{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp09_archetype_breadth150_20260804","cell_id":"context_keyed_representation_switching__chemistry_materials","arm":"BREADTH_PROBE_ONE_SHOT","candidate_id":"context_keyed_representation_switching__chemistry_materials__P1","proposal_index":1,"version":0,"title":"Recipe-Keyed Cure-State Maps for a Shared Inline Spectrometer","problem":"A composites pilot line uses one inline near-infrared spectrometer and cure-state estimator for two recurring filled-epoxy formulations. Because resin chemistry, filler scattering, and spectral baselines differ by formulation, the same measured spectral features can correspond to different degrees of cure. A pooled or recently updated calibration can therefore silently apply one formulation's spectrum-to-conversion correspondence to the other formulation.","actors":["Composite-process operator who selects the batch recipe and observes the active calibration","Materials scientist who owns formulation-specific reference measurements and calibration versions","Process engineer who defines cure endpoints and reviews estimator residuals","Quality engineer who audits batch records and authorizes production use"],"observable_state":"For each batch, the recipe identifier, raw spectra, temperature history, active calibration identity, estimated conversion, and offline reference conversion are observable. The problem is indicated when residuals cluster by formulation, when updating the estimator on one formulation worsens held-out estimates for the other, or when returning to a previously run formulation no longer reproduces its prior calibration results.","consequence":"Wrong-map activation can declare a part cured too early or keep it heating unnecessarily, creating avoidable risk of under-cure, thermal overexposure, scrap, and loss of traceability between the spectral evidence and the endpoint decision.","affected_objective":"Produce traceable and formulation-appropriate cure-state estimates on a shared instrument without allowing calibration changes for one recurring formulation to corrupt the other.","intervention":"Maintain separate, versioned spectrum-to-conversion maps for the two formulations on the shared spectrometer controller. Route each batch to a map using the operator-confirmed recipe identifier cross-checked against batch metadata; display the active map before spectra are interpreted; freeze the inactive map against routine fitting updates; switch only between batches after clearing formulation-specific state; abstain when the recipe key is absent or conflicting; and require interference and re-entry tests before either map can influence cure-endpoint decisions.","structural_mapping":[{"archetype_element":"Shared Substrate Scope","domain_realization":"One inline spectrometer, preprocessing pipeline, controller, and shared wavelength and temperature channels serve both epoxy formulations."},{"archetype_element":"Representation Portfolio","domain_realization":"Two persistent calibration maps separately encode the correspondence from preprocessed spectra and temperature to reference degree of cure."},{"archetype_element":"Context Key Space and Cue Model","domain_realization":"The bounded contexts are formulation A and formulation B; the primary cue is the operator-confirmed recipe identifier, checked against batch and material-lot records rather than inferred from spectra alone."},{"archetype_element":"Shared Invariant Layer","domain_realization":"Instrument wavelength calibration, timestamping, raw-spectrum retention, and approved temperature sensing are shared; formulation-dependent preprocessing coefficients and cure correspondences remain map-specific."},{"archetype_element":"Representation Selection Rule","domain_realization":"A batch is evaluated only when its confirmed recipe key resolves to exactly one approved calibration version; missing, conflicting, or unsupported keys produce no cure estimate."},{"archetype_element":"Representation Isolation Boundary","domain_realization":"Fitting or correcting the active formulation map cannot alter the inactive map's parameters, preprocessing state, validation record, or endpoint threshold."},{"archetype_element":"Switch Transition Guard","domain_realization":"Map changes occur only at a recorded batch boundary after transient buffers and formulation-specific preprocessing state are cleared and the new map identity is acknowledged."},{"archetype_element":"Active Representation Indicator","domain_realization":"The operator display and batch record show formulation, calibration version, validation status, and selection rationale beside every estimate."},{"archetype_element":"Re-Entry Integrity Test","domain_realization":"After formulation B has been processed or updated, archived formulation-A runs are replayed through the preserved A map and compared with its locked reference results before A is reactivated."},{"archetype_element":"Unknown-Context Fallback","domain_realization":"An unknown or conflicting formulation key suppresses automated conversion and endpoint recommendations while retaining raw spectra for later review."}],"mechanism_mapping":[{"mechanism_slug":"Context-to-Map Routing Table","role":"Maps each approved formulation identifier to exactly one approved calibration checkpoint and records the selection in the batch trace.","counterfactual_removal":"Without explicit routing, operators or software could load the most recent calibration and silently interpret a batch through the wrong formulation map."},{"mechanism_slug":"Per-Context Model Checkpoint","role":"Stores formulation-specific preprocessing parameters, conversion model, endpoint threshold, provenance, and validation evidence as an indivisible version.","counterfactual_removal":"Without separate checkpoints, an update fitted to one formulation could overwrite parameters needed for reliable re-entry to the other."},{"mechanism_slug":"Active-Map Status Indicator","role":"Exposes the governing formulation and calibration version before and during estimation.","counterfactual_removal":"Without visible status, a routing error would resemble an ordinary measurement or material anomaly and would be difficult to contest."},{"mechanism_slug":"Cross-Map Interference Regression Suite","role":"Tests whether an authorized update to one formulation changes locked outputs for the other formulation.","counterfactual_removal":"Without the interference test, hidden shared preprocessing or parameter coupling could defeat nominal map separation."