{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp12_substrate_denial72_20260805","cell_id":"prediction_error_learning_calibration__chemistry_materials","arm":"CONSTRAINED_HIGH","candidate_id":"prediction_error_learning_calibration__chemistry_materials__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__chemistry_materials","arm":"CONSTRAINED_HIGH","candidate_id":"prediction_error_learning_calibration__chemistry_materials__CONSTRAINED_HIGH","proposal_index":1,"version":0,"title":"Passive Witness-Coupon Springback Calibrator for Repeated Composite Molding","problem":"Repeatedly molded composite panels can emerge with persistent curvature error when resin lot, fiber moisture, cure shrinkage, or thermal history changes. A fixed compensated mold continues imposing the correction appropriate to an earlier material state, while changing compensation from each raw panel result risks chasing measurement noise and part-to-part variation.","actors":["Composite material undergoing cure and shrinkage","A witness coupon co-cured beside each panel","A passive curvature-comparison fixture","A mechanically adjustable mold insert","Materials or process engineer responsible for the bounded trial"],"observable_state":"After cooling and release, the co-cured witness coupon has a measurable signed curvature relative to a physically stored target contour: over-curved, within a deadband, or under-curved. The fixture's mold-compensation position and ratchet history are also directly inspectable.","consequence":"Persistent signed springback error produces panels outside the intended geometry; indiscriminate correction from every result can instead cause oscillation, excess rework, or incorrect attribution of local panel defects to batch-wide cure shrinkage.","affected_objective":"Maintain panel curvature near its specified geometry across repeated batches while limiting overreaction to coupon noise and preserving traceable, bounded mold compensation.","intervention":"Co-cure a standardized witness strip in a defined location, then place it in a passive differential-curvature fixture containing a master target arc. Contact displacement relative to that arc produces a signed mechanical deflection. A central deadband produces no adjustment for expected variation. Deflection beyond either side of the deadband advances one of two opposed ratchets by a small bounded increment, changing the preload of a flexible mold insert in the compensating direction for the next batch. A viscous dashpot and ratchet pitch limit update gain; hard stops bound total compensation. The comparison, sign separation, memory, and update are embodied mechanically and require no computation or reporting to operate.","structural_mapping":[{"archetype_element":"Prior Prediction Record","domain_realization":"The master target arc plus the mold insert's current preload physically encode the expected witness curvature for the next cure."},{"archetype_element":"Value Reference Frame","domain_realization":"Signed curvature deviation from the specified target contour is the single defined comparison quantity."},{"archetype_element":"Received Outcome Record","domain_realization":"The cooled witness strip retains the realized batch curvature, and the fixture converts it into contact displacement."},{"archetype_element":"Signed Error Signal","domain_realization":"Opposite displacement directions engage separate over-curvature and under-curvature ratchets."},{"archetype_element":"Credit Assignment Window","domain_realization":"A standardized coupon co-cured beside one panel confines the observation to that batch and cure location; unrelated local panel damage is not admitted as the update signal."},{"archetype_element":"Noise and Volatility Filter","domain_realization":"Fixture deadband, replicated witness strips during validation, and dashpot resistance suppress responses to small deviations and brief placement disturbances."},{"archetype_element":"Learning Gain Rule","domain_realization":"Ratchet tooth pitch fixes the incremental correction per qualifying error, while the dashpot limits impulsive advancement."},{"archetype_element":"Update Target","domain_realization":"The ratchet changes only the flexible mold insert's compensating curvature for the subsequent batch."},{"archetype_element":"Prediction History Memory","domain_realization":"Ratchet position is persistent physical memory of accumulated signed corrections."},{"archetype_element":"Positive/Negative Error Separation","domain_realization":"Mechanically opposed pawls preserve error direction and move the insert toward greater or lesser compensation."},{"archetype_element":"Shortcut Learning Guard","domain_realization":"Coupon geometry, placement, release method, and thermal contact are standardized so fixture seating or local edge damage does not become the apparent shrinkage signal."}],"mechanism_mapping":[{"mechanism_slug":"prediction_outcome_delta_log","role":"Realized as the differential gap between the master arc and the co-cured witness strip; the gap exists as mechanical displacement rather than a digital log.","counterfactual_removal":"Without the differential comparison, raw coupon curvature has no signed relation to the stored expectation and cannot select a compensating update."