MM-01Quality-constrained quantizationPreparation blocked and not admitted or tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Find smallest same-model Q4/Q5/Q6/Q8/high-precision format within locked-task noninferiority margin; report quality, RAM, latency, CPU cost.
Planning detail, not a new assignment
close the stopped preparation honestly; inventory available references/licenses without loading models; revise the proposed comparison to fit actual formats and envelope.
Earlier completion criteria
identical model/settings, independent paired evaluator, multiplicity-adjusted interval and margin frozen before run; disclose missing reference formats and budget mismatch. Any placement measurement may be an optional covariate only if independently warranted within an admitted scope; it is not a new prerequisite.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-02Layer-sensitive mixed precisionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Sensitivity-guided per-layer/per-tensor precision (attention/FFN, first/last layers) versus uniform at equal bytes; select on development, confirm sealed.
Planning detail, not a new assignment
specify sensitivity score, deterministic packing and equal-byte rule; validate map on tiny fixture.
Earlier completion criteria
exact byte equality, reproducible layer map, sealed noninferiority and worst-task results.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-03Calibration under domain shiftDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Generic/task-matched/random/shifted calibration for importance-matrix quantization; assess domain shift and reserved tasks.
Planning detail, not a new assignment
define disjoint calibration/development/sealed domains, sampling and dedup audit before calibration.
Earlier completion criteria
provenance/overlap report, same model and budget, shifted-domain uncertainty, reserved tasks remain sealed.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-04Structured pruning and layer removalDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Compare block/channel/layer and unstructured removal; distinguish CPU pruning feasibility from execution.
Planning detail, not a new assignment
freeze mask schema and baseline; check mask and byte counts on tiny fixture before runtime comparisons.
Earlier completion criteria
qualified CPU timing includes conversion/index overhead; matched quality and noninferiority; zero count alone is not speed.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-05Sparse CPU kernel break-evenDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Test sparse layouts, activation sparsity, bit-slice reuse and zero-skipping versus dense after index/dequantization costs.
Planning detail, not a new assignment
lock numerical microbenchmark/oracle; model overhead before small shape grid.
Earlier completion criteria
oracle agreement, overhead-inclusive repeat timings and workload-specific break-even; no one-shape generalization.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-06Native ternary models on CPUDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Reproduce compatible released ternary/BitNet checkpoint inference with specialized lookup CPU kernels; compare usefulness per byte/CPU-second.
Planning detail, not a new assignment
verify primary release, license, tokenizer, format and kernel compatibility via metadata.
Earlier completion criteria
exact artifact identity, supported CPU path, correct output contract, task/resource-normalized result.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-07MoE versus dense CPU economicsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Matched-usefulness MoE/dense comparison including resident weights, active bytes, routing cost and throughput.
Planning detail, not a new assignment
specify utility matching and account total resident bytes including inactive experts/runtime buffers.
Earlier completion criteria
matched utility, full RSS/routed bytes, routing overhead and throughput; active params do not stand in for memory.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-08Expert hot-set caching and prefetchDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Replay traces for locality, ideal-cache bounds, LRU/selective residency/prefetch before runtime hooks.
Planning detail, not a new assignment
define trace provenance; build simulator and ideal-cache bound first.
Earlier completion criteria
misses, bytes, memory and delay across same traces; stop if upper bound misses useful threshold.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-09DRAM and Optane executionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Verify physical memory population and Optane mode, characterize working-set sizes/access patterns, and distinguish Memory Mode's DRAM-cache behavior from separately addressable App Direct placement; no BIOS changes. Current scope: Optane testing is deferred pending replacements. Do read-only prior-record and method review only; no fresh hardware measurement, admission request, probe, or hardware route is proposed.
Planning detail, not a new assignment
inventory existing safe records/method definitions and record exactly what is unknown; leave physical state unasserted.
Earlier completion criteria
source-bound method note clearly separates historical observation from current state; no test or hardware claim. Any future measurement plan requires Jim to revisit the scope and fresh host qualification before a separate admission.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-10Memory mapping and layer streamingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Cold/warm loads, page cache, SSD/SATA, out-of-core streaming/prefetch under caps; preserve models and recovery storage.
Planning detail, not a new assignment
define read-only inventory and synthetic file-layout fixture; cold-cache protocol must not disrupt services.
Earlier completion criteria
RSS/cache/bytes/startup/per-layer latency/I/O contention and correctness; no source or recovery-store writes.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-11KV-cache quantizationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
K/V precision and context length versus RAM/speed/retrieval/reasoning/code accuracy, against higher precision on identical inputs.
