Commit Graph

Select branches

Hide Pull Requests

autonomous/trap-exploration

experimental/inference-server-prototype

main

Mono Color

• 7fc15ff46a Move the round strip out from under the opponent hand; let undo cross turns main coolguyandClaude Opus 4.8 2026-07-15 18:18:47 +09:00
• 774faf2b51 Add the altair-vs-borealis measurement driver coolguyandClaude Opus 4.8 2026-07-15 18:11:27 +09:00
• c8fa89dab9 Pause between rounds, summarise the match, and localise to Korean coolguyandClaude Opus 4.8 2026-07-15 18:11:19 +09:00
• 267db367d8 Remove hints; gate undo/redo behind a setting and limit it to your turn coolguyandClaude Opus 4.8 2026-07-15 17:56:16 +09:00
• 5e177cc369 Stop drag-selecting the game board coolguyandClaude Opus 4.8 2026-07-15 17:36:41 +09:00
• 3e7dda0d22 Make overlapped hand cards legible with a corner index coolguyandClaude Opus 4.8 2026-07-15 17:23:49 +09:00
• 138d7c858d Classic Lost Cities: three-round agent, and borealis in the browser coolguyandClaude Opus 4.8 2026-07-15 07:10:07 +09:00
• 1d3b29aadb Ship borealis to the browser, and add the classic three-round mode coolguyandClaude Opus 4.8 2026-07-15 07:06:29 +09:00
• 8dec6c3fe1 Keep the human game records, and name the models coolguyandClaude Opus 4.8 2026-07-15 06:25:29 +09:00
• 0fb6e3c1a6 Fund the exploiter properly: ours holds, league breaks coolguyandClaude Opus 4.8 2026-07-15 06:21:46 +09:00
• 0c83824243 Correct the exploiter budget: it was a quarter of the target, not half coolguyandClaude Opus 4.8 2026-07-15 05:50:58 +09:00
• e51d14d0fb Measure exploitability: ours is harder to farm than league coolguyandClaude Opus 4.8 2026-07-15 05:37:55 +09:00
• cc12b058e5 Re-measure both-seat vs privileged critic at the same budget coolguyandClaude Opus 4.8 2026-07-15 05:26:38 +09:00
• 4d1c4473b6 Restore the privileged critic: its ablation flips with scale coolguyandClaude Opus 4.8 2026-07-15 04:01:40 +09:00
• 1f51fb8149 Ablate the match stack: both-seat training carries it, the privileged critic hurts coolguyandClaude Opus 4.8 2026-07-15 03:49:24 +09:00
• 4392ec255a Record the old-vs-new head to head in the real three-round game coolguyandClaude Opus 4.8 2026-07-15 03:30:23 +09:00
• 8860f62030 Pay the match total densely and drop the tanh terminal reward coolguyandClaude Opus 4.8 2026-07-15 03:07:18 +09:00
• f170fcdcfd Run the classic self-play trainer and record what the carry probe says coolguyandClaude Opus 4.8 2026-07-15 02:44:04 +09:00
• 9ba5a07a9d Train on matches: match observation, asymmetric critic, both seats, match reward coolguyandClaude Opus 4.8 2026-07-15 02:23:50 +09:00
• 62b755c3f0 Retract the 120-ply round cap and record the missing wiring step coolguyandClaude Opus 4.8 2026-07-15 02:09:49 +09:00
• 7cd299dd1e Add the classic three-round match layer coolguyandClaude Opus 4.8 2026-07-15 02:05:50 +09:00
• 47d0e803f5 Record the Phase 0a gate result and the MAX_STEPS artifact it exposed coolguyandClaude Opus 4.8 2026-07-15 01:57:03 +09:00
• d1c2cf628e Reset envs inside the rollout scan and fix the metrics it breaks coolguyandClaude Opus 4.8 2026-07-15 01:27:02 +09:00
• 8b7348cbce Add hints, per-action undo/redo, and card motion coolguyandClaude Opus 4.8 2026-07-14 21:33:53 +09:00
• be5226bd3b Add seeded deals and card movement animations coolguyandClaude Opus 4.8 2026-07-14 21:10:14 +09:00
• 299c788545 Fix web client interaction and layout bugs coolguyandClaude Opus 4.8 2026-07-14 21:04:27 +09:00
• e47db5eb1e Show placed cards before drawing coolguy 2026-07-14 20:37:59 +09:00
• 891cc113f0 Align web client with pygame table coolguy 2026-07-14 20:28:47 +09:00
