{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp12_substrate_denial72_20260805","cell_id":"prediction_error_learning_calibration__computer_science","arm":"CONSTRAINED_HIGH","candidate_id":"prediction_error_learning_calibration__computer_science__CONSTRAINED_HIGH","decision":"PROPOSAL","abstention_reason":null,"proposal":{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp12_substrate_denial72_20260805","cell_id":"prediction_error_learning_calibration__computer_science","arm":"CONSTRAINED_HIGH","candidate_id":"prediction_error_learning_calibration__computer_science__CONSTRAINED_HIGH","proposal_index":1,"version":0,"title":"Thermo-Mechanical Prediction-Error Damper for Server-Rack Airflow","problem":"A server rack with a manually balanced or raw-temperature-actuated airflow damper can respond incorrectly to changing compute heat loads because it treats outlet temperature itself as the teaching signal. A high outlet temperature that is physically predicted from high electrical load may provoke unnecessary further adjustment, while a smaller temperature rise that is unexpectedly large for the measured load, inlet temperature, and existing damper angle may reveal degraded heat transfer yet produce little adjustment.","actors":["Server rack and installed computing equipment","Heat-producing processors and power components","Rack inlet and exhaust air streams","Rack airflow damper","Facilities engineer responsible for the pilot"],"observable_state":"At a qualifying rack, repeated workload heat pulses produce exhaust temperatures whose deviations from a physically generated forecast have a consistent sign under some combinations of rack current, inlet temperature, and damper angle. A fixed damper retains its prior setting, while a direct bimetallic thermostat responds to raw exhaust temperature without distinguishing an expected hot exhaust from an unexpectedly hot one.","consequence":"Airflow can remain misallocated or oscillate after load changes, creating avoidable thermal-margin loss, thermal throttling risk, or excessive cooling of a rack whose hot exhaust was already expected from its load.","affected_objective":"Maintain computing-component temperatures within established limits while allocating rack airflow according to persistent, load-adjusted thermal mismatch rather than raw exhaust magnitude.","intervention":"Install a bounded thermo-mechanical module on one rack damper. An isolated current transformer drives a proportional resistive heater inside a sealed reference bellows; inlet air temperature and a cam linked to the current damper angle also alter that bellows' pressure, producing a physical forward estimate of the rack's later exhaust temperature. A second matched bellows is exposed to actual rack exhaust. Their opposed forces create signed displacement equal to actual-minus-predicted thermal state. A deadband, thermal mass, and viscous dashpot reject small or short deviations. Sustained positive displacement advances one pawl of a bidirectional ratchet to open the damper; sustained negative displacement advances the opposite pawl to close it. Mechanical end stops bound travel, a spring returns the damper to a safe preset position if the module loses power or pressure, and the retained ratchet position supplies the cross-cycle update memory.","structural_mapping":[{"archetype_element":"Prior Prediction Record","domain_realization":"Reference-bellows pressure is established from upstream rack current, inlet temperature, and damper angle before the corresponding exhaust response develops; its thermal time constant retains that expected state through the comparison interval."},{"archetype_element":"Value Reference Frame","domain_realization":"Value is the rack exhaust thermal state expected for the measured electrical heat input, inlet condition, and available airflow."},{"archetype_element":"Received Outcome Record","domain_realization":"The exhaust bellows directly converts the realized outlet-air temperature into pressure and force."},{"archetype_element":"Signed Error Signal","domain_realization":"Opposed bellows mechanically subtract expected pressure from realized pressure, preserving hotter-than-predicted and cooler-than-predicted directions."},{"archetype_element":"Credit Assignment Window","domain_realization":"A dedicated exhaust pickup, upstream current transformer, and matched thermal response time restrict each comparison to one rack and the heat pulse that preceded its exhaust response."},{"archetype_element":"Noise and Volatility Filter","domain_realization":"Mechanical deadband, thermal mass, and viscous damping prevent small measurement differences and brief exhaust turbulence from advancing the ratchet."},{"archetype_element":"Learning Gain Rule","domain_realization":"Bellows area, lever ratio, pawl pitch, and damper gearing determine how much persistent signed mismatch is required for each retained angular update."},{"archetype_element":"Update Target","domain_realization":"The ratchet changes and remembers the damper's neutral opening angle for subsequent heat-load cycles."