Planning detail, not a new assignment
verify CPU runtime format support; freeze lengths and long-context cases; test tensor round-trip.
Earlier completion criteria
same tokens/model, K/V separately recorded, verified CPU path and task/memory/latency strata.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-12Evidence-aware KV retentionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Sink/heavy-hitter/salience/recency/hierarchical/semantic retention at fixed memory versus full KV and truncation; CPU ports conditional.
Planning detail, not a new assignment
portability review, fixed-byte policy and synthetic attention/retrieval fixture.
Earlier completion criteria
equal memory ceiling, both baselines, retrieval-position strata, document CPU feasibility.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-13Persistent prefix and KV librariesDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Exact prefix reuse, COW branching, save/restore/rehydration; complete cache key and invalidation.
Planning detail, not a new assignment
schema includes model/tokenizer/runtime/settings/prefix hashes; adversarial mismatch fixture.
Earlier completion criteria
identical outputs on reuse; reject stale/incompatible caches; isolation and resource cost; no sealed answers cached.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-14CPU paged-KV allocationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
CPU page allocation/fragmentation under sessions; GPU PagedAttention is reference only.
Planning detail, not a new assignment
synthetic allocator fixture over small page/lifetime grid before inference.
Earlier completion criteria
fragmentation/tail latency/RAM waste/quality versus contiguous allocator; label simulated vs real runtime.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-15Context selection and summary policiesDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Raw, truncation, retrieval, structured/evidence summaries, ordering, lost-middle and TTL compaction at equal tokens on long tasks.
Planning detail, not a new assignment
freeze task provenance and token counter before summaries.
Earlier completion criteria
equal budget, later-task correctness, source traceability, position strata, omission/compression errors.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-16NUMA and CPU placementPreparation complete but not admitted or tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Socket/local/interleaved, first touch, replica, affinity, thread and hugepage comparison on verified topology while protecting production allocations.
Planning detail, not a new assignment
preserve seven-file preparation and 31/31 fixture; resolve active-time uncertainty and 14-call retry reserve capacity; obtain fresh topology/allocation.
Earlier completion criteria
independent fixture replay, current reservation, no production interference, retry plan fits batch.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-17Instruction sets and packed CPU kernelsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Supported AVX2/AVX-512/VNNI, packed kernels/codebooks, W8A8 vs weight-only; no assumed AMX/BF16.
Planning detail, not a new assignment
read-only CPU/runtime feature audit and independent numerical oracle; fixture only verified paths.
Earlier completion criteria
runtime feature check, oracle agreement, unsupported variants excluded, same-shape/settings timing.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-18Separate prefill and decode budgetsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Separate prompt-processing/generation bottlenecks and tune within fixed total CPU resources, including multi-prompt/end-to-end.
Planning detail, not a new assignment
fix total CPU budget and prompt lengths; define phase/queue/end-to-end accounting.
Earlier completion criteria
no borrowing beyond total, tails and quality by prompt/output length; phase speed not substituted for user-visible gain.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-19Deadline-aware batchingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Serial/micro/continuous batch on mixed deadlines; throughput, tails, peak KV and correct completion.
Planning detail, not a new assignment
deterministic arrival/deadline replay; simulate policies before concurrent run.
Earlier completion criteria
correct tasks by deadline, p50/p95/p99, peak RSS/KV, expired work and fairness; not throughput alone.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-20Activation workspace reuseDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Buffer reuse/allocation removal/recompute tradeoffs for activation memory on small fixtures.
Planning detail, not a new assignment
freeze tiny operator inputs, oracle tolerances, and allocation accounting.
Earlier completion criteria
oracle tolerance, live bytes/allocations, no request-state leak, recompute overhead included.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-21Bandwidth-guided autotuningDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Measured roofline; compare grid, bandwidth rules, tiny predictor with independent timing/correctness.
Planning detail, not a new assignment
stable bandwidth probes and bounded grid; independent heldout shapes; predictor later.
Earlier completion criteria
oracle, tune overhead, noise bounds, unseen shapes, no out-of-regime claims.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-22Energy-aware schedulingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Measured or explicitly estimated energy per successful task, energy-delay, thermal drift and scheduling; no power-limit changes.
Planning detail, not a new assignment
list permitted read-only sensors/units; test scheduling on simulated traces.
