alignment patch changes no mechanic or number. It makes the selected integrated physical state authoritative in the machine-readable player supply, distinguishes Default local Power from Advanced Networks in runtime copy, and synchronizes component inventories, profile language, and the explicitly unmeasured complexity forecast. The renamed supply and round fields are contract corrections, not new pieces or new procedures.

The 0.10.1 / 0.7.0-rc.2-test precision patch changes no mechanic or number. It prints final Generator prices on their only legal Energy districts, removes the obsolete Energy discounts, describes Faction programs and Power eligibility without assuming the Advanced Network profile, and uses Escalation availability on one faction-board track everywhere. It also removes duplicate loose-component descriptions created by the selected double-sided and track-based physical forms.

That historical synchronized rules candidate promoted the accepted simplification package. It removes one Generator and all Influence cubes from every faction set, reduces printed Faction programs from twenty-four to twelve, and removes Market Access, Build discounts, Policy Shields, Economic Benchmarks, Experts, Spotlight, and Public Research Grants. Useful effects resolve immediately as Runway, Compute, Trust, Scrutiny removal, movement, Team recruitment, or Mandate. This synchronization proves implementation integrity only; new human and paired simulation evidence must establish balance, negotiation quality, and teachability.

The 0.9.1 / 0.6.0-rc.2-test advancement also changes no physical rule. It corrects the executable selection contract so every unused Core Action and every unlocked, unspent Escalation remains legally selectable even when it has no current resolution. Resolution metadata lets deterministic policies avoid known dead choices without deleting them. A blocked selection exhausts, and a blocked Escalation spends its availability, exactly as the physical commitment rule requires. Because legal decision packets and deterministic sampling change, earlier simulation remains historical evidence rather than balance authority for this executable.

The 0.9.0 / 0.6.0-rc.1-test advancement creates a new Default Game: one-offer immediate trades, one Production Power request per player, and no Era III Realignment. Advanced Play restores all three former procedures as one profile. This is a mechanical revision requiring new three-, four-, and five-player evidence; 0.8.35 and earlier reports remain historical full-rules evidence and do not qualify the new Default Game.

The 0.8.34 / 0.5.0-rc.34-test advancement changes no legal action, cost, resource value, Mandate award, faction ability, setup, or end condition. It adds simulation-only Deal Flow conversion attribution and a fingerprinted deterministic Coalition policy treatment for a preregistered diagnostic. The physical-candidate suffix advances only to preserve immutable synchronized release identity; it is not a balance-rule revision.

0.5.0-rc.20-test selected the historical clarification that ring rotation was district movement rather than Facility relocation for the then-current Grid-Ready rule, the two-source Power inventory, and the evidence-selected rule that Everybody Gets a GPU scores one Mandate per four rivals. It promotes four controlled conversion changes from the winning-path tolerance confirmation: Demis scores less public Mandate at Capability nine and twelve except for five-player Capability twelve; Elon scores one Mandate only when Industrial Velocity actually reduces a completed Facility’s cost; Jensen’s New Architecture self-Compute follows accepted licenses with no automatic base; and Customers four and five score one Mandate each while retaining full income. Every other numerical rule is frozen pending human evidence. That Grid-Ready rule was removed in 0.7.0-rc.6-test. Realignment is retained only in Advanced Play pending comparative human evidence.

0.5.0-rc.21-test changes no playable rule from rc.20. Executable 0.8.20 adds complete realized-value telemetry for Emergency Pause, Audited Deployment, and Responsible Scaling. The corrected fresh-seed faction swaps reject a universal Dario nerf or Sam buff because the Dario effect reverses by backend and negotiation-aware Sam is approximately neutral against Mark.

0.5.0-rc.22-test changes no playable rule from rc.21. Executable 0.8.21 adds per-opponent persona/backend selection to interactive play and an origin-restricted, token-paired localhost bridge for the deployed browser. Every LLM opponent receives an independent bounded decision budget and deterministic fallback. This is runtime access and policy configuration, not a board-game mechanics or balance change.

0.5.0-rc.23-test changes no playable rule from rc.22. Executable 0.8.22 corrects the deployed Simulation Lab’s player-count DOM binding discovered by the live HTTPS-to-localhost browser check.

0.5.0-rc.24-test changes no playable rule from rc.23. Executable 0.8.23 makes weighted and greedy interactive opponents execute directly in the browser. The token-paired localhost bridge remains optional and is selected only for Claude, Codex, hybrid opponents, or server-backed Simulation Lab jobs. The browser imports the same match, decision, persona, and deterministic policy modules as the Node executable; this is a deployment/runtime correction, not a board-game mechanics or balance change.

The rc.17 physical candidate has no mechanics delta from rc.16. It exists because the synchronized evidence documents and executable harness changed; the immutable rc.16 bundle and every report attributed to executable 0.7.1 remain historical evidence.

The rc.18 candidate likewise changes no playable physical rule from rc.17. It synchronizes the fresh-seed-validated Trust Governor and Power Broker profiles with executable 0.7.4, giving subsequent faction probes a clean, reconstructable evidence identity.

The rc.19 candidate changes no playable physical rule from rc.18. Executable 0.7.5 adds the staged AGI funnel, realized faction-ability values, player-count-specific outcome summaries, and preregistered common-seed faction swaps. These are evidence instruments, not rule changes.

The 0.5.0-rc.24-test candidate and executable 0.8.23 retain the three- to five-player product boundary, make four players the balance authority, and require three/five regression coverage. Their playable mechanics are identical to rc.20 and executable 0.8.19.

Executable 0.8.22 also preserves identity classification: dist/runtime/simulation-copy.json is evidence-boundary copy, not a playable rules input. It now contributes to the playtest-kit fingerprint instead of the ruleset and mechanics fingerprints. The new fingerprint is therefore a deliberate identity rebaseline, not evidence of a gameplay delta.

Unified evidence-frame decision

Automated evidence is one sampled seven-factor frame rather than separately labeled “diverse” and “cooperative” batches. Cooperation is classified after a match from promises, betrayals, trades, causal Power support, and supplier placement. Sparse cells are shrunk before interpretation, while a multiplicity-safe bounded confidence sequence supplies the conservative threshold gate.

Adaptive allocation may choose where to sample next, but it cannot change the registered outcome, threshold, or interpretation boundary. Best-response mutation is a diagnostic slice, never a balance authority. Claude or Codex appears only in a committed, explicitly authorized, decision-capped holdout; fresh calls test robustness and read-only cached calls test reproducibility. Strict provider schemas require every declared output field while nullable commentary fields remain semantically optional. Failed calls retain attempted provider provenance through deterministic fallback. Fresh capture uses write-only cache mode so it cannot silently reuse an old answer; the paired replay uses read-only mode and fails on any missing decision.

Open evidence questions

Measure before changing numbers:

Four players is the authoritative balance target. Three- and five-player quality, deferred modules, and numerical balance remain open until controlled evidence exists. Three and five are the suggested full formats and mandatory regression guards for any selected four-player change. Two and six players are playable exploratory configurations, not current balance-authority formats.

The deterministic lean balance and cooperative-AGI study observed rare rules-legal cooperative declarations and narrowed two large faction outliers. Its schema-v3 supplier attribution was not causal and is withdrawn; only clean schema-v4 runs may support supplier-viability claims. Simulation still cannot settle human negotiation, teachability, or duration; those remain physical-test questions.

Blind-play closure decisions

The current rules candidate closes the following previously implicit contracts: