TFPT scientific source excerpt
Original repository path: experiments/theory-contracts/collective-native-marginal-origin-20261026/COLLECTIVE-MACHINE-RUNS.txt
Original file SHA-256: 2ac6e05655b8b20d4b3d4b87d3dad6bdc2b497ef14a5c8ed1cecf28fe4f0568f
Inclusive original lines: 1-61
Line numbers below refer to the original source.
This bounded excerpt is source evidence; it is not a new proof/checker run.
Conversation metadata and personal execution details are omitted where indicated.

000001 | DECLARED RECURSIVE MACHINE: BASELINE, LONG RUN AND RECOVERY EVIDENCE
000002 | 26 October 2026. Existing measurements consolidated into the Atlas.
000003 | MATH.COLLECTIVE-MACHINE.FINITE-RUNS.01
000004 | 
000005 | Scope: the explicitly chosen60x35 complex SOM update already specified by
000006 | RECURSIVE-SOM-AUDIT.txt and premise:declared-recursive-native-som-update.
000007 | This records finite development experiments, not a fundamental selection,
000008 | a universal attractor theorem or an original physical TFPT clock.
000009 | The unmodified source-dependent alignment and chosen rates remain inputs.
000010 | Kohonen/Honkela2007 is background for competition/neighborhood adaptation:
000011 | https://www.scholarpedia.org/article/Kohonen_networks . The particular TFPT
000012 | recursion is the authorized model study, not an attribution to Kohonen.
000013 | 
000014 | 1. Baseline controls
000015 | The original experiment_results.json contains six runs of10000 steps with
000016 | seed7 and20 independent seeds of5000 steps. The baseline uses c=.5,
000017 | feedback=.25,learning=.15,neighborhood=.25 and initial noise .05.
000018 | Final effective ranks and native five-code fractions are:
000019 |  declared recursion:3.7940762,0.9806738;
000020 |  feedback off:4.9998200,0.9992813;
000021 |  frozen memory:18.3358291,0.6388766;
000022 |  interaction off:1.0000000,0.0734841;
000023 |  kernel magnitudes replacing phases:1.0000000,0.1045831;
000024 |  exact initial frame without noise:3.7941953,0.9813609.
000025 | Across20 seeds ranks range3.45815794..4.99994625 and code fractions
000026 | 0.97951570..0.98453656. Effective rank is exp(entropy of the normalized
000027 | signal covariance spectrum), not a spacetime dimension. This finite
000028 | population does not show start-independent convergence to one state.
000029 | 
000030 | 2. Continuing motion and finite response explanation
000031 | The original baseline checkpoint at10000 was continued to100000 steps.
000032 | Final rank3.7940773261 and code fraction.9810936913 stay approximately
000033 | stable, while signal change per step remains.0003771189581. Stationary
000034 | aggregate plots therefore do not establish a fixed microscopic state.
000035 | The actually reconstructed frozen120x120 response has two complex mode
000036 | pairs with periods11201.2123 and64540.3822 iterations. It predicts the next
000037 | 1000 actual nonlinear steps with relative error5.4758727e-14 at this run.
000038 | It is explicitly NOT the full Jacobian or a result for arbitrary histories.
000039 | The later rate audit already shows these periods depend on chosen rates.
000040 | 
000041 | 3. Perturbation and restore
000042 | A signal perturbation of strength.1 gives RMS difference.09890549 initially,
000043 | .01182243 after10 steps and.01101184 after5000. The aggregate rank returns
000044 | near3.794 while a residual displacement remains. This is finite recovery,
000045 | not exact forgetting or a universal robustness theorem.
000046 | The original checker records unit signal norms, lack of memory influence
000047 | when feedback=0,and exact save/cold-load/random-perturb/continue agreement.
000048 | During this consolidation the latter was independently REEXECUTED in a
000049 | temporary file:137 steps,save,fresh load,identical random perturbation and
000050 | 123 further steps. Signals,memories,history and events agree exactly in
000051 | the same current NumPy/runtime environment. No live UI process was changed.
000052 | The stored10000 and100000 checkpoints were loaded and checked against their
000053 | reported measurements. Long runs were not reexecuted during this audit.
000054 | 
000055 | 4. Evidence and qualification
000056 | baseline_machine_evidence/ preserves experiments.py,inspect_dynamics.py,
000057 | machine.py,native inputs,complete measurement JSONs and selected checkpoints.
000058 | baseline_summary.json is a compact extraction; provenance.json pins bytes.
000059 | The later U35/rate findings are essential qualifications: the recursion
000060 | alone chooses neither the absolute native frame nor a physical time unit.
000061 | These baseline results are not evidence for a complete fundamental theory.