Earlier completion criteria
sensor provenance/accuracy, sampling overhead, task denominator, thermal/quality results; estimates labeled and no BIOS/power writes.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-23Exact speculative decodingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Draft/target CPU pairs, draft lengths, selection, acceptance and placement vs target-only with exact contract.
Planning detail, not a new assignment
freeze greedy/sampling equivalence fixture and tokenizer compatibility.
Earlier completion criteria
exact output/distribution contract, rejected work/acceptance, end-to-end quality/latency and contention.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-24Prompt-lookup and n-gram draftingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Lookup, n-gram, suffix and copy-aware drafting vs target decoding.
Planning detail, not a new assignment
deterministic repetition fixture, cache identity/reset, simulated match rates.
Earlier completion criteria
exact target contract, bookkeeping cost, acceptance, memory, task quality; synthetic repetition separated from real workload.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-25Self-speculation and early exitsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Released compatible layer-skip/draft-head and confidence depth; training/architecture changes are separate.
Planning detail, not a new assignment
checkpoint/license/runtime inventory and intermediate-state support; synthetic policy fixture.
Earlier completion criteria
compatibility proven, numerical and locked-task quality, calibrated early-exit failures; no training absent admission.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-26Small-to-large local model routingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Always-large/rules/confidence/specialist cascades at matched utility/resources.
Planning detail, not a new assignment
freeze labels/cost, replay fixed rules, then fit confidence on development only.
Earlier completion criteria
escalation and total cost, worst-group accuracy, false-safe routing, independent heldout gate.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-27Adaptive reasoning stopsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Fixed vs difficulty-aware reasoning budgets and step-verified stops on exact-checker tasks.
Planning detail, not a new assignment
define budget, stop signal and premature-stop categories; simulate on development.
Earlier completion criteria
correctness, early-stop failures, tokens/time, checker independence by difficulty; explanation length is not correctness.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-28Verified candidate searchDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Fixed/adaptive self-consistency, verifier-gated search, diversity and stopping at equal budget.
Planning detail, not a new assignment
cap total candidate tokens/CPU; fixture known false positives; freeze stop policy.
Earlier completion criteria
every candidate cost, diversity, verifier false accepts, correctness under single-candidate budget match.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-29Grammar-constrained generationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Prompt vs grammar/schema constraints including lexicon/output masks; separate valid format from useful correctness.
Planning detail, not a new assignment
freeze schemas, valid/invalid cases, parser oracle and runtime support.
Earlier completion criteria
schema validity, extraction correctness, usefulness and overhead distinct; valid format alone is not success.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-30Verifier-filtered local distillationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Locally generated, independently verified examples for student, rejection sampling, compressed reasoning or drafter; CPU training feasibility first.
Planning detail, not a new assignment
cost worksheet, teacher/student provenance, synthetic checker examples only.
Earlier completion criteria
feasible cap, dedup/no leakage, independent filtering, heldout student vs unchanged/larger model; stop if infeasible.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-31Small-model CPU LoRADraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
CPU LoRA time, optimizer/gradient/activation RAM and tiny heldout gains vs unchanged/larger models.
Planning detail, not a new assignment
derive cost from architecture; tiny fixture and stop cap before training.
Earlier completion criteria
peak RSS/CPU time/adapter bytes, heldout gain and forgetting, baseline parity, contamination audit.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-32Memory updates versus weight updatesDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Retrieval/adapters/RAG fine-tuning/verified bank on new facts/procedures with stale-fact/forgetting checks.
Planning detail, not a new assignment
freeze fact add/update/delete timeline and create provenance-aware fixture; weight updates wait for training gate.
Earlier completion criteria
freshness, provenance, supersession/deletion, retention and resource cost; no adaptation/evaluation leakage.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-33Tiny recurrent and state-space memoryDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Tiny CPU recurrent/state-space probes vs matched transformer/external memory on synthetic checkable sequence tasks.
Planning detail, not a new assignment
exact sequence oracle and fixed time/RAM feasibility.
Earlier completion criteria
synthetic labeling, matched size/compute, scaling/recall failures, fixed stop; no broad reasoning extrapolation.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-34Independent task parallelismDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Task replicas, pipelines, heterogeneous assignment/work stealing/distributed farming vs serial and model sharding.
Planning detail, not a new assignment
simulate synthetic queues; verify per-node allocation interface before distributed trial.
Earlier completion criteria
correct tasks per total CPU-second, transfer/scheduling overhead, fairness, recovery and node receipts.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-35CPU layer and pipeline shardingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
CPU layer/RPC splits on explicitly approved verified nodes vs one-node/constrained baseline.