},{"archetype_element":"Positive/Negative Error Separation","domain_realization":"Separate opposed pawls convert positive mismatch into opening steps and negative mismatch into closing steps."},{"archetype_element":"Ethical Reward Safety Review","domain_realization":"Hard travel stops, fail-safe spring return, electrical isolation, and retention of independent rack thermal shutdowns bound the physical intervention."}],"mechanism_mapping":[{"mechanism_slug":"prediction_outcome_delta_log","role":"The two bellows and differential linkage physically instantiate the prior forecast, received outcome, and signed delta; the ratchet retains the resulting update without a database.","counterfactual_removal":"Removing the reference bellows or differential linkage eliminates comparison with expectation and reduces the device to raw-temperature actuation, breaking the archetype."},{"mechanism_slug":"credit_assignment_trace","role":"Rack-specific current pickup, exhaust sampling location, and matched thermal lag associate a mismatch with the same rack and preceding heat input.","counterfactual_removal":"Without this spatial and temporal coupling, neighboring heat or a later workload pulse could move the wrong rack's damper."},{"mechanism_slug":"learning_rate_schedule","role":"The dashpot, deadband, lever ratio, and ratchet pitch physically bound update frequency and magnitude.","counterfactual_removal":"Without these elements, momentary turbulence or sensor noise could produce persistent damper changes."},{"mechanism_slug":"positive_surprise_capture","role":"The bidirectional ratchet separately retains hotter-than-predicted and cooler-than-predicted updates.","counterfactual_removal":"A one-direction mechanism could open for positive errors but could not learn from negative errors or return airflow when realized heat is below expectation."}],"causal_chain":["Rack current, inlet temperature, and existing damper angle physically establish a reference-bellows state representing expected later exhaust temperature.","The rack converts electrical power into heat, and the corresponding air parcel reaches the exhaust pickup after the rack's thermal delay.","Actual exhaust temperature expands the outcome bellows while the reference bellows supplies the prior expected force.","The opposed linkage produces a signed displacement rather than a response to raw exhaust temperature.","Deadband, thermal inertia, and viscous damping suppress deviations too small or brief to count as stable mismatch.","A sustained signed displacement advances the corresponding ratchet pawl by a mechanically bounded increment.","The retained ratchet position changes damper angle and therefore the airflow available during later heat-load cycles.","When realized exhaust matches the physical forecast within the deadband, neither pawl advances, so an expected outcome does not keep generating updates."],"baseline":"The primary baseline is the existing fixed manual damper. A second diagnostic baseline is a direct bimetallic exhaust-temperature damper that moves from raw temperature relative to a fixed setpoint and contains no independent load-conditioned prediction.","nearest_rivals":["A fixed or periodically hand-balanced rack damper","A direct bimetallic thermostat that responds to raw exhaust temperature","A pressure-responsive passive damper that balances static pressure without predicting rack heat","A firmware-controlled fan curve driven by processor temperature","A digitally modeled predictive cooling controller using sensors, software, and motorized dampers"],"remaining_contrastive_claim":"The candidate's distinguishing causal claim is that persistent damper updates arise only from the signed mechanical difference between an independently generated pre-outcome thermal forecast and the later rack exhaust, with local credit timing and physical gain limits. Fixed dampers do not update, direct thermostats use raw temperature or setpoint error, pressure dampers use pressure imbalance, and digital controllers lose their operative effect when computation is removed. Whether the proposed physical comparison is accurate or useful remains an empirical question.","authority_safety":{"decision_authority":"A facilities engineer authorized for rack airflow hardware may approve a bench prototype and, only after separate review, a noncritical-rack pilot.","authorized_first_step":"Construct and test one externally mounted module on an isolated airflow bench containing a resistive dummy server load; keep the damper's safe preset and all independent over-temperature cutoffs active.","excluded_actions":["Installing the untested module on a production or safety-critical rack","Disabling processor, power-supply, rack, or facility thermal shutdowns","Changing workload-scheduling software to manufacture favorable responses","Increasing fan or electrical equipment operation beyond rated limits","Allowing the learned damper position to exceed mechanically verified end stops"],"halt_rollback":"Stop if the damper binds, changes direction contrary to the signed bellows displacement, reaches an end stop repeatedly, loses bellows pressure, or permits the dummy-load temperature to cross its preset cutoff. De-energize the reference heater, disengage the ratchet, and let the fail-safe spring return the damper to its marked preset angle."