Spaces:
Sleeping
Sleeping
Commit History
-
docs: Amendment 02 framing + Amendment 03 training doc
thewildofficial- Reframe README and specs as centralized planning (not multi-agent) - Document training pipeline alignment for Option A (see PR description) Made-with: Cursor05a5901 -
feat(game): Option A auto-critical, discretionary proposals, remove CRITICAL_FAILURE
thewildofficial- Auto-fund critical in Phase 5; bankruptcy if treasury < sum(Critical) - Remove critical invariant termination and CRITICAL_FAILURE reward penalty - Parliament: drop ProposalAbstention adapter types - Server: remove ABSTAIN_FROM_PROPOSAL from action mapping - Direct allocation: return observation snapshot before next round reset - Sync specs 02,03,04,05,07,09,10 with Option A - Update tests and reward fixtures; add PR_DESCRIPTION_OPTION_A.md Made-with: Cursore4666d9 -
(train) pre-test train
Kr0issantf72fde9 -
Add proposed amendment for training spec alignment
thewildofficialDocuments how the GRPO training pipeline (reward_fn, prompts, dataset generation, benchmarks) must change to work with each game-mechanic option from 01-critical-threshold-abstention-voting-fix.md. Includes detailed compatibility matrix, change order, and implementation estimates. Defers actual training work to post-amendment PRs.5b5a0b2 -
Add proposed amendments for critical threshold and README veracity
thewildofficial- 01-critical-threshold-abstention-voting-fix.md: Options A-F for resolving abstention/rejection/critical contradictions in game mechanics. Recommends Option A (auto-fund critical + discretionary proposals). - 02-readme-veracity-and-theme-alignment.md: Reframe project as centralized planning with parliamentary narrative wrapper. Aligns with hackathon themes #2 (long-horizon planning) and #3.1 (world modeling).dd79c42
-
Merge main and fix benchmark adapter review issues
thewildofficialResolve the main-branch merge conflicts, keep the benchmark plotter artifacts, and make rule-based baselines cast only valid non-duplicate votes so CI and benchmark evidence reflect legal policy behavior. Made-with: Cursor046ac0f -
add spec 13
Kr0issantce5b90a -
Implement baseline benchmark evidence pipeline
thewildofficialAdd seeded in-process policy benchmarks, random baseline coverage, metrics aggregation, and committed result plots so future training work has a reproducible baseline comparison. Made-with: Cursor5f8887d -
chore: scaffold RL adapter pipeline
thewildofficialAdd the adapter, telemetry, evaluation, and training structure so agent work can proceed independently from core game-loop implementation. Made-with: Cursoracb2712 -
fill sectors and events json
Kr0issante6288c6 -
docs(spec): reconcile specifications with original piecewise revenue design
thewildofficialMajor reconciliation updating all core documents to match original design intent: Economy Model (04): - Replace linear efficiency with piecewise revenue curve (Critical→Demand→Surplus→Wastage) - Add per-sector critical threshold (40% of demand = game over) - Revenue depends on Allocation (not Consumption) - Add persistent productivity [0.5, 2.0] with running state - Population-scaled demand: Demand = Baseline × (Pop/Pop₀) × Event_Multiplier - Population dynamics: 1,000,000 initial, birth/death rates - Same-round revenue application Event System (06): - Switch from additive costs to multiplicative demand scaling - Positive events inject cash into treasury (not reduce need) - Add missing events: Unemployment, Teacher Shortage, Market Crash, Contamination, Public Distrust, High Public Enthusiasm - Black swan events: Natural Disaster, Pandemic, Financial Collapse - Realistic distribution: 40% normal, 35% minor, 20% moderate, 4% critical, 1% compound Turn Structure (03): - Same-round r…61c8933 -
docs(spec): add comprehensive RL parliament environment specification
thewildofficialComplete specification document set for multi-agent RL parliament hackathon: - 01_GAME_OVERVIEW: Marxist-inspired cooperative resource allocation concept - 02_GAME_RULES_REFERENCE: Single-page anchor with MUST/MUST NOT rules - 03_TURN_STRUCTURE: 9-phase turn flow with rotating proposal order - 04_ECONOMY_MODEL: Model A government budget execution with realistic macro - 05_VOTING_PROTOCOL: Sequential voting with no self-voting, majority rules - 06_EVENT_SYSTEM: 40% normal, black swan crises, positive events - 07_AGENT_ACTION_SPACE: Valid actions with phase restrictions - 08_OBSERVATION_SPACE: Public/private information distinction - 09_REWARD_MODEL: Prosperity/GDP with collective reward - 10_SUCCESS_CRITERIA: Winning/losing with population growth metrics - 11_GUARDRAILS: Explicit exclusions and anti-patterns - 12_GLOSSARY: 26 defined terms - APPENDIX_A_EXAMPLES: 5 concrete examples with correct math Key design decisions: - Model A: Consumption = min(Need, Allocation), treasury pays onl…449767b