Planning detail, not a new assignment
derive layer/activation bytes; simulate split/failure locally; require exact host approval before RPC.
Earlier completion criteria
per-node RSS, serialization, sync/tail latency, output equivalence and node-loss recovery.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-36Tensor-parallel communication feasibilityDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Derive/measure bytes per token, link/sync cost before implementation; compare whole-task and pipeline.
Planning detail, not a new assignment
analytic model from verified shapes plus synthetic latency sensitivity.
Earlier completion criteria
independently checked bytes/collectives, measured link separately labeled, overlap and decision threshold.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-37Specialist committees and verificationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Cross-family error correlation, voting, draft/verify, independent chains/ensembles at equal compute; count correlated false accepts.
Planning detail, not a new assignment
checker-backed tasks and all-in compute; simulate correlated error fixtures.
Earlier completion criteria
joint errors/false accepts with intervals, latency/correct-task cost, single-model baseline, verifier independence.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-38CPU hybrid retrieval and rerankingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
BM25/vector/hybrid/rerank/graph/adaptive depth at fixed corpus/query/context, including storage indexes and parse cache.
Planning detail, not a new assignment
freeze corpus/version/labels/parser and cache invalidation fixture.
Earlier completion criteria
recall/nDCG plus answer/citations, equal context, build/query cost, stale/delete tests; labels independent.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-39Temporal source-linked evidence graphsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Flat retrieval/tables/graphs with source links, dates, supersession, reliability, contradictions, citation checks; prove graph adds beyond database.
Planning detail, not a new assignment
claim schema and validator; create contradictory/superseding fixture; table baseline.
Earlier completion criteria
temporal/citation/contradiction correctness and maintenance cost, demonstrated graph benefit.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-40Hierarchical episodic and procedural memoryDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Hierarchical summaries, episodic/semantic, verified procedures/scaffolds across sessions with provenance and later correctness.
Planning detail, not a new assignment
keyed multi-session fixture; every summary statement keeps source ID; fixed memory budget.
Earlier completion criteria
recall, provenance through hierarchy, procedure execution, stale conflict handling, equal tokens/memory.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-41Incremental indexing and deletionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Additions/changes/deletes/staleness/parser invalidation versus full rebuild on seeded drift.
Planning detail, not a new assignment
content hash/version/tombstones and deterministic change sequence; clean rebuild oracle.
Earlier completion criteria
incremental result equals rebuild, deleted records unavailable, caches invalidated; quality and cost.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-42Executable database queriesDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
SQL/schema-aware queries, EXPLAIN/validation and deterministic records versus prose retrieval.
Planning detail, not a new assignment
freeze schema/units/exact oracle; readonly query fixture with malformed/null cases.
Earlier completion criteria
query correctness, units, record provenance, safe limits, query vs answer errors separate; cannot mutate source/tests.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-43Calibrated abstention and confidenceDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Calibrate confidence/selective prediction and abstention for semantic uncertainty/source conflicts.
Planning detail, not a new assignment
freeze splits, independent correctness/conflict labels, abstention utility.
Earlier completion criteria
calibration and risk/coverage, false confidence and subgroup/conflict intervals; no threshold tuning on test.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-44Verifier-backed local task solvingPreparation complete but not admitted or tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Model-only/deterministic/model-plus-tools on arithmetic, grounded QA, planning; equalize assistance and attribute gains.
Planning detail, not a new assignment
preserve preparation and 45/45 engineering receipt; resolve independent replay, capacity/CPU access, dispatch and evaluator gaps.
Earlier completion criteria
independent answers, equal help, sealed set, model-only vs full-system separation.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-45Constrained program synthesisDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Generate against fixed specs/tests/schemas/resource limits; compare one-shot/verified search without test editing.
Planning detail, not a new assignment
immutable tests and sandbox contract; adversarial attempts to alter/read tests.
Earlier completion criteria
expected outputs external, hidden property tests, process/network/filesystem containment and resource caps.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-46Reproducible data-analysis agentsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Local SQL/code analysis vs independent numeric answers on locked data/schema/units; rerunnable scripts and receipts.
Planning detail, not a new assignment
tiny frozen synthetic data/oracle, schema/units and rerun manifest.
