v2.20.0: Experiment 17 -- term-order x noise interaction (real positive result)

Experiment 17: term-order x noise interaction check

Experiment 16's own caveat asked whether term-order sensitivity behaves
differently under noise than it does noiselessly. It does.

Method: same original-vs-reversed K=10+10 SYK term ordering as
Experiment 16, but with a depolarizing Kraus channel injected after
each protocol phase (noise_p=0.01), delta averaged over 6 noisy
trials per order (common random numbers between mu signs to isolate
the sign effect from noise-realization variance).

Result -- the first candidate in this whole project that stays
significant as the sample grows, instead of regressing to nothing:

n	r	p
6	+0.811	0.050
20	+0.587	0.0065
30	+0.396	0.030
50	+0.340	0.0158

For comparison, Experiment 16's noiseless version of the identical test
collapsed from p=0.34 (n=6) to p=0.13 (n=30) over the same range --
that one was an honest negative result. This one is not: the point
estimate shrinks as expected with n, but stabilizes around r~0.34-0.40
and stays under p=0.05 at every step from n=6 to n=50.

25/50 (50%) of the n=50 sample are still wrong-signed, so this is a
modest, partial structural handle on the sign -- not a resolution of
it. But it is the first of nine independent candidates tested across
this project to show a real, replicated correlation.

Full write-up: https://tatopenn-cell.github.io/Dense-Evolution-Discovery/wormhole_syk_teleportation/

---

Note: supersedes v2.19.0, whose tag was accidentally pointed one
commit early (Experiment 16 instead of Experiment 17) despite its
release notes already describing this experiment -- this release's tag
correctly includes Experiment 17's actual commit. Created to also
trigger Zenodo's GitHub integration for this repository.

v2.19.0: Experiment 17 -- term-order x noise interaction (real positive result)

Experiment 17: term-order x noise interaction check

Experiment 16's own caveat asked whether term-order sensitivity behaves
differently under noise than it does noiselessly. It does.

Method: same original-vs-reversed K=10+10 SYK term ordering as
Experiment 16, but with a depolarizing Kraus channel injected after
each protocol phase (noise_p=0.01), delta averaged over 6 noisy
trials per order (common random numbers between mu signs to isolate
the sign effect from noise-realization variance).

Result -- the first candidate in this whole project that stays
significant as the sample grows, instead of regressing to nothing:

n	r	p
6	+0.811	0.050
20	+0.587	0.0065
30	+0.396	0.030
50	+0.340	0.0158

For comparison, Experiment 16's noiseless version of the identical test
collapsed from p=0.34 (n=6) to p=0.13 (n=30) over the same range --
that one was an honest negative result. This one is not: the point
estimate shrinks as expected with n, but stabilizes around r~0.34-0.40
and stays under p=0.05 at every step from n=6 to n=50.

25/50 (50%) of the n=50 sample are still wrong-signed, so this is a
modest, partial structural handle on the sign -- not a resolution of
it. But it is the first of nine independent candidates tested across
this project to show a real, replicated correlation.

Full write-up: https://tatopenn-cell.github.io/Dense-Evolution-Discovery/wormhole_syk_teleportation/

v2.18.0 - Wormhole term-order non-commutativity check (Experiment 16, honest negative result)

Tests a different kind of non-commutativity from scripts/channel_order_noncommutativity.py's already-settled noise-channel-order finding (order matters iff at least one channel is non-Pauli): does the ORDER in which the K=10+10 SYK Hamiltonian terms are applied within the Trotterized circuit matter, and does the size of that effect track the sign?

Method: noiseless Trotter protocol run twice per instance (original term order vs. fully reversed), for both mu signs, giving order_sensitivity = |delta_reversed - delta_original|.

An initial n=6 spot-check found the largest point estimate of any candidate tried in this script: r=+0.474 (p=0.342, not significant but notable). Verified on a larger sample given how cheap this experiment turned out to be (~18s/instance). Eighth honest negative result: at the real n=30 (35000 candidates screened to find 30 exact 34/11 matches), the correlation regresses to r=+0.282 (p=0.131) -- still not significant, the same honest-correction pattern as Experiment 11's mode-usage-imbalance finding (r=0.87 at n=6 -> r=0.171 at n=100).

order_sensitivity itself is real and non-zero for every instance (term order genuinely changes the Trotterized circuit's output -- confirming real non-commutativity among the terms), it just doesn't predict the sign. 15/30 (50%) wrong-signed, consistent with Experiment 11's ~49/100.

