{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp12_substrate_denial72_20260805","cell_id":"authority_mentor_relationship_anchoring__computer_science","arm":"CONSTRAINED_MAX","candidate_id":"authority_mentor_relationship_anchoring__computer_science__CONSTRAINED_MAX","decision":"PROPOSAL","abstention_reason":null,"proposal":{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp12_substrate_denial72_20260805","cell_id":"authority_mentor_relationship_anchoring__computer_science","arm":"CONSTRAINED_MAX","candidate_id":"authority_mentor_relationship_anchoring__computer_science__CONSTRAINED_MAX","proposal_index":1,"version":0,"title":"Direct Reference-to-Learner Conductance Transfer for Analog Compute Tiles","problem":"A writable resistive-memory compute tile intended to replace a qualified tile can realize a different current-transfer matrix despite receiving the same nominal programming targets, because cell conductance, contact resistance, and voltage dependence are physical properties of each device. The concrete problem is to reproduce the qualified tile's bounded analog input-output behavior in a spare without relying on a digitized device model or runtime correction software.","actors":["Read-isolated, electrically characterized reference crossbar tile","Writable resistive-memory learner tile","Independent secondary reference tile","Current-limited differential coupling fixture","Device owner and laboratory electrical-safety custodian"],"observable_state":"When identical row-voltage probes are applied, corresponding reference and learner branches produce unequal currents. The mismatch appears directly as a signed bridge voltage across specified operating-voltage and temperature points, and the two tiles consequently produce different column-current vectors for the same physical inputs.","consequence":"The spare implements a different analog matrix and cannot serve as an electrically substitutable replacement within the specified acceptance envelope; downstream numeric outputs can therefore change unless another correction mechanism intervenes.","affected_objective":"Create a physically independent spare whose analog transfer behavior conforms to a bounded reference envelope after the coupling fixture is completely disconnected, while leaving the reference tile unchanged.","intervention":"Place corresponding reference and learner elements in a read-isolated differential fixture and expose both to the same analog excitation. A sign-sensitive transistor bridge converts each instantaneous current mismatch into a bounded programming pulse applied only to the learner's resistive-memory element. The pulse polarity drives ion migration in the direction that reduces the mismatch; its magnitude falls as the bridge approaches null. Repeat the physical exposure across the specified voltage envelope, open the learner's write path, unplug the fixture, and challenge the learner against an independently characterized second reference. No ADC, digital target calculation, firmware, model, or human interpretation closes the adjustment loop.","structural_mapping":[{"archetype_element":"Legitimate Mentor Anchor","domain_realization":"The qualified reference tile physically embodies the accepted transfer behavior. Its authority is limited to the voltage, temperature, and aging envelope in which it was independently characterized, rather than being inferred from label or provenance alone."},{"archetype_element":"Mentee Readiness and Consent Boundary","domain_realization":"Only a writable learner coupon with compatible geometry, safe leakage, and an intentionally installed physical write-enable shunt can enter the coupling path. Removing the shunt makes further material modification impossible."},{"archetype_element":"Relational Safety Container","domain_realization":"High-impedance reference buffers, programming-current clamps, thermal cutoffs, and polarity limits prevent the coupling relationship from writing the reference or overstressing the learner."},{"archetype_element":"Cultural Norm and Value Payload","domain_realization":"The payload is the reference tile's physically embodied conductance and current-voltage response over the declared operating envelope, including behavior not captured by a single nominal conductance code."},{"archetype_element":"Modeled Practice and Judgment Window","domain_realization":"Reference and learner receive the same real electrical stimuli side by side, exposing how the qualified tile responds at each operating point instead of presenting an abstract target table."},{"archetype_element":"Dialogic Interpretation Loop","domain_realization":"The differential bridge responds to the current states of both devices: mismatch generates a signed learner-only write pulse, the learner's material state changes, and the next excitation produces a revised mismatch. This state-dependent exchange continues until the analog deadband is reached."},{"archetype_element":"Autonomy and Exit Safeguard","domain_realization":"A physical disconnect and write-disable shunt terminate all influence from the reference. The learner must then retain and execute the transferred response using its own nonvolatile material state."},{"archetype_element":"Mentor Selection and Matching Criteria","domain_realization":"A reference is eligible only if its package, array topology, operating range, and independent electrical characterization match the learner's intended role."