AHR v2 spatial world and planning repair research checkpoints

Separate 329M full-core research repair with learned spatial readout, same-coordinate residual world, both EBMs, literal Lorentz geometry and bounded discrete-action planning. Final/best weights and measured benchmark/restore receipts. Offline human-replay evaluation; no ARC gameplay score claimed. Verify full SHA256 and byte count. Raw recordings and source notebooks excluded.

AHR v2 preliminary human-playthrough step200 parent checkpoint

Preserved custom329M notebook26 terminal step200 current-frame human-action/click imitation checkpoint and exact verification metadata. Preliminary research state; no baseline win or ARC score claimed. No architecture source notebooks, raw human recordings or credentials are published. Verify asset IDs, sizes and SHA256 because release storage is mutable.

AHR v2 completed human-playthrough research checkpoints

Completed one-hour-bounded continuation: 10,766 additional optimizer updates, global step 10,966, 50m53s on a shared A100. Full 329,334,183-parameter model retained; the current-frame imitation path trains 211,749,274 parameters. All nine exact recovery checks passed before launch.

On 4,096 fixed development rows from four reused held-out families and all 44 eligible recordings: action agreement 36.28%→37.87%, versus 47.68% mask-conditioned baseline; click error 21.95→25.36px; joint NLL 4.5785→5.7308, versus 4.4622 for a global-action-prior/uniform-coordinate control. Final exact clicks 0/1,604. Pixel-shuffle controls do not establish useful state-specific dependence. No baseline win, ARC score or game completion is claimed. The best checkpoint by development joint NLL remains initial step 200.

Both checkpoint payloads (2,650,732,866 bytes), 12 selected metadata files and the public restore receipt passed full anonymous download and SHA256 verification on 19 October 2026. Final/best payloads are also retained on the training server. See the learning_report JSON for paired recording-bootstrap intervals and scope limitations.

These are custom PyTorch recovery states, requiring the matching architecture notebooks/runtime/data identities. World prediction, EBM refinement, Lorentz geometry and planning are inactive in this imitation experiment. No raw human recordings or executable source notebooks are included. GitHub assets are mutable: verify the recorded asset IDs, byte counts and SHA256 hashes in archive_receipts__PUBLISHED_TO_GITHUB.json before loading.

AHR v2 with EBM, Lorentz geometry and planning: final/best; no baseline win

Full AHR pilot: no baseline win (19 October 2026)

The new full-path run retained 329,334,183 parameters and activated differentiable Worker/Scout EBM refinement, literal 1+3 Minkowski/Lorentz event calculations, observed two-frame memory and eight-root/two-step learned-world planning. All 17 exact-recovery and activation checks passed before training. This is a new initialization: the world and geometry modules were freshly initialized, all LayerNorm affine parameters reset, and the verified parent encoder/Queen/Worker/Scout tensors outside LayerNorm preserved. Exact continuation applies within this new run.

It completed 5,500 updates / 22,000 examples in 47 minutes 1 second, including smoke checks and evaluation, within the original one-hour allowance on shared GPU0. This is less than one pass through the 74,122-example pool. The frozen development cohort has 512 observations, 384 multi-legal rows and 196 click rows, balanced across four reused families and covering all 44 eligible recordings. It differs from the historical 4,096-row cohort below.

Fixed development metric	New-run initial	Final step 5,500	Baseline
Human action agreement	38.67%	41.60%	43.75%
Agreement with multiple legal actions	18.23%	22.14%	25.00%
Proper joint action/click NLL, lower is better	12.9404	5.2245	4.4300
Mean click distance, lower is better	28.42 px	31.15 px	23.19 px
Fixed-descriptor world MSE, lower is better	1.8328	0.4616	0.0221

The action baseline is a training-fitted legal-mask prior; joint NLL uses that prior with uniform coordinates; click distance uses a training-fitted constant median; world MSE uses the fixed current-frame descriptor as its successor prediction. Final exact clicks were 0/196. The final action difference from the mask baseline is -2.15 percentage points (exploratory 95% interval -5.46 to +0.98 pp), so superiority is not established. The best checkpoint by development joint NLL is step 2,500, also below the baseline criterion.

Real observations outperform shuffled observation bundles on multi-legal action agreement by 3.13 pp (95% interval +0.52 to +5.80 pp), but not on joint NLL. This is limited evidence of useful observation dependence. Disabling planning slightly improves joint NLL by 0.0069 nats (paired interval 0.0044 to 0.0093); the Lorentz-versus-Euclidean difference includes zero. These are same-checkpoint inference controls, not separately trained architecture ablations. Intervals resample recordings within fixed reused families, with one training seed and no multiple-comparison correction.

World prediction and encoder anchoring improve from their new-run initial states, but final world MSE remains 20.9 times the copy-current error (19.7 times on descriptor-changed rows). The coarse fixed auxiliary target is 16-colour occupancy in 8x8 spatial cells; it is not a direct policy input or full visual reconstruction. The planner uses approximate future transitions and unverified future masks. Computation stays at fixed depth two; persistent event DAG/pheromone memory and useful adaptive halting are not demonstrated. No ARC gameplay, completion rate or competition score was measured.

Public final/best storage verified: the full-path release contains final step 5,500, best step 2,500, 11 selected metadata files and a public restore receipt. Both payloads total 2,652,910,016 bytes. Every complete anonymous payload download and the restore receipt passed byte-count/SHA256 checks. The weights also remain on the training server; the temporary upload credential was removed.

