{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","cell_id":"predictive_residual_processing__chemistry_materials","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"hierarchical_residual_atomistic_trajectory_archive","proposal_index":2,"version":0,"title":"Hierarchical Residual Archive for Atomistic Materials Trajectories","problem":"An atomistic materials simulation can generate a long sequence of full coordinate, velocity, force, and cell-state frames in which predictable lattice motion, thermal vibration, and imposed deformation dominate the stored data. Storage and scientist-review capacity is spent repeatedly representing that expected motion, while localized defect migration, bond rearrangement, nucleation, or an unmodeled collective mode can remain difficult to locate. Simple frame thinning or selected order parameters can reduce volume but can also remove the spatial and temporal evidence needed to reconstruct an unexpected event.","actors":["Materials simulation scientist","Computational chemistry method owner","High-performance computing operator","Trajectory encoder running beside the simulation output process","Trajectory decoder and analysis environment","Independent scientific-integrity reviewer"],"observable_state":"At every selected simulation step, the encoder receives the complete atom identities, unwrapped coordinates, velocities, forces, periodic cell, neighbor relationships, energy diagnostics, timestep, integrator state, and simulation provenance. It freezes a predicted next frame, computes signed residuals in a declared coordinate convention, records uncertainty and model version, and reports whether each residual was encoded, escalated, audited, or forced through the full-state path. The decoder records reconstructed-frame error and synchronization status.","consequence":"Either the archive retains repeated expected motion at substantial storage and analysis cost, or aggressive thinning and feature extraction make an unexpected atomic rearrangement hard to reconstruct, inspect, or distinguish from predictor error. A corrupted or desynchronized residual sequence can additionally yield plausible-looking but scientifically invalid frames.","affected_objective":"Maintain reconstructable, scientifically interpretable atomistic trajectories within a declared storage and analysis-attention budget while preserving localized transformations, topology changes, conservation failures, and evidence that challenges the trajectory predictor.","intervention":"Place a versioned hierarchical predictor between the simulation output process and its trajectory archive. From the last accepted anchor, integrator state, periodic cell evolution, imposed deformation, and local atomic environments, predict the next full coordinate, velocity, and force frame before accepting the corresponding simulated frame. Compute signed atom-level residuals after deterministic atom matching and periodic-boundary unwrapping. Let local predictors explain routine vibration and affine motion; propagate unresolved, precision-weighted residuals to neighborhood and global levels so spatially coherent defect or phase-change signatures remain visible. Store quantized residuals, predictor checksum, uncertainty, and reconstruction metadata instead of every full frame when fidelity conditions hold. Preserve scheduled full anchors plus random and risk-stratified raw frames, and force complete frames for topology changes, neighbor discontinuities, conservation violations, model staleness, version mismatch, excessive cumulative error, or residual structures outside the predictor's validated scope. Use reviewed residual clusters to propose slower predictor revisions, but keep simulation execution and archive-policy changes under human authority.","structural_mapping":[{"archetype_element":"Prediction target definition","domain_realization":"The target is the complete next archived atomistic frame—unwrapped coordinates, velocities, forces, periodic cell, and declared diagnostics—at one specified timestep horizon and fidelity tolerance."},{"archetype_element":"Generative model state and bounded scope","domain_realization":"A versioned predictor combines the previous anchor, integrator state, cell deformation, atom identities, and local-environment models; its authorization is limited to specified material phases, thermodynamic ranges, simulation code, and timestep."},{"archetype_element":"Feedforward expectation before observation","domain_realization":"The predicted frame is frozen from prior state before the simulation's actual output frame is compared with it."},{"archetype_element":"Signed structured comparator","domain_realization":"After deterministic atom matching and periodic-boundary unwrapping, the encoder computes vector coordinate, velocity, force, and cell residuals while separately marking discrete topology and neighbor changes."},{"archetype_element":"Hierarchical residual propagation","domain_realization":"Routine atom-level motion is accounted for locally, while unexplained spatially coherent residuals are aggregated into neighborhood and system-level messages that can identify a migrating defect or collective transformation."},{"archetype_element":"Precision and consequence weighting","domain_realization":"Residual priority depends on numerical uncertainty, atom and region reliability, spatial coherence, persistence, reconstruction impact, and protected scientific-integrity classes rather than displacement magnitude alone."