},{"archetype_element":"Secondary Reference Anchor","domain_realization":"A second independently characterized tile, excluded from the adaptation loop, tests whether the learner copied an idiosyncratic defect of the first reference."},{"archetype_element":"Progressive Autonomy Release","domain_realization":"Correction current decreases intrinsically as bridge imbalance shrinks; uncoupled read trials are then inserted before the write path is permanently opened and the fixture removed."}],"mechanism_mapping":[{"mechanism_slug":"guided_shadowing_with_debrief","role":"Identical physical excitation lets the learner shadow the reference, while the immediate analog bridge imbalance exposes the relevant difference without digitization.","counterfactual_removal":"Without paired exposure and differential comparison, the learner receives no local evidence of the reference's actual response and can only follow nominal or separately represented targets."},{"mechanism_slug":"joint_practice_with_corrective_feedback","role":"Both tiles respond to the same stimulus, and the signed mismatch energy directly produces a learner-only material correction.","counterfactual_removal":"If the corrective coupling is removed, co-exposure merely observes a difference; no conductance state is transferred and the learner remains unchanged."},{"mechanism_slug":"reflective_norm_dialogue","role":"Alternating excitation polarity and multiple voltage points repeatedly tests each correction against the response envelope, allowing subsequent mismatch to reverse or reduce the next physical update.","counterfactual_removal":"A one-point, one-direction exposure could null a single current while leaving polarity asymmetry or nonlinear behavior unconstrained."},{"mechanism_slug":"mentor_rotation_or_second_opinion_channel","role":"The independently held-out reference challenges the learned state after uncoupling and distinguishes shared acceptable behavior from first-reference idiosyncrasy.","counterfactual_removal":"Acceptance against the training reference alone cannot reveal faithful copying of that reference's private defect or drift."}],"causal_chain":["Fabrication variation or drift leaves the learner's resistive elements with an input-output response different from the qualified reference.","The fixture applies the same analog voltage to each matched reference-learner pair while electrically isolating the reference from write-level fields.","Kirchhoff-law current imbalance drives a differential transistor stage, producing a signed electrical signal without estimating a model or consulting stored targets.","The signal gates a current- and duration-limited programming pulse into the learner element only.","The pulse's electric field moves mobile ions in the learner, changing its nonvolatile conductance in the direction selected by the mismatch sign.","The changed learner state alters the next bridge comparison; as the responses converge, bridge imbalance and correction energy diminish physically.","Opening the write shunt and disconnecting the fixture removes the reference relationship completely.","If the material state persists, the learner independently reproduces the bounded transfer behavior; disagreement with the held-out reference reveals overcopying or failed transfer."],"baseline":"The primary comparison baseline is conventional write-verify calibration: measure learner cells, represent desired conductances as digital targets, calculate programming updates, and retain a calibration table or firmware correction. An additional physical baseline trims the learner against fixed precision resistors using the same programming-energy budget.","nearest_rivals":["Digital write-verify using ADC measurements and computed per-cell target updates","Runtime ADC plus firmware lookup-table compensation for analog error","Closed-loop physical trimming against precision resistor and voltage standards rather than another compute tile","Thermal stabilization, fabrication binning, or selection of an already matched spare","Redundant analog columns with downstream averaging or error correction"],"remaining_contrastive_claim":"The defensible claim is mechanistic and conditional, not a claim of novelty or superiority: this candidate uses a qualified device's contemporaneous analog response as the local physical setpoint and converts pairwise mismatch energy directly into nonvolatile learner-state change. It remains distinct from digital calibration only if that transfer persists after complete uncoupling, and distinct from fixed-standard trimming only if paired reference exposure reproduces the declared response envelope under an equal write-energy comparison.","authority_safety":{"decision_authority":"The device owner may authorize only sacrificial, nonproduction coupons, while the laboratory electrical-safety custodian sets non-overridable current, voltage, temperature, and pulse-count limits. Production acceptance remains outside this experiment's authority.","authorized_first_step":"A current-limited probe-station trial using two read-only reference coupons and four sacrificial learner coupons, with the reference write rails physically absent.","excluded_actions":["Connection to production hardware or live workloads","Use of customer or sensitive data as excitation","Any write-capable electrical path to either reference tile","Bypassing current, voltage, pulse-count, or thermal limits","Acceptance based solely on the reference used for adaptation","Calibration outside the reference's characterized operating envelope","Irreversible package modification of nonsacrificial devices"],"halt_rollback":"A hardware cutoff opens the learner write path on excess current, temperature, reference-state change, or increasing mismatch over consecutive bounded exposures. The fixture is unplugged immediately. A learner is reset only within its prevalidated endurance window; otherwise the sacrificial coupon is quarantined rather than forced back to its initial state."