Selectivity Curve Sweep¶
Protocol — instantiates Selectivity-Window Calibration
Runs the selector across a planned range of the control parameter and plots target yield against non-target capture.
Before anyone can say where a selector works, someone has to deliberately push it past both edges of where it works. Selectivity Curve Sweep is the discovery protocol that does exactly that: it varies one control parameter across a planned grid, holds everything else fixed, and at each setting measures target success and non-target capture as separate channels — never a single blended score. The output is the archetype's founding artifact, a two-sided response curve on which a lower bound (where discrimination first becomes usable) and an upper bound (where extra intensity stops helping and turns indiscriminate or reversed) can be read directly. Its defining move is that it goes on purpose into the bad regions on both sides; a sweep that only samples where things already look good never learns the shape of the window it is trying to find.
Example¶
A process chemist is dialing in a Suzuki–Miyaura cross-coupling that should build a desired biaryl. The instinct is "run it hot to push conversion," but conversion alone hides what else is forming. The sweep sets a temperature grid from 40 to 120 °C in 10-degree steps, keeps catalyst loading, solvent, and time constant, and at each temperature assays two things separately: yield of the target biaryl and the burden of side products (protodeboronation and homocoupling impurity). Plotted together, the picture is not a ramp. Below about 60 °C target yield is near zero — the reaction is too sluggish to discriminate. From roughly 70 to 90 °C target yield is high and impurity stays low: the window. Past 100 °C the impurity climbs steeply, and above about 110 °C the side pathway dominates so hard that target yield itself falls — a reversal. The deliverable is a labeled curve with a lower bound near 65 °C and an upper bound near 95 °C, the raw evidence every later mechanism will lean on.
How it works¶
- Pick one axis, freeze the rest. Choose the single control parameter to vary; hold confounders constant so the curve is attributable (or run a small factorial when interactions are suspected).
- Plan a grid that brackets both edges. Extend the range deliberately past where you expect the window to end on each side, so the sweep actually captures the lower onset and the upper collapse rather than a plausible-looking middle.
- Measure target and non-target separately at every point. Two (or more) channels: target yield, and each non-target/collateral outcome. A one-number success metric is the enemy here.
- Replicate near the edges and read the bounds. Add repeats where the curve turns fastest; mark the lower bound where target crosses its minimum-effective level and the upper bound where non-target crosses its maximum-acceptable level or where the target response reverses.
Tuning parameters¶
- Step size / sampling density — finer steps resolve a narrow window and sharp turns but cost more runs; coarse steps are cheap but can step clean over a thin band.
- Range span — how far past each expected edge the grid reaches. Too narrow and the sweep never sees a bound; too wide and effort is spent deep in obviously-bad territory.
- Replication near turns — repeats where the curve bends most; more replication tightens the bound estimate at the price of throughput.
- Non-target channels measured — which collateral outcomes are tracked. Add a channel and you can see a new failure mode; every channel adds assay cost.
- One-factor vs. small factorial — sweeping a single axis is simplest; a factorial reveals interactions (temperature × time) but multiplies runs.
When it helps, and when it misleads¶
Its strength is converting a monotonic "more is better" intuition into an evidenced, two-sided curve — and, uniquely among the discovery tools, it is the one that can catch a reversal, because it keeps sampling past the upper edge instead of stopping at the first good result. It is the raw material a formal specification, an audit, or a drift chart all presume already exists.
Its failure mode is under-sampling: a grid too coarse or too narrow can miss a thin window entirely, and a one-factor-at-a-time sweep is blind to the interactions that a response-surface design would surface.[n1] The classic misuse is stopping the sweep the moment a setting looks "good enough," which guarantees you never observe the upper edge and leaves the window one-sided by construction. The guarding discipline is simple to state and easy to skip under time pressure: always sweep past both edges, and space the grid finely enough to resolve the turns.
How it implements the components¶
control_parameter_axis— it names and operationalizes the one parameter being varied, defining the axis the whole calibration hangs on.selectivity_response_map— its primary output: the target-vs-non-target curve across the swept axis.lower_selectivity_bound— located empirically where target response first crosses its minimum-effective level.upper_selectivity_bound_and_reversal_guard— located where non-target burden crosses its ceiling or where the target response itself reverses.
It discovers the band but does not codify it into an operating_window_rule, stratified_window_table, or window_version_register — that formalizing job belongs to Operating Band Specification.
Related¶
- Instantiates: Selectivity-Window Calibration — the sweep supplies the raw response map the rest of the pattern presumes.
- Sibling mechanisms: Operating Band Specification · Challenge-Panel Cross-Reactivity Test · ROC or Precision–Recall Surface Review · Bycatch Audit · Window Drift Control Chart · Selective Admission Band Protocol
Editorial Notes¶
Form Classification¶
Form family: Experiment, Test & Rehearsal
Rationale: Selectivity Curve Sweep operates as an active test, trial, simulation, drill, or rehearsal that generates evidence through a deliberate attempt or perturbation because it runs the selector across a planned range of the control parameter and plots target yield against non-target capture.
Independent corroboration: The frozen evidence defines Selectivity Curve Sweep as 'Runs the selector across a planned range of the control parameter and plots target yield against non-target capture', so its operative form is Experiment, Test & Rehearsal.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Sweeping a decision variable and plotting desired against undesired capture is statistical operating-characteristic analysis.
Related originating lineages:
- Chemistry & Materials Science — Separation science measures selectivity and yield across process conditions.
- Data Science & Analytics — Data science, analytics, and operational monitoring supplies a parallel or contributing lineage for the mechanism's defining operation: runs the selector across a planned range of the control parameter and plots target yield against non-target capture.
- Engineering & Design — Engineering qualification uses parameter sweeps to locate acceptable operating regions.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: runs the selector across a planned range of the control parameter and plots target yield against non-target capture.
- Medicine & Healthcare — Diagnostic ROC practice supplies a familiar target-detection versus false-capture curve.
- Operations Research — Operations research, optimization, and queueing analysis supplies a parallel or contributing lineage for the mechanism's defining operation: runs the selector across a planned range of the control parameter and plots target yield against non-target capture.
Review resolution: The blind reviewers agree that statistics_experimental_design is the primary origin and differ only on alternate origin disagreement, origin mode disagreement, domain reach disagreement, encyclopedia synthesis disagreement. I preserve every independently explained alternate from both records rather than imposing a numeric cap. I retain convergent because the combined record shows independent disciplinary development. The broader reach of multi_domain records portability separately from historical provenance, and encyclopedia_synthesis=true preserves the affirmative synthesis judgment where either reviewer identified one.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Reconciled after independent review; medium confidence.
Notes¶
[n1] Design of experiments (and its response-surface variant) plans the sampling grid so that a window and any parameter interactions are actually resolvable, rather than varying one factor at a time and hoping the shape reveals itself; the "process window" is the industrial name for the band this protocol is hunting. ↩