Replace PR #998 with OPC native causal packed-memory non-record packet

[asuramaya]

Title: Replace PR #998 with OPC native causal packed-memory non-record packet

This replaces the earlier Conker-5 non-record claim in PR #998 with a fresh opc-native causal packed-memory packet packaged as a self-contained track_non_record_16mb record folder.

New record folder:
records/track_non_record_16mb/2026-03-30_OPC_CausalPackedMemory_NativeFullSpecClean

What changed:

• the measured run is now genuinely open_predictive_coder native
• no vendored Conker runtime is used for train/eval
• the packet includes its own native legality/replay adapter

Kernel basis:

• upstream open_predictive_coder: https://github.com/asuramaya/open-predictive-coder
• vendored open_predictive_coder snapshot ref: 4072074288fa279b655c11c30f8fca2e1859f925
• model family: packed unigram/bigram/trigram memory with fitted global mixture weights

Full spec run:

• train shard: official fineweb10B_sp1024 train shard
• validation shard: official fineweb10B_sp1024 validation shard
• run_id: opc_native_tokens100000000_62021846

Metrics:

• pre_quant_val_bpb: 2.48898442865273
• val_bpb: 2.48898442865273
• test_bits_per_token: 6.062992566022187
• bytes_model_int6_zlib: 2,705,939
• bytes_total: 6,146,877
• train_tokens: 100,000,000
• eval_tokens: 62,021,846
• train_time_sec: 52.381125042
• mixture_weights: [0.0, 0.9, 0.1]

Hardware:

• machine: MacBook Pro
• chip: Apple M3 Max
• memory_bytes: 38,654,705,664
• os: macOS 26.3 (25D125)

Submission surface:

• includes README.md
• includes submission.json
• includes results.json
• includes train.log
• includes train_gpt.py
• includes root artifact model_artifact.npz
• includes audit_tokens.npy
• includes opc_native_detect_adapter.py
• includes vendored open_predictive_coder
• includes vendored opc_parameter_golf_submission

Audit status:

• submission audit: pass
• legality trust achieved: strict
• legality satisfied: true
• replay repeatability: pass
• artifact is under the 16,000,000 byte cap
• total packet bytes are under the 16,000,000 byte cap

Scope note:

• this is a non-record / unlimited-compute submission
• this packet is an opc-native descendant, not a parity wrapper
• the record folder is self-contained apart from the standard external dataset root passed to train_gpt.py

[asuramaya]

Edited March 29, 2026:

Not closing this PR. I am updating this same pull in place.

The old Conker-5 claim is invalid, but the PR itself is now being reused as the correction surface for the fresh legal Conker-11 rerun and attached audit bundle.

Current tools and public packet:

• conker-detect: https://github.com/asuramaya/conker-detect
• conker-ledger: https://github.com/asuramaya/conker-ledger
• detector packet: https://github.com/asuramaya/conker-detect/tree/main/examples/parameter-golf-pr998-replacement-2026-03-29
• validity bundle: https://github.com/asuramaya/conker-ledger/tree/main/examples/parameter-golf-pr998-replacement-2026-03-29

[MatoTeziTanka]

[RETRACTED 2026-04-11] — This IMPORT_FAIL was a false positive. Root cause: sibling module exists in same records/ folder; runner sys.path bug. Your code is not broken. See correction below: #998 (comment)

---

Community Review — Replace PR #998 with OPC native causal packed-memory non-record packet

Compliance: NEEDS AUTHOR ACTION — train_gpt.py fails to import on CT2038 (Python 3.10 / torch 2.10.0+cpu)

What I found: The CPU smoke test on CT2038 (proteus-engine, 128 GB RAM, Triton 3.6.0, flash_attn stub, cutlass_evt_fusion stub) failed at the import step with:

ModuleNotFoundError: No module named 'opc_parameter_golf_submission'

A few of the common patterns I've seen for this class of error in the 2026-04-11 sweep:

• PEP 701 f-string nesting — e.g. log(f" {cat}: {", ".join(...)}") is valid Python 3.12+ but invalid Python 3.10 because the inner ", " re-enters the outer double-quote context. One-character fix: ', ' instead of ", ". See PR Record: SP8192 + Improved Parallel Residuals + Muon 0.97 + LR 0.03 + Legal TTT — val_bpb 1.07785 (3-seed mean) #1541 / Record: SP8192 + Triple Recurrence + Banking + Fused MLP + Muon 0.97 — val_bpb 1.0778 (3-seed mean) #1523 for reference.
• Missing flash_attn variants — e.g. from flash_attn_interface import flash_attn_varlen_func when the wrapper script only stubs flash_attn_func. Not a PR defect on H100s, but the eval image / CPU preflight path needs a guarded import.
• Local compiled extension — e.g. import cutlass_evt_fusion from a records/*/cutlass_evt_fusion/ subfolder that isn't on the import path at smoke time. Usually an import-order issue inside the script.
• Actual syntax error — typo, missing bracket, etc.

Recommendation: Could you run python3 -c "import py_compile; py_compile.compile('train_gpt.py')" on your records-folder train_gpt.py under Python 3.10 specifically? The eval image is Python 3.10 per Issue #17 / the README, so any parse error on 3.10 blocks the submission at import time before any of the scored-eval logic runs.

Once the parse/import issue is fixed, I'll re-run the compliance audit through the normal pipeline. No other flags identified yet because the audit halts at the import step.

---

Reviewed by @MatoTeziTanka — The Agora. CPU smoke test (CT2038 proteus-engine, 2026-04-11): IMPORT_FAIL — ModuleNotFoundError: No module named 'opc_parameter_golf_submission'. Classification via classify_prs.py AST-based classifier; full compliance audit deferred until the import issue is resolved. Auto-drafted from a template and spot-checked before posting.

[MatoTeziTanka]

Retraction — this IMPORT_FAIL was a bug in my smoke runner

Sorry @asuramaya, this one's on me. I re-audited the IMPORT_FAIL I posted above and it was a false positive — the fault is in how my CPU smoke runner set up sys.path, not in your code.

What happened:

The runner imported your records/track_non_record_16mb/2026-03-30_OPC_CausalPackedMemory_NativeFullSpecClean/vendor/opc_parameter_golf_submission/train_gpt.py with only the script's folder implicitly on sys.path, so when your file did from opc_parameter_golf_submission import ... it couldn't resolve the sibling opc_parameter_golf_submission.py that lives in the same opc_parameter_golf_submission/ directory. The error I reported — ModuleNotFoundError: No module named 'opc_parameter_golf_submission' — looked like a missing file, but I re-checked the head SHA e00a207 and records/track_non_record_16mb/2026-03-30_OPC_CausalPackedMemory_NativeFullSpecClean/vendor/opc_parameter_golf_submission/__init__.py is right there, committed to the PR, next to train_gpt.py.

Verified at head e00a207:

records/track_non_record_16mb/2026-03-30_OPC_CausalPackedMemory_NativeFullSpecClean/vendor/opc_parameter_golf_submission/__init__.py   ← sibling module, exists
records/track_non_record_16mb/2026-03-30_OPC_CausalPackedMemory_NativeFullSpecClean/vendor/opc_parameter_golf_submission/train_gpt.py   ← imports it

On the real eval image (Python 3.10, records/*/ as the working dir), this import resolves correctly because the records folder ends up on sys.path via the standard cwd-driven import or via the eval harness's per-record entry point.

Your PR is not broken by this error. I'm retracting the IMPORT_FAIL classification. I'll re-queue the full compliance audit (BPB check, n-gram / TTT / SLOT flags, etc.) on the current head and post findings separately.

Again — sorry for the noise. These community reviews only work if I actually read what I'm reviewing, and I didn't in this case.

[asuramaya]

@MatoTeziTanka not a problem, thank you for taking the time to review. the next model will be cleaner, this was more of a proof of concept.