Concept study / Tests & research methodsConceptual illustration. Shapes, colors, and motion are not measured data.
Materials can retain signatures of earlier driving, and experimental work on sheared suspensions provides one concrete example. Such findings motivate a careful version of Hill’s question: which present states preserve information about which past inputs, for how long, and under what transformations? They do not establish that an amount of energy carries a universal biography. This proposed benchmark begins with a bounded encoding and decoding task. The immediate objective is to detect a reproducible history-dependent state while ruling out direct leakage through temperature, instrument settings, labels or environmental carryover. More ambitious interpretations require additional experiments and a specified mechanism.
Encode histories under matched conditions
Assign samples or independent runs randomly to two or more training histories. Match total duration and relevant input budgets where feasible, while documenting any differences that cannot be removed. After training, apply the same probe and record the response. Include untrained controls and sham training. A useful first version uses a well-characterized physical simulation or a published benchmark dataset before any new apparatus is built. The point is to validate inference and decoding, not to create an impressive signal whose origin is unknown or whose controls were chosen afterward.
Read out without leaking the answer
Train a decoder on one partition and evaluate it on wholly separate runs. Keep the training labels hidden from preprocessing decisions that might encode them accidentally. Filename order, acquisition date, instrument warm-up and sample identity can all become shortcuts. Compare a history-aware model with a model supplied only the present controlled inputs. Also test a known relaxation or hysteresis model. Above-chance decoding is useful only after showing that it reflects the intended physical state rather than an overlooked administrative or environmental record of the experimental assignment.
Measure retention and erasure
Repeat the readout after several declared delays, perturbations and attempted resets. Plot decoding performance with uncertainty against delay and transformation severity. A reset that erases the signal helps identify its substrate; a reset that preserves it motivates further controls. Neither outcome proves a cosmic information field. If matched energy states retain different histories, the result points toward additional state variables rather than energy alone. The proposed publication should name the material, protocol, decoder and retention conditions, and state exactly which transformations were tested and which remain outside the experiment.
The mathematical lens
Evaluate both on independent runs. Chance accuracy depends on label balance and the chosen metric.
Turn it into a question.
- Question
- Can a later physical readout recover an assigned earlier history?
- Method
- Randomize training, hold out complete runs, test leakage and measure retention after specified resets.
- What would count against it?
- Performance vanishing under leakage controls or independent runs defeats the claimed benchmark result.
A memory result belongs to its encoding, substrate and readout conditions.
Follow the original work.
These sources support the mechanisms and limits discussed here. They do not endorse Hill’s framework. Notes distinguish research proposals, reviews, and observations.
Continue the investigation
Edition 02 · 18 October 2026 · AI-assisted synthesis. See editorial method.