},{"archetype_element":"Reconstructive residual channel","domain_realization":"The archive contains the correction and compatible predictor identity needed for the decoder to rebuild each frame as predicted state plus residual."},{"archetype_element":"Residual-driven learning","domain_realization":"Reviewed residual clusters enter a replay set used to test or propose changes to local-motion models, thresholds, or declared scope; model revision is slower and separately governed from archival encoding."},{"archetype_element":"Synchronization and validity controls","domain_realization":"Checksums, ordered frame identifiers, validity clocks, and complete anchor frames prevent residuals from being decoded against an incompatible or stale trajectory state."},{"archetype_element":"Independent raw audit and fallback","domain_realization":"Random complete frames and risk-stratified windows are retained outside the residual selection logic, and validity failures restore full-frame archival for the affected interval."},{"archetype_element":"Explicit capacity and error budget","domain_realization":"Storage saved by residual encoding is evaluated together with predictor computation, metadata, anchors, audits, replay, fallback frames, and scientist review under component-wise and cumulative reconstruction limits."}],"mechanism_mapping":[{"mechanism_slug":"hierarchical_prediction_error_loop","role":"Organizes atom, neighborhood, and system-level predictors so expected local motion is absorbed at the lowest valid level while unresolved coherent error rises for analysis.","counterfactual_removal":"Without the hierarchy, the archive could encode temporal deltas but would not concentrate attention on multiscale defect and phase-transition structure."},{"mechanism_slug":"predictive_codec","role":"Maintains compatible encoder and decoder predictors and stores quantized trajectory residuals plus reconstruction metadata.","counterfactual_removal":"Without the codec, residuals would be anomaly indicators rather than a representation from which complete trajectory frames can be rebuilt."},{"mechanism_slug":"model_version_checksum_handshake","role":"Gates every residual on the exact motion model, preprocessing convention, atom ordering, and parameters used by the encoder.","counterfactual_removal":"Without the handshake, a residual could be applied to a different atom mapping or predictor and silently create a plausible but false atomic configuration."},{"mechanism_slug":"precision_weighted_error_gate","role":"Allocates storage and review priority using residual magnitude, uncertainty, coherence, consequence class, and encoding cost while logging suppressed residual mass.","counterfactual_removal":"Without precision weighting, ordinary high-amplitude thermal motion could crowd out smaller coherent signals of defect motion or incipient transformation."},{"mechanism_slug":"anomaly_detection_model","role":"Scores local and collective residual patterns against the predictor's validated envelope and identifies candidate events or model failures for review.","counterfactual_removal":"Without residual-pattern screening, the codec could reconstruct within average tolerance while leaving informative localized departures difficult to find."},{"mechanism_slug":"residual_comparison_test","role":"Checks residual sequences for bias, autocorrelation, spatial correlation, regime dependence, and differences from a simpler kinematic predictor and raw audit frames.","counterfactual_removal":"Without comparison testing, structured model error could be treated as random thermal noise and repeatedly suppressed."},{"mechanism_slug":"periodic_full_state_resynchronization","role":"Stores complete anchor frames on a fixed and triggerable cadence, limiting the duration of loss, quantization, or decoder-state divergence.","counterfactual_removal":"Without full anchors, one missing or incorrectly decoded correction could contaminate the remainder of a trajectory."},{"mechanism_slug":"shadow_raw_channel_sampling","role":"Retains random and risk-stratified complete frames independently of the predictor and measures what the residual archive would have reconstructed.","counterfactual_removal":"Without independent raw frames, the residual system would evaluate omissions only through the representation choices it made itself."},{"mechanism_slug":"model_drift_monitoring","role":"Monitors changes in residual distributions, reconstruction calibration, thermodynamic regime, local environments, and model age.","counterfactual_removal":"Without drift monitoring, a changing phase or simulation regime could remain encoded against a model whose scope had silently expired."},{"mechanism_slug":"raw_signal_fallback_switch","role":"Forces full-frame archival when synchronization, reconstruction, topology, conservation, validity, or protected-integrity conditions fail.","counterfactual_removal":"Without fallback, the archive would continue suppressing full state precisely when reconstruction was least defensible."},{"mechanism_slug":"prediction_error_replay_buffer","role":"Stores residual clusters with surrounding full context and provenance for event analysis, regression tests, and separately approved predictor revision.","counterfactual_removal":"Without replay, transient predictor failures and atomic transformations could not be examined consistently across model versions."