Persistent observations and a minimal deterministic controller

Question

How much calibration saving is explained by ordinary persistent observations and deterministic scientific execution, and what does the LLM add?

Hypothesis

At matched correctness, stored authentic observations can reduce revisit calibration. A deterministic procedural controller may achieve the same task without model inference. These predictions are tested, not assigned as new-fixture outcomes.

Registration

Locally frozen at 565e8472c5275e4561efd3cffa78c2fd9fe21915 before model dispatch. This revision publishes registration before execution; actual public verification receipt is retained separately and dispatch is gated on that receipt. Generator seeds 2026101402 and 2026101403, fixed policies, fixtures, order, scoring, budgets and recovery rules. No backdated public availability.

Methods

Two independent blocks, three fresh apparatuses and six tasks A,B,C,A,C,B per block. Exactly the same fixtures across four conditions. Fresh stores per block/condition; fresh task contexts; all controllers restart after task three. Cyclic condition rotations balance each condition across serial positions three times. Same frozen scientific runtime and model configuration as earlier memory-new-fixtures-v1. Full generator, filters, memory restrictions and recovery are in the registration artifact. Neutral identifiers assigned independently of dynamics. Constructed outcome-changing revisits and sufficient-calibration guarantee restrict the tested distribution. All 23 rejected candidates and reasons retained; 7 in block one, 16 in block two. No model calls for fixture selection.

Baselines

LLM_PERSISTENT/LLM_EPISODIC choose actions with the same model. RULE_PERSISTENT/RULE_EPISODIC implement the documented schedule with zero inference. Persistent memory contains authentic observations only; episodic access is current-task only. Exact apparatus identity matching; estimates recomputed every task. Constant labels and historical-label-copying references included.

Evaluation

48 scheduled task executions, 12 per condition, 24 involving the LLM. Main unit six apparatuses in two blocks; dependent revisits reported separately. +1 correct, -1 incorrect, 0 explicit abstention; missing terminals and failures separate. Report calibrations, selected sources, scientific operations, retrievals, rejected requests and resources for every task. Compare matched-correct persistent/episodic and rule/model performance. No precise population estimate.

Resource ceilings

192 total model requests including auxiliary; 1600000 full logical input tokens including cached input; 98304 generated; 512 generated and 300 seconds per request; 14400 seconds actual loading/provider-request wall time. Eight decisions per model task, eighth finish only. Zero auxiliary calls, warmup, transport retries, downloads or paid inference. Rule/offline runs zero inference. Loading nested in request elapsed, counted once. Preparation/reporting outside inference budget.

Stopping conditions

Unknown usage or operation execution stops the entire partial study. Committed checkpoint or persisted known response recovery must use the same allowance and immutable journal fences. Canceled dispatch stays spent and unknown dimensions remain reserved. No task replacement, adaptive prompt/fixture change or outcome-selected retry.

Executing code

Isolated worktree and new frozen runtime directory; production digest e668e42492ade662b3b3f2f2f5754cca54be45c01b94ef1fcff6fb603d87ad64. Same scientific operations, numerical semantics, sensor precision, controller identities, final-slot behavior and late-response fences. Sealed hashes included; private source withheld.

Configuration

qwen3.8:27b-q4_K_M, digest 25b843619e944cd0ae6069f94ff4e5e26a16e109ccbc0a66a0f05979ed70098e; seed 424242, temperature 0.7, top_p 0.8, top_k 20, min_p 0, context 8192, cap 512, think=false. Six-decimal sensors, full inclusive parameter intervals; calibration geometries x=0.25 and 0.3, six steps, strict target separation reduction >0.0001. Rule baseline uses the same public APIs and observation provenance.

Complete outcomes

Execution pending. Offline earlier frozen-fixture verification measured both rule controllers 6/6 correct with three persistent versus six episodic calibrations. These results concern earlier fixtures only. The retained previous model comparison remains 6/6 each, five versus seven calibrations, 40 versus 35 requests. Its five extra requests reconcile as three extra retrievals, two fewer calibrations and four extra rejections, with other accepted operation counts equal. Accounting is not a counterfactual trajectory claim.

Realized resources

New-study inference usage zero at registration. Historical uncertain reservations remain separately held: 34852 input tokens, 1024 generated tokens, 600 seconds. Actual model requests/tokens, scientific calls, calibration simulation, exclusive storage/journal/retrieval, other processing, guards, loading and total task wall time will be reported separately. Rule journal decisions are not provider requests. Hypothetical added calibration-cost scenarios use seconds/calibration and remain separate from measured simulator timing.

Limitations

Six apparatuses, two selected simulator blocks, stable simple dynamics, exact source identity matching, engineered changed outcomes and each legal calibration sufficient. Results are conditional on this constructed distribution. Matching rules show this task admits a simpler implementation; they do not establish absence of LLM value elsewhere. An LLM advantage requires faithful baseline and equivalent information. No consciousness inference. Token counts imply no FLOPs, energy, monetary cost or provider determinism.

Reproducibility

Nine new focused tests and 21 journal/identity/recovery tests passed in separate interpreters. A combined first attempt had historical fixture-order adapter collisions; separate interpreters resolve the harness conflict without protocol changes. Offline earlier-fixture baseline verification forbids provider connections. New fixtures generated reproducibly with retained rejections. Closeout will reconstruct all requests from journal event prefixes, verify pure rule responses, recompute scientific facts and scoring, and replay accepted operations without effects.

Revision record

2026-10-14, r1: prospective public-safe registration. Execution has not begun. A subsequent immutable revision will report all results including negative, failed or incomplete outcomes.

Reviewed evidence & downloads

• registration — derived. Selected project-authored factual data and policy description; redistribution permitted.
• rule policy — derived. Selected project-authored factual data and policy description; redistribution permitted.
• summary — derived. Selected project-authored factual data and policy description; redistribution permitted.
• tool contracts — derived. Selected project-authored factual data and policy description; redistribution permitted.

These are selected public derivatives; raw transcripts, internal configuration and private source are withheld. Hashes describe the actual downloads, not unmodified original evidence.

Artifact manifest and source provenance · Permanent revision URL

Substantive versions

• Revision 1