},{"mechanism_slug":"negative_prediction_error_review","role":"The under-curvature side of the comparator engages only its corresponding pawl, preserving a worse-than-expected signed direction.","counterfactual_removal":"Without direction-specific engagement, under-curvature and over-curvature could produce the same adjustment and destabilize compensation."},{"mechanism_slug":"positive_surprise_capture","role":"The over-curvature side engages the opposed pawl, physically retaining the opposite signed deviation as a reverse increment.","counterfactual_removal":"Without the opposed path, the fixture could learn in only one direction and could not follow reversals in material drift."},{"mechanism_slug":"learning_rate_schedule","role":"Fixed ratchet pitch, dashpot resistance, and hard stops bound the size and speed of mechanical updates.","counterfactual_removal":"Without bounded gain, one anomalous coupon could cause a large mold change or drive the insert beyond its validated range."},{"mechanism_slug":"credit_assignment_trace","role":"The one-batch co-cure linkage and dedicated fixture position associate the correction with the relevant cure exposure rather than with a later mixed batch.","counterfactual_removal":"Without batch linkage, the ratchet could update the next mold setting from a coupon produced under unrelated conditions."},{"mechanism_slug":"shortcut_probe_holdout_set","role":"During validation, deliberately mis-seated and edge-damaged coupons serve as physical probes to confirm that locating features and rejection gauges block those incidental cues.","counterfactual_removal":"Without shortcut probes, fixture seating error could masquerade as cure-induced curvature and receive mechanical credit."}],"causal_chain":["The current insert preload and master arc embody the expected curvature response for the next repeated cure.","A standardized witness strip experiences the batch's resin, moisture, temperature, and shrinkage conditions.","After release and cooling, the strip physically retains the realized curvature outcome.","The differential fixture converts target-versus-realized curvature into a signed displacement.","Displacement inside the deadband leaves both pawls disengaged, so an essentially expected outcome causes no update.","A reliable displacement outside the deadband engages only the pawl matching its sign.","The ratchet advances by one bounded increment and mechanically stores the correction.","The adjusted insert imposes opposite pre-curvature on the next molded panel, counteracting persistent springback bias.","Repeated same-sign errors accumulate bounded correction, while reversed errors unwind it through the opposed ratchet."],"baseline":"A fixed compensated mold selected during initial qualification and left unchanged across material lots, with panels inspected after molding and nonconforming parts reworked or rejected.","nearest_rivals":["Manual measurement of panel or witness curvature followed by technician adjustment of mold shims","Optical metrology connected to a software-controlled servo mold","Active control of cure temperature or pressure to hold a prescribed thermal-mechanical trajectory","Reformulation or tighter incoming control of resin, fiber moisture, and layup materials","A compliant mold that passively accommodates shrinkage but retains no signed error memory"],"remaining_contrastive_claim":"For a repeated molding process whose dominant batch-level curvature bias is represented by a standardized co-cured witness strip, a passive signed-error comparator with a deadband and bounded bidirectional mechanical memory can track persistent springback drift without requiring digital inference or treating every raw curvature observation as an equal-strength correction. This claim does not extend to spatially heterogeneous defects, non-repeatable layups, or drift that the witness coupon fails to represent.","authority_safety":{"decision_authority":"A materials or process engineer with authority over the laboratory mold and its validated compensation envelope.","authorized_first_step":"Build and operate a benchtop, low-energy fixture on inert molded strips or coupons; permit only one-tooth adjustments within premarked hard stops while independently measuring curvature.","excluded_actions":["Installation on production tooling","Use on pressure-vessel, flight-critical, implantable, or other safety-critical composites","Adjustment beyond the insert's validated elastic range","Changing cure chemistry, temperature, pressure, or safety interlocks","Using a cracked, delaminated, mis-seated, or dimensionally nonconforming witness coupon","Allowing the mechanism to actuate during mold loading, cure, or personnel access"],"halt_rollback":"Lock the ratchet and return the insert to its indexed baseline position if errors alternate without convergence, coupon and panel curvature cease to correlate, the fixture binds, a hard stop is reached, or any insert damage appears. The indexed baseline and removable ratchet permit manual rollback without altering cure controls."