},"negative_tests":{"strongest_counterevidence":"An equal-energy precision-standard trim or sham-paired fixture yields the same retained response envelope, or the direct-paired learner agrees only with the training reference and not the independent reference. Either result removes the claimed causal contribution of reference-anchored co-exposure.","problem_falsifier":"Independent raw electrical metrology shows that the replacement discrepancy originates after analog conversion—for example in quantization, digital arithmetic, or software scaling—or shows that nominal programmed conductance fully predicts the physical transfer response within the declared envelope.","intervention_falsifier":"Under safe bounded pulses, mismatch fails to move consistently toward the bridge deadband, the reference state changes, the learner diverges when tested at untrained voltage points, or the transferred state disappears immediately after physical disconnection.","risks":["Copying a defective or atypical reference response","Write-cycle wear or irreversible learner drift","Electrical or thermal damage from unstable correction pulses","Oscillation around the bridge deadband","Overfitting to one voltage, temperature, or reference tile","Reference disturbance through inadequate read isolation","False independence if both references share correlated fabrication error","Retention loss after uncoupling"]},"next_evidence_step":"On an isolated probe station, characterize two read-only reference coupons and four sacrificial learner coupons at a small, predeclared set of voltages and temperatures. Within each learner, assign matched cell blocks to direct reference coupling, equal-energy precision-standard trimming, and sham coupling. Enforce fixed pulse-count, current, and temperature ceilings; then physically disconnect every adaptation path. Measure raw current-voltage responses immediately and after one bounded retention interval against both references. The first-step decision is whether direct-coupled blocks enter and remain within the predeclared envelope for both references without measurable reference change; halt on any safety-limit breach or sustained divergence.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"No other proposals or experiment cells were inspected. Internally, this candidate realizes the archetype as direct analog-material state transfer between hardware systems, rather than as human mentorship, policy, information routing, or software-mediated learning.","revision_record":{"parent_version":null,"progress_targets_addressed":["Constructed an essential physical-material causal effect","Preserved bounded authority, paired exposure, corrective exchange, secondary anchoring, and autonomy release","Specified serious rivals, counterfactuals, safeguards, and falsifiers","Kept prior art unsearched and claims mechanism-limited"],"conceptual_changes":["Initial version recasts the mentor as a bounded, qualified physical reference and the mentee as a writable nonvolatile compute tile.","Tacit transmission is realized as transfer of a multi-point electrical response rather than transmission of descriptions or digital coefficients."],"operational_changes":["Specified a read-isolated differential bridge, learner-only programming path, passive safety limits, and complete physical disconnection.","Added an independent secondary reference and equal-energy physical control condition."],"evidence_changes":["No external evidence or prior-art evidence was introduced.","Defined a bounded coupon experiment with raw electrical endpoints and explicit halt conditions."],"claim_changes":["Made no claim of novelty, prevalence, demand, or effect size.","Limited the proposal to a falsifiable contrast between direct physical co-exposure, digital calibration, and fixed-standard trimming."]},"substrate_contract":{"primary_allowed_process":"PHYSICAL_MATERIAL","counterfactual_independence":"Remove all software, algorithms, databases, dashboards, inference, reporting, incentives, authorization rules, and procedural enforcement. If the hard-wired fixture is energized, the reference-learner current difference still arises from circuit physics, the differential stage still applies a bounded electric field to the learner, mobile-ion redistribution still changes its conductance, and the nonvolatile state still remains after disconnection. Human authorization and measurement records affect safety and evaluation only; they do not produce the transfer.","forbidden_channel_audit":"The operative path contains no ADC, firmware, stored target, model, recommender, database, dashboard, network, software controller, or human decision based on sensor output. Analog transistors transduce instantaneous mismatch directly into learner-only write energy; no inferred state is computed. Logging instruments may assess the experiment but are outside the intervention loop. Safety roles, write permissions, and acceptance procedures are wrappers, while the essential effect is electric-field-driven material reconfiguration."}}}