}],"causal_chain":["Full trajectory frames repeatedly encode atomic motion that the integrator and local-environment model can already predict.","A scoped, versioned hierarchy predicts the next atomistic frame from the last synchronized state before the actual frame is accepted.","The encoder compares the complete simulated frame with that expectation and preserves signed atom-, neighborhood-, and cell-level residuals.","Precision weighting separates numerical or thermal variation from reliable, coherent, or protected departures without treating amplitude as the sole criterion.","Compatible decoding reconstructs routine frames from prediction plus correction, reducing repeated representation of expected motion.","Unresolved residual clusters rise through the hierarchy and direct scientist attention toward localized defects, collective transformations, or model misspecification.","Independent raw samples, residual-structure tests, checksums, and complete anchors expose suppressed information and bound accumulated reconstruction error.","Validity, topology, or scientific-integrity triggers restore full-frame archival, while reviewed misses can support a separately approved predictor update."],"baseline":"Archive complete atom identities, coordinates, velocities, forces, cell state, and diagnostics at a fixed output cadence; optionally apply ordinary lossless file compression; then locate events through manual trajectory inspection or predefined global order parameters. The comparison must include storage, encoding and decoding computation, audit frames, anchors, fallback volume, reconstruction checking, and scientist-review effort.","nearest_rivals":["Ordinary lossless trajectory compression, which reduces byte redundancy without maintaining an explicit scientific predictor whose residuals also drive event review and model correction.","Simple coordinate differencing, which stores change from the previous frame but need not predict from dynamics, represent uncertainty, synchronize model versions, or audit what is suppressed.","Temporal subsampling, which retains fewer complete frames but can omit short-lived rearrangements between stored frames.","Selected collective variables or order parameters, which provide compact summaries but generally cannot reconstruct an unanticipated atomic configuration.","Standalone trajectory anomaly detection, which flags unusual frames without making prediction plus residual the archival representation.","Adaptive simulation or rare-event sampling, which changes where computation is spent rather than governing how every produced trajectory frame is represented and reconstructed."],"remaining_contrastive_claim":"This proposal makes a synchronized multiscale dynamics prediction the implicit portion of an atomistic archive and stores the signed correction as both reconstructive code and teaching signal. Its defining controls are hierarchical error propagation, compatible decoding, independent raw frames, full anchors, and forced decompression. It is therefore narrower than trajectory thinning, order-parameter monitoring, generic compression, anomaly detection, or adaptive simulation.","authority_safety":{"decision_authority":"The materials simulation scientist owns the scientific interpretation and decides whether a residual cluster represents a physical event, numerical artifact, or model-boundary failure. The computational-method owner approves predictors and fidelity criteria. The computing operator may force full archival for integrity or recovery. The encoder may route data and request fallback but may not alter the simulation Hamiltonian, integrator, timestep, boundary conditions, or publication record.","authorized_first_step":"Run the proposed archive in shadow mode on one bounded atomistic materials trajectory while retaining every complete frame as the authoritative record; prohibit the residual system from changing the simulation or deleting raw output.","excluded_actions":["Changing forces, potentials, timesteps, thermostats, constraints, or simulation control based on residuals during the first evidence step","Deleting or replacing authoritative full frames during shadow evaluation","Treating reconstructed frames as ground truth before comparison with retained full frames","Suppressing topology changes, neighbor discontinuities, conservation failures, nonfinite values, or provenance gaps","Activating a new predictor, threshold table, atom-matching rule, or quantizer without recorded approval","Extending the model to an unvalidated material phase, thermodynamic range, code version, or timestep","Using absence of a residual as evidence that a simulation frame was produced or decoded successfully"],"halt_rollback":"Stop residual encoding for the affected interval and preserve complete frames if a checksum or frame sequence fails, atom matching is ambiguous, reconstruction crosses a component-wise or cumulative budget, residuals develop sustained spatial or temporal structure, topology changes, protected diagnostics fail, the simulation leaves model scope, or a raw audit reveals an omitted event. Reconstruct again from the last verified full anchor, retain the failed encoder state for diagnosis, and require scientist and method-owner approval before resuming with the last accepted model version."