Sep 23 17:09 Refresh the publish history 17:08 Publish 1895 files, replacing 825 cells Sep 8 21:10 Reindex the mmc_pps_ee cells for meco_gr, meco_ko and meco_ru 17:33 Refresh predictor card for M20 Monte Carlo cells and current index Correct MC examples and cap, regenerate inventory tables from tags.parquet, and update renamed loading examples. Predictor data is unchanged. 14:19 Relative shard names in the pool manifests 14:13 Refresh the publish history 14:13 Environment, write time, model revision and device from records 13:51 Full commit hashes in the index 13:44 Refresh the publish history 13:43 Reconstructed compute commits for the 2026-08 MC cells 10:32 Card: environment fields 10:11 Refresh the publish history 10:10 Provenance columns on the index, publish history, card for the new home Sep 7 21:37 Word entropy on the deferred wave's corpora 19:10 Word entropy on the corpora the G-only and deferred waves covered 13:02 Retire the unseeded continuation entropy cells 13:02 Retire the unseeded continuation entropy cells 13:02 Retire the unseeded continuation entropy cells 13:02 Retire the unseeded continuation entropy cells 13:02 Retire the unseeded continuation entropy cells 13:02 Retire the unseeded continuation entropy cells 13:02 Retire the unseeded continuation entropy cells 12:52 Forward word look-ahead entropy under per-stimulus seeding Sep 6 19:17 WikiText-103 unigram surprisal (Pimentel et al. 2023 table) on nine further English corpora, for the Meister et al. 2024 replication 16:30 Retire word entropy at the 30 sub-token cap 01:09 Restore the index rows for cells this publish did not regenerate 01:00 Monte Carlo entropy: per-stimulus seeding, and word entropy at a 20 sub-token cap Sep 2 13:07 Add the expectation-equivalent priority MMC unigram surprisal cells (mmc_pps_ee) and retire the seed-100000 canonical-only generation Aug 30 21:34 Normalize unigram surprisal cells by the pool word count 21:29 Remove superseded k4096 unigram surprisal cells 18:15 Add text_unit and value_unit columns 01:23 Add files using upload-large-folder tool 01:22 Add files using upload-large-folder tool 01:22 Add files using upload-large-folder tool 01:21 Add files using upload-large-folder tool 01:21 Add files using upload-large-folder tool 01:21 Add files using upload-large-folder tool 01:21 Add files using upload-large-folder tool 00:43 Add text_unit and value_unit columns Aug 26 21:07 transplant campaign: per-character charbeam surprisal on nine new eye-tracking corpora, reduced-ladder logit lens (11 models x 12 corpora) Aug 25 15:31 lens campaign complete: pythia-12b logit lens over the four datasets 13:41 word-infoval: celer (full 200-shard campaign), dundee + natural_stories (trained-window cap) 11:37 Add text_unit and value_unit columns 04:12 lens campaign (68 cells dense in lens_layers/), simadj alpha arrays (simadj_esim/), per-character charbeam cells, bsc word-infoval 01:35 mgpt long-context batch (26 MECO cells) + wt103 unigram cells (4 Pimentel corpora) 01:04 Add text_unit and value_unit columns Aug 24 19:23 A6: recompute the responsivity family as one draw-batch (58 cells, gpt2 + mGPT x 29 datasets, uniform Mila L40S hardware) and the 23 placeholder-era MC cells on meco_ee/fi/gr (N512 sampled with chunk-size 25) 15:19 Add text_unit and value_unit columns 02:01 Vintage fixes: recompute the 21 placeholder-era closed-form cells on meco_ee/fi/gr under the current loader context and refresh the geco wordfreq unigram cell (43 stale multi-word-AOI rows) Aug 23 17:00 Add the 26 sentence-scope cells (mGPT surprisal + next_word_entropy, 13 MECO wave-1 languages, _sent tags; anchored against the authors' mgpt_sc columns) 