Earlier completion criteria
answer within prespecified tolerance, rerun from locked inputs, units/filters/source hashes recorded; executable vs model-only separated.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-47Cost-aware experiment selectionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Compare random/round-robin/Bayesian/bandit/information/resource planners and fixed/model-directed selectors using simulation first.
Planning detail, not a new assignment
independent objective and synthetic/history queue replay, fixed cost/stops.
Earlier completion criteria
utility/regret per resource, budget violations, stability; planner cannot certify its own score.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-48Bounded failure triage and code repairDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Injected isolated failures; restart-only vs log diagnosis/limited patch; preserve failures and untouched regressions.
Planning detail, not a new assignment
immutable known-fault fixtures, exact allowlist, at most two repair attempts.
Earlier completion criteria
hidden untouched regression groups pass, expected tests cannot be changed, all attempts kept.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-49Checkpoint recovery and result commitmentPreparation complete but not admitted or tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Isolated interruption/journal/lease/identity/idempotent commit/replay; test actual interruption types.
Planning detail, not a new assignment
preserve interface handback and six NOT_RUN groups; specify minimal shared dispatch/reservation/commit contract; fixture crash/timeout/stale worker only.
Earlier completion criteria
atomic accounting, lease fencing, idempotent receipts, raw retries; do not claim reboot support untested. Existing evidence found no shared dispatch path.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-50Protected evaluation and research integrityPreparation complete but not admitted or tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Immutable holdout, separate evaluator, anti-tamper/contamination and independent receipts; candidates cannot rewrite tests or self-certify.
Planning detail, not a new assignment
preserve hold recommendation; resolve evaluator custody, atomic accounting, path/hardlink/symlink safety, stale fencing, independent replay.
Earlier completion criteria
independent identity/custody, cumulative usage, CPU/model/helper evidence and outside replay.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-51Learning from negative experimentsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Complete positive/negative ledgers vs positive-only/stateless planners for repeated errors and justified next steps.
Planning detail, not a new assignment
append-only ledger schema and synthetic repeated-failure/exception histories.
Earlier completion criteria
full denominator, repeat mistake rate, next-step calibration, no dropped nulls/inconclusives.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-52Unattended endurance and driftDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Bounded soak for leaks, pressure degradation, task/input drift, injection and restart, with quality and telemetry.
Planning detail, not a new assignment
finite synthetic endurance/fault simulator; specify supervisor stops before live test.
Earlier completion criteria
resource trends, graceful stop, cursor integrity, injection containment and quality; simulation labeled.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-53PC hybrid-core schedulingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Compare P/E placement/thread/prefill/decode on qualified PC.
Planning detail, not a new assignment
literature/protocol only until a PC is explicitly available/authorized.
Earlier completion criteria
verified topology/affinity, same binary/model/task, thermal/resource record; no Xeon extrapolation.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-54CPU runtime and build comparisonsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Compatible CPU runtime/build/compiler/allocator/portability comparisons after same-model/settings equivalence.
Planning detail, not a new assignment
inventory commits/backends/features/formats; output oracle; no production install/change.
Earlier completion criteria
equivalence first, fallback path and provenance, performance separate; CUDA presence does not qualify CPU.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-55Distributed prefill and KV migrationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Simulate then conditionally test split prefill/cache transfer/checkpoint migration vs one-node/recompute.
Planning detail, not a new assignment
synthetic KV identity/transfer model and approved-node inventory; no network test before grant.
Earlier completion criteria
cache compatibility, bytes/time, recompute baseline, output equivalence, per-node receipts/recovery.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-56CPU model and adapter mergingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Merge compatible task-arithmetic/TIES/evolutionary checkpoints/adapters; compare every constituent and protect task holdouts.
Planning detail, not a new assignment
identity/license compatibility; validation-only selection; estimate merge/repack costs.
Earlier completion criteria
per-task/worst-task vs all constituents, heldout untouched, CPU I/O recorded.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-57Continual learning and forgettingDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Sequential adapters/replay/regularization/routing on fixed sequences; old-task forgetting, new quality, cost/errors.
Planning detail, not a new assignment
freeze arrival sequence/splits; tiny exact fixture; compute feasibility before training.
Earlier completion criteria
forgetting curve/new gain/routing/cost and no omitted degraded tasks or leakage.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-58CPU preference optimizationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Tiny CPU DPO/preference and quantized adapter-scale vs SFT/unchanged on approved deterministic preferences.
Planning detail, not a new assignment
verify fixture authority/provenance, derive cost, freeze arms; stop if infeasible or labels unavailable.