Full write-up: docs/wormhole_syk_teleportation.md#experiment-16-term-order-non-commutativity-check-run-2026-08-08

v2.17.0 - Wormhole N-scaling check: N=8 vs N=12 (Experiment 15)

Tests whether the sign-dependent instance variance from Experiments 10/11 persists, worsens, or shrinks at a larger Majorana count -- the one structural lever never pulled until now.

The exact backend is infeasible at N=12 (dim^3 diagonalization cost, ~4096x slower than N=8). Both N=8 and N=12 are evaluated here via the Trotterized gate-circuit backend (~19s/call at N=12, measured directly) for a clean, backend-matched comparison, rather than the exact-backend N=8 numbers used in Experiments 10/11 (mixing backends would confound N-scaling with a separate, already-quantified backend effect from Experiment 9).

Scope: n=6 instances per N (not n=100, infeasible at this cost), K_TERMS fixed at 10. The paper's own 34/11 selection criterion has no exact match at N=12 (verified: 0 of 3000 candidates) -- N=12 instances are selected by closest achievable match instead.

Result: wrong-sign rate identical (2/6 at both N, too small a sample to trust as a rate), but mean |delta| drops from 0.00765 (N=8) to 0.00034 (N=12) -- roughly a 22x reduction, present in every N=12 instance individually. Consistent with the signal weakening toward a thermodynamic limit, but equally consistent with the paper's fixed default parameters becoming increasingly sub-optimal at larger N -- this first look cannot distinguish the two explanations.

Full write-up: docs/wormhole_syk_teleportation.md#experiment-15-n-scaling-check-n8-vs-n12-run-2026-08-07

v2.16.0 - Wormhole qubit-coupling topology check (Experiment 14, honest negative result)

Tests whether the actual qubit-coupling topology of each instance's K=10 SYK quads -- which specific modes get coupled together, not just how many terms commute or how many terms touch a given mode -- predicts the sign-dependent instance variance.

A weighted 8-mode co-occurrence graph is built per instance (edge weight = how many quads couple that pair of modes). An ad hoc check before committing to this design caught a real methodology problem: a binary version of the graph (edge iff any co-occurrence) saturates to the complete graph K8 for most instances, useless as a discriminator -- the weighted count is used instead.

A second honest check, done after computing the real n=100 numbers: two of the four candidate features (max weighted degree, weighted degree std) turn out to be an exact linear rescaling of Experiment 11's mode-usage-count features (weighted degree = 3x usage count, verified to 1 part in 1e15) -- not new information. The two genuinely new features -- n_zero_pairs (how many of the 28 possible mode pairs are never coupled) and algebraic connectivity (the Fiedler value) -- also fail to correlate: r=+0.159 (p=0.114) and r=-0.141 (p=0.163).

A 6th and 7th candidate explanation ruled out, on top of Experiment 11's two, Experiment 12's structurally-trivial phase diagnostic, and Experiment 13's two. Seven independent hypotheses tested against the sign-dependent instance variance from Experiments 10/11; none hold up.

Full write-up: docs/wormhole_syk_teleportation.md#experiment-14-qubit-coupling-topology-check-run-2026-08-07

v2.15.0 - Wormhole mechanistic check: message-mode participation and operator growth rate (Experiment 13, honest negative result)

Tests two more candidate explanations for the sign-dependent instance variance from Experiments 10/11, this time grounded directly in the protocol implementation rather than a generic aggregate statistic, reusing Experiment 11's own n=100 instance set and delta values (no re-screening needed).

Feature A -- message-mode participation: the Jordan-Wigner mapping (dense_evolution.fermions.majorana_pauli_terms) shows Majorana modes 1 and 2 map onto qubit index 0, exactly the qubit the message is injected into and read out from. Counts how many of each instance's K=10 SYK quads touch those two modes specifically.

Feature B -- operator growth rate: reuses Experiment 12's own Gamma_P/Heisenberg-evolution machinery (run_size_winding_check) to get mean operator size at t=0.7 and t=1.2 -- real, non-trivial, instance-varying data Experiment 12 already computed but never correlated against the sign.