},"negative_tests":{"strongest_counterevidence":"An independent full-frame review finds a localized rearrangement, transient defect, topology change, or collective mode that the residual hierarchy suppressed or reconstructed ambiguously while all production validity indicators remained nominal.","problem_falsifier":"Complete trajectories at the required cadence fit within the declared storage and review bounds, and scientists can reliably locate decision-relevant events without thinning or lossy summaries; there would then be no binding representation or attention problem for this intervention to solve.","intervention_falsifier":"When held to the predeclared component-wise fidelity, event preservation, provenance, and fallback requirements, the residual archive plus predictor computation, anchors, audits, and review costs no less than full-frame archival, or it fails to reproduce the scientific classifications supported by the complete trajectory.","risks":["A predictor trained on stable lattice motion may explain away the earliest signal of a phase transition.","Periodic-boundary unwrapping or atom correspondence errors may appear as physical residuals or conceal real motion.","Quantization errors can accumulate between anchors and distort long-timescale diffusion or vibration statistics.","A hierarchy can duplicate, attenuate, or misattribute the same event across atom, neighborhood, and global levels.","Thresholds based on familiar defects can systematically discount an unanticipated rearrangement.","The encoder and decoder may agree on the same misspecified model, so a valid checksum does not establish physical correctness.","Random raw sampling may miss a short rare event, while risk-stratified sampling may reproduce known blind spots.","Residual-weighted replay can overrepresent spectacular events and underrepresent routine conditions needed for calibration.","Fallback frequency may erase the storage benefit in highly anharmonic, reactive, or nonstationary regimes.","A reconstructed archive may be mistaken for an untouched simulation record unless provenance remains explicit." ]},"next_evidence_step":"Pre-register a shadow evaluation for one simulation code and version, one material system, one thermodynamic regime, one timestep, one trajectory duration, and one archive cadence. Retain every complete frame. Freeze the hierarchical predictor, atom-matching convention, quantizer, thresholds, anchor cadence, and fidelity budgets before evaluation. Include ordinary trajectory intervals and copied-frame challenge windows containing controlled coordinate displacements, defect translations, affine cell changes, brief collective modes, atom-order permutations, missing residuals, checksum mismatches, and protected topology or conservation flags; do not feed challenge copies back into the simulation. Decode the residual archive from anchors and compare it frame-by-frame with full truth. Have a blinded simulation scientist classify predeclared physical and numerical events from the full and reconstructed paths. Measure total stored representation, reconstruction error by state component and regime, cumulative drift between anchors, protected-event capture, audit disagreement, residual structure, review burden, and correct fallback behavior. Reject operational use if any protected challenge does not force a complete frame, any decision-relevant event becomes unreconstructable or differently classified, any sequence failure propagates beyond its permitted anchor interval, or the governed residual path does not fit the same resource bounds better than the full-frame baseline.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 concerns live Raman observations from a physical reaction experiment and uses matched spectral predictors to reduce remote transmission and chemist-monitoring traffic while preserving safety instrumentation. This proposal instead concerns archival representation and scientific interrogation of computational atomistic trajectories. Its predicted object is a multiscale atomic state rather than an experimental spectrum; its principal actors are simulation scientists, method owners, and computing operators rather than bench and remote chemists; its scarce resources are trajectory storage and analysis attention rather than a laboratory communication channel; and its causal path uses integrator-conditioned atom, neighborhood, and system-level residuals to expose defect motion or collective transformations. It can be adopted for simulation archives without installing Raman instrumentation or changing reaction monitoring, and proposal 1 can be adopted without running or encoding an atomistic simulation. It is therefore an independent opportunity rather than a feature or implementation variant of proposal 1.","revision_record":{"parent_version":null,"progress_targets_addressed":["Generate one additional complete candidate at proposal index 2","Address a materially different chemistry-and-materials problem","Use a distinct hierarchical atomistic archival intervention and causal path","Explain diversity from the earlier sealed proposal","Preserve authority, safeguards, falsifiers, and bounded first evidence"],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}