},{"mechanism_slug":"Rollback to Prior Map Snapshot","role":"Restores the last approved formulation-specific checkpoint if a new version fails shadow or re-entry checks.","counterfactual_removal":"Without rollback, recovery would require reconstructing a prior calibration or continuing with an unverified update."}],"causal_chain":["The two recurring epoxy formulations create different spectrum-to-conversion correspondences on the same inline instrument.","A pooled mutable calibration or implicit model selection allows the dominant or most recently updated formulation to govern both contexts.","The resulting wrong correspondence biases conversion estimates and can shift cure-endpoint recommendations.","An explicit recipe key routes each batch to one persistent formulation-specific map, while the active-map indicator makes that choice observable.","Checkpoint isolation prevents fitting activity in the active formulation from rewriting the inactive formulation's correspondence.","Guarded batch-boundary switching prevents mixed preprocessing state, and abstention blocks unsupported contexts.","Cross-map regression and re-entry replay test whether the inactive map remained intact after intervening use or updates.","Only maps that pass those checks become eligible for separately authorized endpoint use."],"baseline":"One pooled spectrum-to-conversion model is maintained for both formulations, with formulation data incorporated through ordinary refitting and no separately preserved re-entry checkpoint or visible active-map identity.","nearest_rivals":["A single global calibration with formulation identifier included as an input feature","Manual loading of separate calibration files without enforced routing, isolation, or re-entry tests","Two dedicated spectrometers and controllers, one per formulation","A continuously blended ensemble of formulation models selected by spectral similarity","A permanent migration to one standardized resin formulation and one calibration"],"remaining_contrastive_claim":"The candidate is specifically a persistent, discrete portfolio of formulation-dependent spectrum-to-cure correspondences on one shared measurement substrate: an explicit recipe key selects one visible map, inactive maps are protected from updates, and return to an earlier formulation is verified. Separate files alone, a pooled model, spectral blending, or a one-time calibration migration would not supply that selection-isolation-re-entry structure.","authority_safety":{"decision_authority":"The materials scientist may propose calibration versions, but the process and quality owners jointly authorize any version or endpoint threshold for production; the operator may confirm or reject the recipe key and may always withhold automated estimation.","authorized_first_step":"Run a read-only offline replay using archived raw spectra, recipe metadata, and reference-conversion measurements from up to 20 completed cure runs for each of the two formulations. Compare the locked pooled baseline with two isolated candidate maps under correct, swapped, missing, and conflicting recipe keys; then repeat the formulation-A replay after fitting only formulation B.","excluded_actions":["Changing heater commands, cure duration, or production endpoint thresholds","Allowing the candidate maps to stop or release a live batch","Inferring formulation from sensitive operator attributes or from spectra when the explicit recipe record is unavailable","Overwriting the existing approved calibration or archived raw data","Propagating a fitted parameter across maps without separate validation","Adding further formulation contexts during the bounded test"],"halt_rollback":"Halt the replay if recipe provenance or reference-conversion records are incomplete, if preprocessing cannot be reproduced, or if fitting one map changes the other's locked outputs. Discard the experimental copies and retain the approved baseline and immutable archive; no plant-control state is changed."},"negative_tests":{"strongest_counterevidence":"A preregistered held-out replay could show that one fixed global calibration meets the same formulation-specific error and endpoint-classification criteria as the isolated maps, remains stable after balanced updates, and exhibits no formulation-structured residuals.","problem_falsifier":"The problem is falsified for this scope if, after controlling for instrument drift and temperature history, the spectrum-to-reference-conversion relationship is invariant across the two formulations and updates using either formulation do not degrade the other's locked cases.","intervention_falsifier":"The intervention is falsified if correctly keyed isolated maps do not outperform the locked baseline on predeclared formulation-specific criteria, if recipe-key errors are common enough to offset any calibration benefit, or if re-entry outputs change after activity confined to the other map.","risks":["Incorrect or stale recipe metadata can activate the wrong calibration with an appearance of procedural legitimacy.","Excessive isolation can duplicate maintenance work and prevent valid improvements to shared instrument preprocessing.","Small reference datasets can make formulation-specific maps less stable than the pooled baseline.","Map/version proliferation can burden operators and weaken auditability.","A visible formulation label may encourage staff to trust map selection without checking sensor quality or domain limits.","Batch-boundary state clearing may add latency or omit necessary instrument-history corrections.","Reference conversion measurements may themselves be biased, causing each isolated map to preserve a different error."]},"next_evidence_step":"Before examining outcomes, define formulation-specific conversion-error and endpoint-classification acceptance bounds. Execute the authorized offline replay on no more than 40 archived runs total, including deliberate key swaps, unknown-key abstention, an update confined to formulation B, and subsequent re-entry to formulation A. Record routing accuracy, abstention behavior, per-formulation error, cross-map output changes, and rollback reproducibility; do not connect results to live control.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not assessed because runtime isolation prohibits inspection of other proposals or experiment cells; this object is derived only from the supplied archetype and domain card.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}