Result: neither correlates. Message-mode participation r=-0.012 (p=0.90); operator growth rate at t=1.2 r=+0.126 (p=0.21). Both are far from the conventional 0.05 threshold -- a clean null on both counts, ruling out a 4th and 5th candidate explanation (on top of Experiment 11's two and Experiment 12's structurally-trivial phase diagnostic). Flagged explicitly: this is the 4th/5th candidate tested on the same n=100 sample used in Experiment 11, a real multiple-comparisons risk that would matter more had either come back significant.

Full write-up: docs/wormhole_syk_teleportation.md#experiment-13-mechanistic-check----message-mode-participation--operator-growth-rate-run-2026-08-07

v2.14.0 - Wormhole size winding diagnostic (Experiment 12, honest negative result)

Directly computes arXiv:2604.10090's own 'size winding' diagnostic (Sec. S6, Eqs. S18-S22): expands a Heisenberg-evolved single-sided Majorana operator in the Majorana-string basis and checks the phase coherence R(l)=|q(l)|/P(l) and phase arg(q(l)) of the winding size distribution, across the same 6 SYK instances used in Experiments 8/10, at 4 post-quench times each.

Verified first on 3 individual seeds (spanning correctly- and wrong-signed instances) before running the full sweep. The basis normalization Tr(Gamma_P Gamma_Q^dagger) = 2^|P| * dim * delta_PQ was verified directly via numerical Hermiticity/orthogonality checks, since the PDF-extracted paper formula did not reproduce as literally written (likely a lost exponent).

Result: R(l)=1.0000 and arg(q(l))=0.0000 exactly (to floating-point precision) for every instance and every time tested -- a third theory-motivated diagnostic, after mode-usage-imbalance and the level-spacing r-statistic in Experiment 11, that fails to explain the sign-dependent instance variance from Experiments 10/11. Mean operator size (t) does show genuine chaos-consistent growth followed by finite-size recurrence, confirming real operator-growth dynamics distinct from this null phase-coherence result.

Full write-up: docs/wormhole_syk_teleportation.md#experiment-12-size-winding-run-2026-08-07

v2.13.0 - Wormhole large-sample (n=100) ensemble sign check (Experiment 11)

Large-sample version of Experiment 10's check, matching arXiv:2604.10090's own reported ensemble size. Result: 49/100 (49%) of exact 34/11-selection-matched instances are wrong-signed at the paper's own default parameters -- far stronger than Experiment 10's 2/6, essentially a coin flip, not a generic feature of the ensemble. Two candidate structural explanations (Majorana mode-usage imbalance, spectral level-spacing chaos statistic) tested for correlation -- neither holds up at this sample size (an earlier n=6 look had suggested mode-usage imbalance correlated strongly, r=0.87 -- that does not replicate at n=100, r=0.171, p=0.09). See README Section 21 / docs/wormhole_syk_teleportation.md.

v2.12.0 - Wormhole cross-check vs. arXiv:2604.10090's ensemble-robustness claim (Experiment 10)

Direct comparison against the source paper's own 'Ensemble robustness' section, which claims the sign-dependent asymmetry is 'a generic feature of the ensemble' from 100 disorder realizations. Controlling for a real confound in Experiment 8 (its baseline point was itself tuned on seed=61), re-evaluating all 6 of our 34/11-selection-matched instances at the paper's own stated default parameters (t0=0.3, mu=12, t1=0.60) still leaves 2 of 6 instances wrong-signed -- directly contradicting the 'generic feature' claim for this subset. Seed 2166 is wrong-signed at every evaluation point tested across this whole write-up. See README Section 21 / docs/wormhole_syk_teleportation.md.

v2.11.0 - Wormhole signal vs. realistic depolarizing noise (Experiment 9, honest negative result)

Real Trotterized gate circuit at seed=61's converged optimum (t0=0.70, mu=17.0, t1=0.36), with a stochastic depolarizing Kraus channel injected after each of the protocol's three phases. Honest negative result: the sign-dependent signal decays and crosses zero between p=0.01 and p=0.02 -- already at p=0.01 the mean signal is smaller than its own trial-to-trial standard deviation, statistically indistinguishable from zero at a noise level within range of current NISQ hardware. See README Section 21 / docs/wormhole_syk_teleportation.md.