Experiment v0.1 · analysis & integrity · locked
The easiest way to fake a discovery is to choose the analysis after seeing the data, so we chose it first.
Latency has a fixed threshold and a fixed dwell window. An active node has a fixed threshold and a fixed dwell. Exclusions have a closed list of allowed reasons, and 'it performed badly' is not one of them. The condition map stays sealed until the numbers and the exclusion list are frozen.
Status
analysis plan locked · no data existsANALYSIS PLAN LOCKED · NO HARDWARE DATA EXISTS · PHYSICAL EVIDENCE NONE
Results must not be able to depend on post-hoc smoothing or window selection. Each rule below is deterministic: given the same raw run file, any two people running the frozen analysis must obtain identical numbers to the last digit.
No moving goalposts · the pipeline
analysis plan locked · no data existsData flows one way. Once a stage is passed it cannot be re-entered, and no parameter downstream of the raw file is ever chosen while looking at the output.
- 01RAW DATA
Every field of the per-run record, written once, hashed, never edited.
- 02FROZEN ANALYSIS
A1 → A10 applied as code. No parameter is chosen while looking at the output.
- 03one-wayBLINDED SCORE
Metrics and the exclusion list computed against condition codes only, then frozen.
- 04one-wayUNBLIND
The sealed condition map is opened. Timestamped, one-way.
- 05one-wayVERDICT
The verdict table below is read off mechanically. There is no discretionary step.
Required raw per-run record
A run that does not carry every field below is not a run. It is written once, hashed, and never edited.
| Field | Type | What it is for |
|---|---|---|
| run_id | string | Unique, assigned before the run begins. |
| blinded_condition | code | Condition code only. The map is sealed. |
| topology_id | string | Which of the ≥ 3 reconfigured topologies at this N. |
| N | int | 16, 32 or 64. |
| topology_seed_hash | hex | Seed and hash of the frozen substrate graph. |
| endpoints_a_b | ids | Source and destination node IDs for this trip. |
| trigger_time | timestamp | Origin for every latency measurement. |
| trigger_amplitude | float | Injected seed amplitude, for the recruitment curve. |
| controller_events | array | Every controller read AND write, with node IDs, scope and timestamps. |
| node_state_traces | array | Per node, over time: chosen observables — amplitude, phase, local c and r. |
| payload_input | vector | The transmitted structured waveform / test vector. |
| payload_output | vector | The decoded waveform at B. |
| power_traces | array | Current, voltage and power for the oscillator rail and the control rail, separately. |
| fault_injections | array | Edge, node or address corruptions, with timestamps. |
| restoration_time | duration | Measured by the frozen restoration rule below. |
| invalid_flags | array | Invalid-run flags with explicit machine-readable reasons. |
Frozen analysis rules · A1 → A10
Deterministic by construction: given the same raw run file, two analysts must obtain identical numbers.
T is the FIRST destination timestamp at which the payload match exceeds the pre-registered fidelity threshold AND remains above it for the fixed dwell window. Not the peak, not the best window, not an eyeballed onset.
Threshold and dwell window are numbers fixed after P0 calibration and before P4. No manual window selection at any point.
A node is ACTIVE when its local C-state or coupling observable crosses the fixed threshold for at least the minimum dwell. From that indicator: peak N_active, integrated active-node-time, and f_active = max_t(N_active) / N.
One threshold and one dwell for every condition and every N. No per-run tuning.
The pre-registered normalized correlation / reconstruction error against the transmitted structured waveform. ≥ 99% is required for a primary pass, and the FULL distribution is published, not the mean.
Metric definition and its noise calibration frozen before unblinding.
All local states and couplings return within the fixed tolerance of the pre-trigger baseline and stay there for the fixed dwell, with NO global reset event anywhere in the log.
Tolerance, dwell and the restoration window length are fixed in advance.
An automated pass over controller_events fails ANY run containing a read or write outside that node's allowed local neighbourhood and address inputs. The global positive control is exempt from the rule but is permanently labelled CONTROL and can never be reported as a C* result.
The audit is code, run before scoring, not a human judgement call.
Integrate the measured electrical power for the oscillators AND for the coupling / control hardware across the full run. Report BOTH gross energy and baseline-subtracted energy.
Controller cost is never folded away, never amortised, never omitted, however unfavourable.
Fit R_T = T_C* / T_C0 and f_active across N = 16, 32, 64 with the same estimator. The primary directional hypothesis requires BOTH slopes negative.
Estimator, confidence intervals and the bootstrap procedure (resamples, statistic, seed) are locked before any data arrives.
The analysis receives condition codes only. The condition map is revealed after the metrics AND the exclusion list are frozen.
Unblinding is a one-way, timestamped event. Nothing in the analysis may change after it.
Permitted only for pre-registered instrumentation failure, saturation or clipping, missing samples, clock-sync loss, or an explicit locality-rule violation.
'Bad performance' is NEVER an exclusion criterion. A slow, noisy or failing run is a result.
Publish ALL valid runs, plus every excluded run ID with its machine-readable reason.
The excluded list is published even when it is embarrassing, and especially then.
Verdict logic · read off mechanically
- — Local C* beats C0 on latency, with R_T decreasing across N
- — f_active decreases across N
- — Payload fidelity ≥ 99%
- — False-trigger rate ≤ 1%
- — Restoration passes with no global reset
- — Zero illicit global controller writes
- — The effect beats the random-local-switching control beyond the pre-registered uncertainty
- — Faster than C0, but the sparsity or locality signature fails, or the behaviour matches the global / random switching controls
- — Sparsity signature holds, but there is no latency improvement over C0
- — The primary joint signature is not reproduced
NO HARDWARE DATA EXISTS. Nothing on this page reports a measurement, and no bench outcome — including the most favourable one — would license any statement about spacetime, distance, or faster-than-light transport. Physical evidence for Adjacency Theory remains NONE.
Take the checklist with you
The full record schema, the ten frozen rules, the pipeline order and the verdict table, as a plain-text checklist you can carry to a bench. It documents an analysis plan. It contains no data, because none exists.
ANALYSIS PLAN LOCKED · NO HARDWARE DATA EXISTS · PHYSICAL EVIDENCE NONE