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence would be repeated trials showing that witness-strip signed curvature poorly predicts the corresponding panel's signed springback, or that the mechanical update oscillates more than a fixed mold under the same drift sequence.","problem_falsifier":"The problem framing is falsified if panel curvature error is predominantly random, local, or caused by trimming and fixturing after cure rather than by a recurring batch-level shrinkage state observable in the co-cured coupon.","intervention_falsifier":"The intervention is falsified if, under imposed repeatable drift and blinded coupon ordering, correctly signed ratchet updates do not reduce subsequent signed curvature bias relative to a locked-ratchet control, or if mis-seating probes routinely cause updates.","risks":["The witness coupon may not represent the panel's thickness, fiber architecture, thermal gradients, or tool contact.","Mechanical friction, backlash, wear, or thermal expansion may bias the comparator.","A deadband that is too narrow may chase noise; one that is too wide may ignore persistent drift.","Repeated increments may conceal a resin, moisture, cure, or tooling fault that should instead halt production.","Insert compensation may correct mean curvature while worsening local waviness or residual stress.","Opposed pawls may both partially engage if the neutral geometry is poorly designed.","Coupon handling damage may be misattributed to cure shrinkage.","Stored mechanical position may be mistaken for evidence that the underlying material process is stable."]},"next_evidence_step":"Run a bounded benchtop sequence of 24 inert or non-safety-critical molding cycles divided among stable, deliberately higher-shrinkage, and deliberately lower-shrinkage conditions. Randomize condition order within short blocks; co-cure two witness strips per cycle; independently measure strip and panel curvature; compare a locked-ratchet fixture with the one-tooth adaptive fixture; and test predefined mis-seating and edge-damage probes. Evaluate sign agreement, false-update frequency, oscillation, hard-stop approach, and whether next-cycle signed panel bias changes in the mechanically predicted direction. Stop on insert damage, binding, or three consecutive sign reversals.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not assessed against other proposals because runtime isolation forbids inspecting them; this candidate is derived only from the supplied archetype and domain card.","revision_record":{"parent_version":null,"progress_targets_addressed":["Construct one causally defensible chemistry/materials candidate","Preserve prior-prediction, signed-error, filtering, credit-assignment, and bounded-update structure","Ensure the essential effect is physical and survives removal of forbidden wrappers","Specify falsifiers, rivals, safeguards, and a bounded evidence step"],"conceptual_changes":["Translated prediction-error learning into passive mechanical comparison and persistent mold-geometry adaptation."],"operational_changes":["Specified a co-cured witness strip, master arc, deadband, opposed ratchets, dashpot, flexible insert, indexed baseline, and hard stops."],"evidence_changes":["Defined a 24-cycle locked-versus-adaptive benchtop test with imposed drift and physical shortcut probes."],"claim_changes":["Limited the contrastive claim to repeated batch-level curvature drift represented by the witness coupon and made no novelty, prevalence, demand, or effect-size claim."]},"substrate_contract":{"primary_allowed_process":"HYBRID_OTHER_ALLOWED_PRIMARY","counterfactual_independence":"The primary process is passive mechanical and spatial transduction coupled to physical mold-shape modification: cured-material curvature displaces a comparator, direction-selective pawls increment mechanical memory, and changed insert geometry alters the next panel's forming boundary condition. Removing software, algorithms, databases, dashboards, reporting, incentives, authorization workflows, and procedural enforcement does not remove this compare-select-store-adjust causal path. Human activity is needed to load coupons and authorize a bounded trial, but not to calculate the error, choose its sign, or execute the mechanical update.","forbidden_channel_audit":"No algorithm, model, database, recommender, digital controller, information-routing system, incentive, policy, training program, or downstream analytics supplies the operative correction. Measurement is direct contact metrology whose displacement mechanically actuates the update. Optional written records or independent gauges may evaluate safety and evidence but are not in the essential causal chain. If the comparator required a person or software system to interpret a reading and decide the adjustment, the proposal would fail this contract; the specified opposed-pawl linkage avoids that dependency."}}}