Earlier completion criteria
labels unchanged, compute matched, heldout behavior/regression and resource record.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-59Formal proof verificationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Local generate-check-repair with real proof checker; informal text is not proof.
Planning detail, not a new assignment
read-only tool/version/library discovery and tiny public theorem fixture.
Earlier completion criteria
checker-accepted artifact and theorem deps, reproducible env, bounded repairs, compute per checked result.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-60CPU document extraction and OCRDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Deterministic parsing/table extraction/CPU OCR or vision/local model on frozen documents with field keys and regions.
Planning detail, not a new assignment
define frozen set/independent keys, inspect CPU support, build region fixture.
Earlier completion criteria
field precision/recall, region, coverage/unsupported fields and parser/OCR/model effects separate.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-61Embedding and index compressionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Embedding quantization/vector-index compression/embedder adaptation at fixed corpus/query conditions.
Planning detail, not a new assignment
freeze corpus, queries, labels, embedding model and build; compression round-trip fixture.
Earlier completion criteria
recall/nDCG/downstream correctness, RAM/storage/latency/build; adapt only development queries.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-62Low-rank weight compressionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
CPU low-rank approximation plus quantization vs uniform at equal bytes, including decomposition/repacking and task quality.
Planning detail, not a new assignment
tiny matrix oracle; compatible tiny model and full cost estimate; rank selected on development.
Earlier completion criteria
total factor bytes, conversion CPU/RAM, inference support and heldout quality; matrix error is not task quality.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-63Dense-to-sparse MoE conversionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Probe FFN activation overlap, clustering, gating and upcycling; post-hoc quality and speed are separate hypotheses.
Planning detail, not a new assignment
development activation overlap and routing simulator; keep conversion distinct.
Earlier completion criteria
heldout routing stability, bytes/params, quality and measured speed separate; no conversion/training absent grant.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-64Sleep-time and anticipatory computationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Verified background summaries/likely answers/caches vs on-demand at equal CPU, measuring hit value, staleness, correctness and waste.
Planning detail, not a new assignment
simulate demand traces and TTL/invalidation; no persistent worker until admitted.
Earlier completion criteria
useful hit/task rate, stale harm, total background+foreground CPU, invalidation correctness.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-65Statistical power and adaptive evaluationDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Paired uncertainty, variance, seeds, sequential/multiplicity and subsampling vs frozen full suite without relaxing criteria post-result.
Planning detail, not a new assignment
estimands/units from evaluator contract; simulate declared effect/power and freeze stop method.
Earlier completion criteria
interval/coverage and error calibration, all seeds/items counted, fixed threshold/no peeking.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-66Programmatic tasks and contamination controlsDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Schema tasks with independent oracles; dedup, consistency, train/test separation and contamination; no weak-oracle tasks in acceptance.
Planning detail, not a new assignment
seeded generator and oracle provenance; near-duplicate/split overlap fixture.
Earlier completion criteria
independent correctness, deterministic regeneration, contamination audit, uncheckable synthetic tasks explicitly exploratory.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-67CPU-scale paper reproductionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Select CPU-feasible published work, verify primary source/code, preserve assumptions and reproduce principal result; GPU-only excluded unless validated CPU port.
Planning detail, not a new assignment
literature screen for data/license/protocol/compute; deviation table before resource request.
Earlier completion criteria
primary-source trail, protocol/data/seed/metric match, deviations, independent reproduction/uncertainty; demo not reproduction.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-68Input-adaptive precisionDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Tiny CPU token/layer precision and activation thresholds vs fixed precision at matched average cost; isolate numerical and switching overhead.
Planning detail, not a new assignment
deterministic development policy, switch/numerical oracle, fixed matching rule.
Earlier completion criteria
numerical bounds, measured switching overhead, matched average resources, sealed noninferiority; no holdout tuning.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
MM-69Funding and research evidence packagesDraft and not tested.
No scientific execution. Status reported in the combined portfolio dated October 19, 2026.
Research question from the earlier plan
Research-only funding map and evidence packages; validate deadline, eligibility, org requirements, cash vs credits, unknowns; draft only, no contact/submission.
Planning detail, not a new assignment
source/date schema; official program research; map only accepted evidence and gaps.
Earlier completion criteria
dated primary citations, unknowns visible, award/application not implied, no outreach; accepted experiment receipts only.
Planning details retain their October 14 evidence date. Preparation complete means not admitted and not tested; no proposed performance outcome is established.
No matching topic. Clear the search to see all 69.