Manuscript Psy03-C@1 · Psy03-C-v1.tex · section · d3380b90
Claim-status and debugging ledger
Claim-status and debugging ledger · lines 2652–2722
| and exact boundaries of the principal claims. | ||
|---|---|---|
| Claim | Status | Boundary |
| Claim | Status | Boundary |
| The recurrent source is the earlier hidden-noise MLP unrolled | Excluded | Different input and hidden perturbation interfaces. |
| Ordinary Gaussian stress makes hidden states unbounded | False under stated controls | External ReLU invariant cube; external tanh range bound. |
| Bounded states prove contraction | False | A separate gate-aware or tube certificate is required. |
| Zero recurrent matrix removes every carry path | False | Update-gate interpolation retains the preceding state. |
| Recurrent gate spectral radii certify stability | Insufficient | Full cell Jacobian includes gates, candidate, and biases. |
| Twenty repeated steps provide twenty independent task samples | False | One latent point is repeated; extra per-step corruption is separate. |
| Rotation/scaling loss proves cognitive impairment | Not established | Invertible shift preserves ideal label information. |
| Fine and high-noise tasks have exact reference ceilings | Proved ideal laws | Equal-prior Cartesian Gaussian centers. |
| Blend implements conflicting voice labels | False for this evaluator | Voice labels are not read. |
| More distractor weight monotonically destroys information | False in the ideal noiseless blend | Exact collision-dependent Bayes reference. |
| Lower clean-to-probe drop means better probe performance | False generally | Report absolute accuracy and checkpoint contrasts separately. |
| Battery confidence follows from recurrent step count | False | Independent unit is the complete latent example block. |
| Scar correlation establishes scar causation | Not established | Paired admitted interventions and common probe law required. |
| Three-seed and four-rate iso controls are unused | False in this version | They are used; matching tolerance remains ignored. |
| Quantum carrier contraction erases different scar distributions | Not established | Persistent classical labels can retain distinguishability. |
| Delayed phase-label retention is exact | Proved finite pilot | Specified channels, encoding, inputs, and readout. |
| Phase sensitivity proves quantum advantage | Not established | No scalable baseline or resource separation. |
| Clinical mechanism or inherited analytic admission | Outside scope | No biological correspondence or characteristic construction. |