• ef4b9d82b0 Focus project on JAX PPO coolguy 2026-07-14 20:09:03 +09:00
• 79273f7eb3 Add on-device web client coolguy 2026-07-14 19:49:03 +09:00
• 9517fd3ba8 Document JAX PPO model capacity experiment coolguy 2026-07-12 15:21:09 +09:00
• f854d97764 Add match replay and JSONL export coolguy 2026-07-12 15:14:24 +09:00
• 7e481ca064 Add polished Lost Cities table GUI coolguy 2026-07-12 14:58:59 +09:00
• 3b86f81bf6 Add JAX PPO opponent to classic GUI coolguy 2026-07-12 04:14:02 +09:00
• 19560029b5 Add final cycle report and human play CLI coolguy 2026-07-06 00:14:24 +09:00
• 16cc31676c Add diminishing returns diagnostic report coolguy 2026-07-05 18:32:47 +09:00
• fa9a1c5286 Add gates 1-2 audit and repair workflow coolguy 2026-07-05 17:55:22 +09:00
• 94e9ac1854 Add JAX PPO league self-play v1 coolguy 2026-07-05 06:33:22 +09:00
• 6037b650f3 Add JAX PPO ladder v2 expert pass coolguy 2026-07-05 02:06:22 +09:00
• 7fe0bfdcfe Add JAX PPO ladder verification pass coolguy 2026-07-05 00:17:49 +09:00
• bb52ef9ec1 Document JAX PPO ladder results coolguy 2026-07-04 23:19:41 +09:00
• 4c0c2e9add Reset finished JAX PPO environments coolguy 2026-07-04 22:33:51 +09:00
• 9e27f42f27 Add JAX PPO static-opponent trainer coolguy 2026-07-04 22:30:16 +09:00
• 768f74693d Plan JAX PPO static-opponent ladder coolguy 2026-07-04 22:14:37 +09:00
• b37b841eef Record CUDA JAX throughput coolguy 2026-07-04 21:56:24 +09:00
• 72893250ec Record full JAX differential verification coolguy 2026-07-04 20:39:01 +09:00
• 30ccc3cf41 Batch JAX differential verification coolguy 2026-07-04 19:42:05 +09:00
• f872204b13 Document JAX engine and benchmark coolguy 2026-07-04 19:38:09 +09:00
• ac54f98189 Add JAX engine verification tests coolguy 2026-07-04 19:38:04 +09:00
• 1d7758b7d3 Add JAX Lost Cities rules engine coolguy 2026-07-04 19:37:27 +09:00
• 8f380928e6 docs(research): split SO-ISMCTS into its own group in catalog coolguy 2026-05-12 00:22:10 +09:00
• 2a8a082a12 Catalog research notes for new agents coolguy 2026-05-11 21:04:44 +09:00
• 44b8faba3d docs(research): SO-ISMCTS BC ceiling write-up from 2026-05-11 autonomous session autonomous/trap-exploration coolguy 2026-05-11 20:45:10 +09:00
• cba6caee2f Add mixed-opponent self-play with opponent-aware MCTS coolguy 2026-05-11 20:42:13 +09:00
• b9fc5693a4 Normalize PUCT Q + add mirror-descent policy target coolguy 2026-05-11 16:31:02 +09:00
• d850070ed4 Add KL anchor to BC reference policy in trainer coolguy 2026-05-11 15:24:14 +09:00
• 9fdfa88b23 Add lost-cities-ismcts pretrain: behavior-clone heuristic into network coolguy 2026-05-11 13:36:28 +09:00
• 33c44c708e Add --resume-from for warm-starting training from a checkpoint coolguy 2026-05-11 10:41:08 +09:00
• 0d35341bbe Cycle 5 setup: long run with use_rollout_value=false + 768x4 network coolguy 2026-05-11 07:56:35 +09:00
• a501a93223 Cycle 4: revert use_rollout_value=true, scale network 512x3 -> 768x4 coolguy 2026-05-11 07:03:45 +09:00
• 200129d16d Add diagnostic value-head metrics: rmse, target stats coolguy 2026-05-11 06:43:50 +09:00
• 8b7ed66ffd Cycle 3 prep: fix search() ignoring parallel_simulations + flip use_rollout_value=false coolguy 2026-05-11 06:42:20 +09:00
• be0c1a8d62 Add training.value_loss_weight (default 1.0) for value loss reweighting coolguy 2026-05-11 06:34:05 +09:00
• 169d4dcb14 Cycle 1: c_puct 3->5, dirichlet_eps 0.25->0.4 produced first natural-end wins coolguy 2026-05-11 05:53:54 +09:00
• f289997c1c Add Dirichlet root noise + standalone eval CLI, fix self-play stall trap coolguyandClaude Opus 4.7 2026-05-11 05:17:32 +09:00
