RESEARCH RELEASE 0.2 / 20 OCTOBER 2026
Protocol draft—not registered or run. Technical lead appointment, numerical acceptance tolerances, apparatus definition and specialist sign-off remain prerequisites.
Decision and scope
First qualify a known-force optical instrument (MF-01A), map disturbances (MF-01B), then compare a measured optical momentum distribution with a measured force difference (MF-01C). Graphene integration follows optical and thermal characterization. Alpha 1 new-field hardware is on hold: no microscopic state-coupling mechanism or experimentally allowed operating point has been established.
Observables and system boundary
An external light source illuminates the sample. All outgoing orders, transmitted light, scatter and thermal emission cross the sample boundary and carry momentum. Supports, cables, beam stops and sources remain in the wider ledger. Any detector or beam stop on the balance must be included in its force prediction. Define the signed full force difference D = mean(F_B) - mean(F_A). For an ideal symmetric reference define amplitude a = D/2. Never compare uncertainty for a to the full difference D. With N independent windows equally allocated to A/B, u(D)=2 sigma/sqrt(N); u(a)=sigma/sqrt(N). Common biases may cancel; state-correlated effects do not automatically cancel.
Design freeze before acquisition
Record serial numbers, sample stack, dimensions, mass, optical alignment, incidence angle, wavelength, polarization, spot profile, intercepted power, gate states, temperature sensors, pressure and gas species, wiring, lever arms, sampling rate, window length, settling rules and calibration chain. Select safe irradiance using measured device limits; no 10 W graphene rating is implied. A 1064 nm passive commissioning source is provisional; the published graphene comparison operates at 7.281 micrometers. A qualified experimental lead must choose components and safety controls.
Prediction and competing explanations
Measure the complete angular scattering distribution in both states; calculate the vector momentum integral and its covariance. Freeze a predicted D, operating envelope and null geometry before unblinding force data. Measure absorption independently or from a closed optical energy budget. Fit matched-heater transfer functions, including phase and frequency dependence, to characterize thermal radiation, mount distortion and center-of-mass drift. Separate electrical patch drift, cable forces, magnetic effects, residual-gas forces and vibration with matched controls.
Run structure
Use balanced randomized A/B blocks with recorded seed; include beam-off electrical switching, fixed electrical state with optical modulation, matched-heater controls, dummy samples, reversal and pressure/temperature sweeps. Conceal condition labels from the primary analyst where feasible. Calibration before and after, independent days and reassembly are required. Choose dwell time from measured transients and drift, not an assumed instrument bandwidth.
Sample-size and uncertainty freeze
Select N from the candidate signal, qualified per-window covariance and a predeclared coverage/power objective. 100 nN independent-window noise, N=128 and a 0.05 nN amplitude systematic standard uncertainty yield a 17.324 nN approximate 95% half-width: insufficient for the 13.343 nN reference amplitude. A 217-window illustration passes only that arithmetic comparison; it is not a power analysis or a qualified test duration. Use measured autocorrelation or block resampling to estimate effective N. Report Allan/deviation-versus-averaging-time diagnostics where appropriate. A systematic floor does not average down.
Analysis and exclusion rules
Lock the measurement model and analysis code hash before decisive data. Preserve raw data. Exclude a window only for a prospectively recorded instrument fault (saturation, loss of beam lock or logging failure); retain a separate record of each exclusion and the all-window sensitivity analysis. Do not reject observations because they disagree with a desired force. Propagate correlated input distributions. Report signed D, standard uncertainty, interval method, coverage, fit residuals and all tested configurations. A 1.96 factor is only an approximate Gaussian interval, not proof of detection.
Decision gates
Advance only if independent force calibration, optical closure and artifact controls are adequate for a predeclared tolerance; signal scaling, reversal and null tests agree; residuals meet the locked model checks; and the result repeats across days and reassembly. An unresolved artifact means HOLD. A null or disagreement remains a reportable result. Independent replication requires another team/instrument and is a separate gate. Nothing here qualifies a flight system.
Illustrative Monte Carlo, not empirical GUM compliance
model/review.py uses 200,000 draws, seed 20261020. Power N(10,0.05) W; order efficiencies jointly normal means (0.078,0.084), standard uncertainties (0.003,0.003), correlation 0.5; angles N(-12,0.2) and N(45,0.2) degrees. These are analyst assumptions, not published measurement uncertainties. Gaussian tails are a local approximation; validate bounded distributions for real inputs. The full B-A residual example uses absorption Uniform(0.85,0.95), effective differential thermal angular coefficient N(0,0.01), and fixture coefficient 2 nN/K times differential temperature N(0,0.02) K. The thermal coefficient and temperature have correlation 0.6. Patch, gas and electrical residuals are respectively Uniform(-0.1,0.1), Uniform(-0.2,0.2), Uniform(-0.05,0.05) nN. Readout standard uncertainty is 2*10/sqrt(128) nN. Distributions not explicitly correlated are independent. None of these allocations authorizes a detection claim; their values must be measured or defensibly bounded. The model omits unresolved changes in other optical orders, so it is not a complete graphene prediction.
Release and ownership
Founder/program sponsor: Christopher Hill. Experimental PI, force-metrology lead and independent reviewer: not appointed. Sponsor acceptance and specialist review remain pending. AI assisted research synthesis, code, calculations and writing using ChatGPT/Codex on 20 October 2026; the exact model identifier was not exposed. No human peer review, NASA submission, affiliation, grant eligibility, DOI deposition or preregistration is claimed. Preserve raw and processed data, calibration, drawings, BOM, software environment, deviations and signed review records in any future release.
Research basis: Joint Committee for Guides in Metrology, 2008 · Joint Committee for Guides in Metrology, 2008.
Download the protocol draft↗Inspect uncertainty calculations↗