Research Stopping Rule¶

Inquiry-stopping rule — instantiates Regress Termination Rule

Sets in advance when data collection, literature search, or analysis has done enough for its purpose — by information saturation, a preset threshold, or a resource limit — so inquiry ends before it either never closes or is cut off arbitrarily.

There is always one more database to search, one more subject to enrol, one more analysis to run — so the gathering itself can regress without end. A Research Stopping Rule ends that by fixing, ahead of time, what will count as enough collecting for this purpose: typically the point where new data stops changing the answer (saturation), backed by a resource or time cap so the search cannot run forever even if saturation is slow. What makes it THIS mechanism is that it governs the process of inquiry, not the adequacy of the collected evidence or the readiness of any downstream decision — and that it is committed before results are seen, which is exactly what stops the rule from bending to a convenient answer.

Example¶

A team runs a systematic review on whether a dietary supplement reduces a chronic symptom. Left open, the literature search would never end. So they pre-register the stopping rule before screening a single study: keep adding databases and search terms until two successive rounds turn up no new study that shifts the pooled estimate by more than a preset margin — saturation — with a hard backstop of a fixed search budget in reviewer-weeks, whichever comes first. When the last two rounds return only already-included trials, the search stops; the unindexed grey literature they didn't have budget to chase is recorded as a known gap rather than pursued indefinitely. Because the rule was set in advance, no one can quietly stop the moment the estimate looks favorable — the terminus was chosen before anyone knew which way it would point.

How it works¶

Define "enough" as diminishing returns. State the saturation condition: how little new information across how many successive rounds means further collection isn't worth it for this purpose.
Attach a resource backstop. Set a time, budget, or sample cap that bounds the search even if saturation is slow — a second, independent brake.
Pre-commit before results. Fix the rule before outcomes are visible, so the stopping point is a decision about method, not about the answer.
Record what was left uncollected. Log the sources or subjects the rule excluded, so the gap is visible to whoever uses the result.

Tuning parameters¶

Saturation threshold — how little new information triggers the stop. A tight threshold (almost nothing new) is thorough but slow; a loose one is fast but risks stopping before the picture stabilizes.
Resource-cap size — how large the backstop budget is. A generous cap lets saturation govern; a tight one may force a stop before saturation, biasing what got collected.
Pre-commitment strictness — fully pre-registered versus adaptively adjustable. Rigid pre-registration resists motivated stopping; adaptivity handles surprises but reopens the door to bias.
Purpose calibration — what the inquiry is for. A quick scoping review can stop early; a safety-critical synthesis demands near-exhaustion before the rule fires.

When it helps, and when it misleads¶

Its strength is killing two opposite pathologies at once: the endless search that never yields a conclusion, and the arbitrary cut-off that stops whenever someone tires. By making the terminus a pre-committed rule, it lets inquiry close on principle.

Its signature hazard is optional stopping — halting collection or analysis at the moment the result looks best, which inflates false findings and is a live methodological problem wherever data accumulate over time.^[1] The mirror failure is a resource cap that bites before saturation, so the sample is truncated in a way that skews the answer while looking rule-governed. And the rule is run backwards when it is written (or revised) after peeking at interim results. The discipline that keeps it honest is to pre-register the rule before outcomes are visible, keep the saturation criterion and the resource backstop distinct, and disclose the residual gap rather than presenting a truncated search as exhaustive.

How it implements the components¶

sufficiency_threshold — it calibrates how little marginal information counts as enough for the inquiry's purpose, tightening or loosening the saturation bar.
stopping_criterion — the saturation-or-cap condition is the operational rule that halts collection.
timebox_or_resource_limit — the budget/time backstop bounds the search independently of saturation, so inquiry cannot run indefinitely.

It does not grade the quality and independence of the evidence once collected (regress_chain / evidentiary sufficiency_threshold → Evidence Sufficiency Rubric), gate the downstream decision (stopping_criterion for the decision → Decision Closure Criteria), or name the residual doubts left after closure (uncertainty_residue → First-Principles Statement).

Instantiates: Regress Termination Rule — it terminates the collection regress at a pre-committed point.
Sibling mechanisms: Evidence Sufficiency Rubric · Decision Closure Criteria · Timeboxed Inquiry Record · Burden-of-Proof Rule · Axiom Set · First-Principles Statement · Five Whys with Stop Rule · Governance Authority Chain · Review Trigger Register · Assumption Log

Notes¶

Two nearby siblings are distinct in what drives the stop. A Timeboxed Inquiry Record is driven primarily by a deadline and exists to document what was reached when time ran out; a Research Stopping Rule is driven primarily by information saturation, with the resource cap only a backstop. Reaching for the timebox when saturation is the real question — or vice versa — stops inquiry for the wrong reason.

References¶

[1] In statistics, optional stopping — continuing to collect or analyze data and halting when the result crosses a desired line — inflates false-positive rates; pre-registering the stopping rule before seeing outcomes is the standard corrective, which is why pre-commitment leads the method here. ↩