IBM Phase 2 B-C Scaling Manifest — 2026-05-05¶
This manifest preregisters the next continuation run after the completed reduced A+G replication. It is limited to scaling blocks B and C only.
Status¶
| Field | Value |
|---|---|
| Status | Preregistered; not submitted at document time |
| Backend | ibm_kingston |
| Approved blocks | B (n=6) and C (n=8) |
| Deferred blocks | A, D, E, F, G |
| New readout baseline | none; reuse completed A+G same-day n=6,8 readout controls |
| Dry-run command | PYTHONDONTWRITEBYTECODE=1 /home/anulum/.local/bin/python scripts/phase2_full_campaign_ibm.py --dry-run --backend ibm_kingston --skip A D E F G |
| Live command | PYTHONDONTWRITEBYTECODE=1 /home/anulum/.local/bin/python scripts/phase2_full_campaign_ibm.py --confirm-promo-active --backend ibm_kingston --skip A D E F G --max-live-depth 900 |
Scientific purpose¶
This run tests whether the DLA parity asymmetry observed and replicated at
n=4 survives at n=6 and n=8.
It can validate:
- early system-size scaling of the parity-sector asymmetry;
- whether the middle-depth sign remains positive beyond
n=4; - whether the depth window shifts or narrows with system size.
It can falsify:
- a simple scalable-parity interpretation if
n=6andn=8lose the sign; - a monotone scaling story if the effect is present at one size but absent at the other;
- any claim that Phase 2 already established
n=6-12scaling.
Circuit inventory¶
| Block | n | Depths | Sectors | Reps | Shots | Circuits |
|---|---|---|---|---|---|---|
| B | 6 | 4,8,14,20 |
even, odd | 20 | 4096 | 160 |
| C | 8 | 4,8,14,20 |
even, odd | 15 | 4096 | 120 |
| Total | 6,8 | preregistered above | even, odd | mixed | 4096 | 280 |
Dry-run result on 2026-05-05:
| Metric | Value |
|---|---|
| Circuits | 280 |
| Simulator ISA depth min | 65 |
| Simulator ISA depth max | 269 |
| Simulator ISA depth mean | 160.1 |
| Simulator gate-count min | 198 |
| Simulator gate-count max | 1396 |
| Simulator gate-count mean | 670.7 |
The live command must pass the pre-submit backend-transpilation gate before any
IBM job is submitted. The live max-depth guard is 900; if exceeded, the script
must abort without submission.
Promotion rules¶
The run is not promoted until all gates below pass:
| Gate | Requirement |
|---|---|
| Raw counts | 280 B-C circuits with 4096 shots each. |
| Job binding | Every circuit is bound to an IBM job identifier. |
| Reproducer | scripts/analyse_phase2_scaling_bc.py recomputes all published values from raw counts. |
| Integrity | Raw JSON and summary JSON have SHA-256 hashes recorded in data/phase2_scaling_bc/README.md. |
| Ledger | Hardware ledger names the exact artefacts and scope. |
| Scope | Promote only n=6,8 scaling evidence; no D-E, GUESS, broad advantage, frontier, or multi-QPU claim. |
Expected interpretation¶
Positive result: same-sign asymmetry at n=6 and/or n=8, especially at middle
depths, supports early scaling beyond n=4.
Null result: constrains the claim to n=4 replication and marks larger-size
scaling as unestablished.
Mixed result: identifies a size/depth window and should be reported as such.