Funnel Process¶
Workflow — instantiates Coarse-to-Fine Search
Implements the archetype in review or product workflows by moving many candidates through cheap early screens before detailed evaluation of a smaller set.
A Funnel Process arranges evaluation as a fixed sequence of stages, each costlier and more discriminating than the last, each sized so that the batch it passes forward is one the next stage can actually afford to examine. Its distinctive object is not any single screen but the economics of the pipeline: pass-rates and per-stage costs are engineered together so that expensive capacity downstream is never flooded, and cheap capacity upstream is never the bottleneck. Where sibling mechanisms center on how a candidate is refined, the funnel centers on how many survive each cut and what each cut costs — throughput as a design variable, with a running tally of what happened to everyone who entered.
Example¶
A drug-discovery team starts with a library of a million compounds and one question: which few might inhibit a target enzyme. Testing each in a full biochemical workup is unthinkable, so they build a funnel. Stage one is a high-throughput primary assay — cheap, automated, run in plates, tolerant of noise — which flags roughly one percent as "active." Stage two re-tests those ten thousand in dose-response to confirm the signal is real and dose-dependent, killing most as artifacts. Stage three runs the surviving few hundred through selectivity and early ADME/tox panels; stage four takes a dozen into animal pharmacokinetics.
Each stage costs perhaps ten times the last per compound but sees a tenth as many, so the total bill stays bounded while fidelity climbs. Crucially, the team logs the fate of every compound at every stage — how many entered, how many advanced, why the rest were dropped — so that six months later they can ask whether the primary assay's one-percent threshold was throwing away real chemistry. The funnel's value is precisely that it makes a million-item search affordable and leaves a ledger of the cuts.
How it works¶
- Order the screens cheapest-first. Put the highest-throughput, lowest-cost test at the mouth of the funnel and the most expensive, highest-fidelity test at the neck.
- Size each stage to the next. Set each stage's pass threshold so the batch it forwards matches the capacity (and budget) of the stage that receives it — the funnel is balanced, not just narrowing.
- Track conversion. Record entries, advances, and drop reasons at every stage — the coverage ledger that makes the pipeline inspectable and its thresholds tunable.
- Stop feeding a stage when its budget is spent or the survivor slate is small enough that the remaining stages can absorb everyone left.
Tuning parameters¶
- Stage ordering — which test comes first. Front-loading the most discriminating cheap test removes the most volume early; a poorly ordered funnel pays deep-stage costs on candidates a later-but-cheaper test would have killed.
- Per-stage pass threshold — strict early cuts protect downstream budget but raise the false-negative rate; loose cuts preserve recall but overwhelm the neck.
- Number of stages — more stages give a smoother cost ramp and more chances to catch value, but each transition adds handoff overhead and another place to lose a true positive.
- Batch size vs downstream capacity — how large a cohort each stage forwards. Oversized batches stall the expensive stages; undersized ones leave costly capacity idle.
When it helps, and when it misleads¶
Its strength is making a very large search economically possible: cheap tests do the bulk removal, expensive tests see only a curated few, and total cost stays governed by an explicit budget rather than by how many candidates showed up. The conversion ledger turns the pipeline into something that can be audited and improved over time.
Its failure modes come from the early screen doing too much load-bearing work. A cheap first assay has real false-negative rate — a genuine hit killed by an assay artifact never reaches the confirming stages — and because the funnel's economics reward volume removal, thresholds tend to drift toward whatever the cheap test measures rather than what actually matters. That drift is Goodhart's law: once the early proxy becomes the gate, it stops tracking the thing it stood in for.[n1] The classic misuse is optimizing stage-to-stage conversion rates as if they were the goal, which quietly rewards a funnel that passes easy-to-score candidates over one that passes valuable ones. The discipline is to re-test a sample of rejects against the final criterion, calibrate the early screen against known-good exemplars, and judge the funnel by end-of-pipe yield, never by its middle-stage conversion metrics.
How it implements the components¶
promising_region_filter— each stage's pass/fail screen is a filter that marks which candidates earn the next, costlier evaluation.evaluation_budget— the funnel's defining act is sizing each stage's throughput against a cost budget, so scarce deep-evaluation capacity is never overrun.stop_rule— feeding a stage stops when its budget is exhausted or the survivor set is small enough for the remaining stages to take everyone.coverage_record— the entry/advance/drop-reason ledger documents the fate of every candidate for later audit and threshold tuning.
It supplies no machinery for deepening an individual candidate's analysis (refinement_step in the design sense) and no built-in re-opening of a rejected candidate (false_negative_check, backtracking_path) — see Design Downselection and Diagnostic Narrowing.
Related¶
- Instantiates: Coarse-to-Fine Search — the staged, budget-balanced many-to-few pipeline.
- Sibling mechanisms: Multi-Resolution Search · Coarse Grid Search · Diagnostic Narrowing · Portfolio Screening · Progressive Candidate Review · Design Downselection · Search Tree Pruning with Refinement
Editorial Notes¶
Form Classification¶
Form family: Protocol, Workflow & Routine
Rationale: Funnel Process operates as a repeatable ordered procedure or handoff sequence that coordinates action because it implements the archetype in review or product workflows by moving many candidates through cheap early screens before detailed evaluation of a smaller set.
Independent corroboration: The frozen evidence defines Funnel Process as 'Implements the archetype in review or product workflows by moving many candidates through cheap early screens before detailed evaluation of a smaller set', so its operative form is Protocol, Workflow & Routine.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Organizational & Management Science
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Staged screening funnels are longstanding organizational selection and process-design devices.
Related originating lineages:
- Operations Research — Sequential screening and coarse-to-fine search formally optimize evaluation cost across stages.
Review resolution: Both reviewers agree that organizational_management is primary: Staged screening funnels are longstanding organizational selection and process-design devices. I retain operations_research only where the reviewers identify formative lineage, not every later application. I resolve origin_mode as convergent because materially independent traditions developed the same operational structure. I resolve domain_reach as multi_domain because it transfers across several fields but is not a domain-free primitive. Encyclopedia synthesis is false because the exact generalized packaging is already established enough that encyclopedia-specific synthesis is not required.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Goodhart's law — "when a measure becomes a target, it ceases to be a good measure." In a funnel, a cheap early screen chosen as a proxy for final value becomes the gate candidates are optimized (or selected) against, so it drifts away from the value it was meant to predict; this is the mechanism behind proxy-drift failures. ↩