},"negative_tests":{"strongest_counterevidence":"If the reference bellows merely behaves as a disguised temperature setpoint, rather than producing a forecast independently conditioned on prior rack current, inlet temperature, and damper angle, the system is an ordinary thermostat and does not preserve prediction-error learning.","problem_falsifier":"The proposed problem is falsified for the test rack if upstream current, inlet temperature, and damper angle do not produce a repeatable leading relation to later exhaust temperature, or if raw-temperature control and signed forecast-error control issue the same direction and persistence of adjustment across the bounded test conditions.","intervention_falsifier":"The intervention is falsified if actual-minus-predicted sign does not reliably determine differential-linkage direction; near-zero predicted errors still advance the ratchet; reversing the error sign fails to reverse updates; or locking the reference bellows leaves substantially the same adaptive behavior.","risks":["Reference-bellows drift can create a persistent false error.","Thermal lag mismatch can assign an exhaust outcome to the wrong heat pulse.","Ratchet wear, stiction, or backlash can bias updates or prevent fail-safe return.","Opening one rack's damper can reduce airflow available to neighboring racks in a shared plenum.","A ruptured capsule or failed reference heater can move the mechanism unless the fail-safe spring is independently verified.","A narrow thermal reference frame can favor exhaust matching while missing component-local hotspots."]},"next_evidence_step":"On an isolated airflow bench, run a preregistered matrix of twelve dummy-load cycles covering three electrical heat steps, two inlet temperatures, and two repetitions. Before each exhaust response, record the reference-bellows displacement with an independent scale; then record exhaust-bellows displacement, signed differential, pawl direction, and retained damper step. Include locked-ratchet and direct-bimetal baselines. The bounded test passes only as a mechanism check if reference state precedes the corresponding outlet response, sustained error sign predicts pawl direction, sign reversal reverses the update, and within-deadband cycles produce no retained step; it makes no claim about production benefit.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"No comparison with other proposals was performed because runtime isolation forbids their inspection. Internally, this candidate is defined as a rack-local thermo-mechanical adaptive device whose memory and signed comparison are embodied in bellows, damping, and a ratchet rather than software, reporting, incentives, or procedure.","revision_record":{"parent_version":null,"progress_targets_addressed":["Construct one concrete computer-systems problem with a physical primary mechanism.","Preserve prior prediction, received outcome, signed error, credit assignment, filtering, bounded gain, and retained update.","Make the essential effect counterfactually independent of forbidden software and governance channels.","Specify serious rivals, falsifiers, safeguards, and a bounded first evidence step."],"conceptual_changes":["Initial version maps prediction-error learning to repeated physical calibration of server-rack airflow rather than to model training or software feedback."],"operational_changes":["Initial version specifies a current-heated reference bellows, exhaust bellows, differential linkage, dashpot, bidirectional ratchet, end stops, and fail-safe spring."],"evidence_changes":["Initial version limits evidence collection to an isolated dummy-load bench matrix with locked-ratchet and raw-temperature baselines."],"claim_changes":["Claims are limited to the proposed causal contrast and testable mechanism; novelty, prevalence, demand, and effect size remain unclaimed."]},"substrate_contract":{"primary_allowed_process":"HYBRID_OTHER_ALLOWED_PRIMARY","counterfactual_independence":"If all software, algorithmic inference, databases, dashboards, reporting, incentives, authorization rules, and procedural enforcement are removed after safe installation, rack current still induces reference-heater energy, the sealed bellows still transduce predicted and realized thermal states into opposed forces, the dashpot and deadband still filter transients, the ratchet still retains signed updates, and the damper still changes airflow. These physical, energetic, mechanical, and measurement processes therefore carry the essential causal effect independently of forbidden wrappers.","forbidden_channel_audit":"The intervention contains no processor, firmware, learned model, database, network connection, dashboard, recommendation, or human-in-the-loop actuation. The current transformer and thermal bellows do not merely report measurements: their electrical heating, expansion, force subtraction, damping, and ratcheting directly move and retain the airflow-control element. Human authority is confined to safe installation, testing, and removal and is not the mechanism that performs prediction comparison or updating."}}}