Inactive AHR diagnostic checkpoint archive

Public archive of the author's inactive AHR v2 research checkpoints.

This release preserves a failed grounded-policy repair (steps 1, 401 and 600) and an early synthetic-puzzle diagnostic snapshot (step 6097). These are recovery and audit artifacts, not evidence of strong ARC gameplay.

Assets are uploaded with a restore receipt mapping original relative paths to asset IDs, byte counts and SHA256 values. Downloads must be verified against the receipt before restoring the matching notebook-bound experiment. Raw human playthrough data and source notebooks are excluded. No license grant for upstream datasets is implied.

AHR v2 delta-consistent full-architecture research checkpoints

Action balancing and corrected world gradients: no baseline win (19 October 2026)

The complete 329,350,583-parameter AHR v2 remains intact: Queen transformer, Worker and Scout EBMs, literal Lorentz geometry, observed memory, learned spatial world model and bounded two-step planning. Informative actions are sampled uniformly, with one forced action per eight examples. Conditional coordinate supervision is separated from branch action weights; world heads use different successor bootstraps and changed-cell supervision.

A first action-balanced trial completed 3,446 updates / 55,136 sampled examples. Its calibrated policy recovered 34/68 optional clicks, but its raw policy recovered 0/68. It also exposed a concrete implementation error: subtracting a detached current state from a forecast containing that same identity base created an unwanted encoder gradient. The corrected loss differentiates both terms. Identity-limit and nonzero-residual finite-difference tests verify the correction; initial forward predictions remain bitwise identical.

The corrected trial completed 2,516 updates / 40,256 sampled examples in 26.64 minutes, including 1,812 joint updates. All three actual stage recovery gates passed. It used the same parent, seed, data, sampler and learning rates, with a separately registered shorter allowance. This makes terminal cross-trial comparisons descriptive; matched TRAIN updates are an additional mechanism diagnostic.

Same 512 development observations	Action agreement	Joint NLL, lower is better	Mean click distance
TRAIN-fitted baseline	43.75%	4.4300	23.19px
Parent 4,385	35.16%	4.9727	23.67px
First action-balanced 3,446	40.43%	4.9127	22.93px
Corrected terminal 2,516	35.74%	4.9493	23.26px
Preregistered TRAIN-selected 704	36.52%	4.9797	23.61px

Baseline cells use metric-specific references: mask-table action probabilities with uniform legal coordinate density for joint NLL, and the TRAIN-fitted coordinate median for click distance. They are not one predictor yielding every cell.

Neither corrected endpoint beats the unchanged baseline criterion. Optional clicks remain 0/68, raw and calibrated. The terminal raw joint NLL worsens from 5.0000 to 5.1244; TRAIN-fitted calibration lowers it to 4.9493. Loss reduction after calibration does not demonstrate that the raw policy learned useful click decisions.

Grounding is more stable: corrected world forecast MSE is 0.08603, versus 0.21326 in the first action-balanced trial and 0.08911 at initialization. Encoder alignment is 0.06875, versus 0.19342 and 0.07334 respectively. At matched TRAIN update 2,204, world MSE is 0.08472 versus 0.12296. However, the corrected terminal forecast still loses to neural-current persistence 0.08382 and fixed-descriptor persistence 0.02211. Useful action-dependent dynamics beyond persistence are not established. Distinct world heads alone do not prove calibrated epistemic uncertainty.

Corrected final/best weights and benchmark/restore metadata · First diagnostic trial.

Checkpoint preservation verified: both corrected final/best payloads total 2,653,937,277 bytes. All 22 payload/metadata assets and the separate restore receipt passed complete anonymous byte-count/SHA256 verification at 2026-10-19T22:33:44.406082+00:00. Latest two resumable checkpoints plus best remain on the training server. The temporary upload credential has been removed. GitHub releases remain mutable; verify full hashes on restore.

These are offline human-playthrough imitation and one-step descriptor diagnostics, using the same 512 observations from 44 recordings in four reused development families. This is not an ARC-AGI-3 gameplay score, game-completion result or new-game generalization test. The TRAIN-family selector has only four optional-click labels; ancestors were exposed to its families. One seed, reused development data and recording-bootstrap intervals conditional on these families limit conclusions. Inference controls establish component activity, not a causal architecture advantage.

AHR v2 action-balanced full-architecture research checkpoints

Separate 329M full-architecture action-balanced research repair with detached conditional click training, informative action-stratified sampling, changed-transition spatial dynamics, three independently bootstrapped successor heads, both EBMs, literal Lorentz geometry and bounded discrete-action planning. Final/best weights and measured benchmark/restore receipts. Offline human-replay evaluation; no ARC gameplay score claimed. Verify full SHA256 and byte count. Raw recordings and source notebooks excluded.

Diagnostic qualification: world delta gradient

This action-balanced trial is a preserved negative diagnostic: final action agreement is 40.43%, versus a 43.75% baseline; joint NLL is 4.9127, versus 4.4300 (lower is better). TRAIN action calibration recovers 34/68 optional click choices while the raw policy remains at 0/68. World descriptor MSE regresses to 0.21326, versus 0.02211 for persistence. An audit found an asymmetric detach in the changed-state loss: the forecast contains the current-state identity base, but its subtraction detaches that state, leaving an unintended encoder gradient. A separate delta-gradient trial corrects this; its results must be assessed separately. These are exploratory measurements on reused human observations, not ARC gameplay scores.