What changed:
the world, the system,
or your view?
This system keeps doing the same thing.
Is it stuck, or is its world staying the same? Examine the evidence. Compare the interpretations. Then look behind the measurements.
World
What is happening, before anyone tries to explain it.
Revealing changes only your view. The observers never receive the hidden labels.
Measurement
Two imperfect views: one of the target, one of the occupied state.
8 vectors per packet / noise variance 1.00
Bright squares show measured relationships, not certainty. A short, noisy packet can resemble the wrong organization.
Interpretation
Same measurements. Different treatment of the target's history.
Target/state mismatch probability: 6.1%
This is not KL remainder.
Before the reveal, what would you conclude?
One trace need not identify one mechanism.
The science behind the scene
Dashed line: the original fitted plug-in M1, using a score from the current U packet but no target-history model. Solid line: archived J0. Emissions and transition parameters were fitted on independent training data. Selecting a different archived joint fit does not change the study decision.
World, then evidence, then inference
P(H_t = h | H_(t-1) = h') = Q[h', h] P(Z_t = B | H_t, Z_(t-1)) = sigmoid(a(2H_t - 1) + b(2Z_(t-1) - 1)) U_t ~ N(0, Sigma_U[H_t]); Y_t ~ N(0, Sigma_Y[Z_t])
H and Z each take labels A or B. A pairs channels 1-2 and 3-4; B pairs 1-3 and 2-4. These are predefined synthetic organizations, not discovered neural units. A source has variance 1 and within-pair covariance 0.6. Independent noise is added to the diagonal.
predict joint (H,Z) -> assimilate current U -> score current Y -> assimilate Y P(H != Z | measurements) = P(A,B | measurements) + P(B,A | measurements)
The joint filter carries all four probabilities. Its target estimate can use both measurement streams. Multiplying separately estimated target and state marginals would incorrectly ignore their dependence. A model update has no assigned physiological duration.
Probability is not remainder
R(K | p) = inf over q in Q(K) of D_KL(p || q)
The mismatch probability shown here is a binary-state diagnostic, not this KL quantity, not an estimate of all unresolved possibilities, and not a measurement of C or M. The simulation probes one operational problem within the framework.
Read the actual measurements and probabilities
Each grid is a packet correlation estimated from zero-mean sample second moments, not an EEG trace. The complete covariance packets are included in an exported run.
| Ch. | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| 1 | 1.000 | -0.057 | -0.338 | -0.735 |
| 2 | -0.057 | 1.000 | -0.075 | -0.362 |
| 3 | -0.338 | -0.075 | 1.000 | 0.196 |
| 4 | -0.735 | -0.362 | 0.196 | 1.000 |
| Ch. | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| 1 | 1.000 | -0.643 | -0.089 | -0.473 |
| 2 | -0.643 | 1.000 | -0.211 | 0.165 |
| 3 | -0.089 | -0.211 | 1.000 | 0.562 |
| 4 | -0.473 | 0.165 | 0.562 | 1.000 |
After this packet: target B 3.0%, occupied B 7.8%, mismatch 6.1%. Joint entropy 0.3599 nats. Current conditional Y log scores: joint -55.22852, comparison -55.49819 nats/update (higher is better). These probabilities do not include parameter uncertainty; a single score does not select a mechanism.
What this does and does not establish
A computed illustration
This module uses standard Gaussian likelihoods and hidden-state filtering. An equivalent conventional implementation has the same predictions. Model uncertainty, parameter uncertainty, and uncertainty about the permitted model families are not the same thing.
A bounded research result
In v0.4's fresh cohort, the joint comparison selected extra inertia in 0/24 no-inertia cases at each setting. It selected real inertia in 24/24 favorable, 2/24 moderate, and 0/24 severe cases. These counts depend on the design and decision rule. Non-selection does not establish absence.
An untested empirical bridge
The two separable streams, Gaussian emissions and predeclared identities are assumptions. No human data, unknown sensor mixing or biological mechanism is tested here. No result establishes a measure of conscious experience.
