Hidden relational structure, recovered from shot statistics on IBM Heron r2 · ibm_fez. Confirmation batch: 7/7 above chance, mean cluster recovery 52.04%.
Three boxes. Same numbers, public page.
For 3 groups, chance is 33.33%. Cluster scores sit in the ~42–59% range. The grouping is visible in the data.
Confirmation batch: 7/7 seeds above chance. Mean cluster recovery 52.04% (95% CI 48.04–56.04%).
Shallower circuits, more shots, decoder work, and newer Heron-class chips expand how fully the structure appears on device.
Chance baseline: for 3 groups, a fair cluster score on blind labels is about 33%. Our cluster scores are usually in the ~42–59% range.
Same experiment family: hide a relational pattern (who is linked to whom, in groups) in a quantum circuit, run many shots, then recover that pattern from the measurement statistics.
| Run | Setup | Typical cluster score* | Typical link (edge) F1* | Transpiled depth (contexts 0–2) |
|---|---|---|---|---|
| A | 32 qubits, 8192 shots per context, 3 separate Fez jobs | ~53% / 59% / 50% | ~0.23 / 0.22 / 0.19 | ~282 / 235 / 334 |
| B (tuned) | 24 qubits, 16384 shots per context, shallower circuits | ~50% / 42% / 54% | ~0.31 / 0.21 / 0.25 | ~155 / 236 / 222 |
| C (confirmation batch) | 24 qubits, 8192 shots, observed fraction 0.6, 7 seeds (independent jobs) | mean 52.04% (95% CI 48.04–56.04%) | n/a (partial-observation mode) | ~389–461 |
* Cluster score = match of recovered groups to hidden groups (after fair label alignment). Edge F1 = balance of finding true links vs extra links (median threshold on correlations).
Takeaway: Run C is the quality check: 7/7 seeds above chance (chance = 33.33% for 3 groups), mean 52.04%. Repeatable structure recovery on IBM Heron r2 · ibm_fez.
Other Fez hardware runs. A Kochen–Specker (Mermin–Peres) contextuality test also ran on Fez: 18 circuits, 2 qubits. With QPC reducer: row +1, col −1 (ideal). Raw: col3 −0.72. Task framing: Quantum contextuality task. Noise Reducer.
They belong to the same QPC family (information in how the circuit is built and what you measure). They ask different questions:
| Holographic memory line | QPC-SWR v1 (this test) |
|---|---|
| Can we store a pattern in a spread-out way and still reconstruct it from partial or redundant readout? | Can we hide a relationship map (groups + links) and see it again in pairwise statistics from many samples? |
| Success read as reconstruction accuracy, majority vs single-qubit, etc. | Success read as structure recovery (clusters + edges). |
What SWR adds: it targets relational / many-body style structure in correlations, a complementary lens to pattern reconstruction. Both lines sit on the public holographic family pages.
Site-wide statement. We report the simulator or intended behavior where we have it, and the measured processor outcomes on IBM Heron r2. See also the Holographic Memory report and Quantum-native task pages.
ibm_fez).On IBM Heron r2 · ibm_fez, QPC structure-recovery tests show a repeatable hardware signal. In the confirmation batch, all 7/7 runs were above chance, with mean recovery 52.04% (95% CI 48.04–56.04%). Hidden relational structure is recovered from measurement statistics on the chip.