• dafb7a6a1d Add central ISMCTS scheduler prototype experimental/inference-server-prototype coolguy 2026-05-11 03:24:54 +09:00
• 1ee329250e Add ISMCTS inference server path coolguy 2026-05-11 03:16:18 +09:00
• 651175e5bd Add multi-process self-play, eval workers, MCTS Cython port coolguyandClaude Opus 4.7 2026-05-11 02:39:39 +09:00
• 0999d34277 Wire W&B tracking into ISMCTS trainer coolguy 2026-05-10 23:13:03 +09:00
• 812dace1e3 Extend analyze.py for ISMCTS metrics coolguy 2026-05-10 23:08:51 +09:00
• 25a3fba53f Use Deep CFR diagnostics for IS-MCTS eval coolguy 2026-05-10 23:04:05 +09:00
• bec59dfc3c Record §12 R3 SO-ISMCTS mini Lost Cities PoC result coolguy 2026-05-10 22:54:03 +09:00
• e69f3165b6 Add SO-ISMCTS mini trainer coolguy 2026-05-10 22:46:22 +09:00
• 5acda3f272 Record R2 (heuristic_balanced opponent) early termination + IS-MCTS pivot coolguy 2026-05-10 22:18:17 +09:00
• 7d59398159 Add heuristic_balanced opponent_policy to interleaved scheduler coolguy 2026-05-10 21:08:12 +09:00
• 12d10fd9c8 Record full_depth past-self pool experiment from prior repo coolguy 2026-05-10 20:34:31 +09:00
• 52ef321274 Record anchor_safe015 self-play league mixing experiment from prior repo coolguy 2026-05-10 20:27:48 +09:00
• 79f3ca0a58 Record regret_matching_epsilon=1e-4 zero-pit history from prior repo coolguy 2026-05-10 20:25:38 +09:00
• 3b6da3bcbc Record R1 (discard_only opponent) result + R2 direction coolguy 2026-05-10 19:04:12 +09:00
• e3f2423f46 Add discard_only opponent_policy + analyze.py merges coolguy 2026-05-10 17:39:00 +09:00
• 4ac74e2501 Broaden gitignore for Cython outputs and local session files coolguyandClaude Opus 4.7 2026-05-10 15:37:03 +09:00
• 004b913a7b Rename bot family, curate analyze plots, tier evaluation cadence coolguyandClaude Opus 4.7 2026-05-10 15:32:55 +09:00
• 0457efdf29 Honour all_negative_fallback in interleaved scheduler; sync default.yaml coolguyandClaude Opus 4.7 2026-05-10 15:01:46 +09:00
• b6863b3ba0 Fix ColorSharedNetwork to use real per-color encoding layout coolguyandClaude Opus 4.7 2026-05-10 14:42:32 +09:00
• f63c4b8059 Strip heuristic input features and add selectivity diagnostics coolguyandClaude Opus 4.7 2026-05-10 14:26:16 +09:00
• 1593135313 Record first-open replay result coolguy 2026-05-10 02:44:10 +09:00
• 8217cecd19 Speed up first-open memory sampling coolguy 2026-05-09 23:57:51 +09:00
• c570fc3ea5 Add first-open replay reweighting coolguy 2026-05-09 23:46:55 +09:00
• 7c3a3499dc Add first-open counterfactual audit coolguy 2026-05-09 17:06:06 +09:00
• ba54032703 Add first-open target audit coolguy 2026-05-09 16:36:43 +09:00
• c246260a8b Document Deep CFR selectivity findings coolguy 2026-05-09 12:38:05 +09:00
• b33a55c76d Align interleaved outcome targets and add open diagnostics coolguy 2026-05-08 17:25:27 +09:00
• b5550d3840 Add current-vs-average eval diagnostic coolguy 2026-05-08 17:14:21 +09:00
• 21ad8a4d28 Document Deep CFR baseline analysis coolguy 2026-05-08 17:01:05 +09:00
• aa898edfc2 Increase default Deep CFR eval cadence coolguy 2026-05-08 16:33:43 +09:00
• dd88564b8d Add W&B grouping options coolguy 2026-05-08 16:22:52 +09:00
• fe01704b60 Add deterministic Deep CFR traversal mode coolguy 2026-05-08 16:09:03 +09:00
• 17d149c47c Document Deep CFR reproducibility policy coolguy 2026-05-08 15:53:38 +09:00
• 0f85fa85b3 Close librarian: full archive promote-survey + parallel dispatch coolguyandClaude Opus 4.7 2026-05-08 02:33:51 +09:00
• 5c221fb3c6 Accept first survey batch + add commit-hash post-processing coolguyandClaude Opus 4.7 2026-05-08 01:52:08 +09:00