Connection to CCC and CCR
CCC asks about presence, identity, closure and stable description, with a separate cosmological conjecture. CCR develops an operational program for organized content. This example illustrates why access and target history constrain an interpretation. It does not prove or disprove C, M, or cosmogenesis.
Source, settings and reproducibility
Frozen research output with regenerated measurements. The browser displays cached reference probabilities. It does not train a model or rerun the 288-case study.
{
"module": "0.6.0",
"mode": "replay",
"case": "fresh_memoryless_moderate_00",
"cohort": "fresh",
"replicate": 0,
"sequence": 0,
"updates": 512,
"sourceVersion": "0.4 primary",
"sourceSHA256": "f89198e99c3d2c39d3bbb4665fb6dcc9c459fb00eab22b1495022ae5d51a6f26",
"seedEntropy": [
2026090704,
0,
1,
0
],
"selectionRule": "mean conditional Y validation log gain > 0.005 nats/update AND one-sided 95% t lower bound > 0 over eight independent sequences",
"displayedJoint": "J0",
"fitFlags": {
"joint": [
{
"converged": true,
"optimizerWarnings": 0
},
{
"converged": true,
"optimizerWarnings": 0
}
],
"plugin": [
{
"converged": true,
"optimizerWarnings": 0
},
{
"converged": true,
"optimizerWarnings": 0
}
]
},
"jointParameters": {
"CU": [
[
[
2.0141024291260403,
0.5963949022785234,
0,
0
],
[
0.5963949022785234,
2.0413638009228547,
0,
0
],
[
0,
0,
1.963198897808428,
0.5644726286982282
],
[
0,
0,
0.5644726286982282,
1.922427432007178
]
],
[
[
1.993962693516233,
0,
0.6417943117137044,
0
],
[
0,
1.9874216109216585,
0,
0.6017744241197145
],
[
0.6417943117137044,
0,
2.0143545306204276,
0
],
[
0,
0.6017744241197145,
0,
2.052720544542622
]
]
],
"CY": [
[
[
1.9815051909704098,
0.6128300902275976,
0,
0
],
[
0.6128300902275976,
1.9470293950494253,
0,
0
],
[
0,
0,
2.0789115984776636,
0.6615457153660484
],
[
0,
0,
0.6615457153660484,
1.9909243280343067
]
],
[
[
1.9236069310306474,
0,
0.5706082015118743,
0
],
[
0,
1.9702263418525612,
0,
0.5023519553497073
],
[
0.5706082015118743,
0,
1.9731394471061654,
0
],
[
0,
0.5023519553497073,
0,
2.0245528911282857
]
]
],
"q": [
[
0.9382671229284822,
0.061732877071517735
],
[
0.07441501563305666,
0.9255849843669434
]
],
"pi": [
0.6800735050652533,
0.3199264949347467
],
"a": 2.0772195649514376,
"b": 0
},
"pluginParameters": {
"C": [
[
[
2.000218955536514,
0.6243139755989062,
0,
0
],
[
0.6243139755989062,
1.9362456695116697,
0,
0
],
[
0,
0,
2.0808856635067463,
0.6637339314475732
],
[
0,
0,
0.6637339314475732,
2.006394547282432
]
],
[
[
1.905573091501818,
0,
0.5349583954555523,
0
],
[
0,
1.980996237895397,
0,
0.4802157656065041
],
[
0.5349583954555523,
0,
1.9749424146783756,
0
],
[
0,
0.4802157656065041,
0,
2.0066150705716157
]
]
],
"theta": [
1.0235926922864254,
1.9359430603226175,
0
],
"D": 1
}
}The archived case is not chosen for a flattering result: this release uses fresh replicate 00, first test sequence, for both generators and all three settings. Case numbering conceals the mechanism until the reveal. Original study decisions are retained, including non-selections and numerical warnings.
Read the research report / Shared reconstruction specification