Research Continuation Gate¶
Review gate — instantiates Marginal Stop Rule
A review gate that decides whether another experiment, pilot, or refinement cycle is worth running.
A Research Continuation Gate is a checkpoint that decides whether to run one more experiment, pilot, or refinement cycle — and its defining move is that it values the next cycle by its expected information, not its expected result. A cycle is worth running not because it might succeed but because it would meaningfully narrow what the team doesn't yet know, enough to justify its cost and the more informative experiment it displaces. The gate is built for deep uncertainty: it asks how much the next cycle would move the team's beliefs, discounts that by how noisy the read will be, and compares it to the best alternative use of the same rig time. When another cycle would mostly repeat what is already known, the gate stops the line.
Example¶
A materials lab has run a dozen synthesis-and-test cycles chasing a higher-conductivity battery electrolyte. The early cycles were richly informative — each ruled out whole families of formulations. Now the gains are small and the measurements noisy: the last three cycles nudged conductivity within the error bars and didn't clearly separate the two remaining hypotheses. The team wants to run cycle thirteen "to be sure," and the equipment is already booked.
The Research Continuation Gate reframes the decision around information. What would cycle thirteen actually resolve? Given the measurement noise, it probably wouldn't discriminate between the surviving hypotheses — its expected information gain is low and its uncertainty band wide. Meanwhile a differently designed experiment, using a sharper characterization method, would discriminate them, for comparable rig time. The gate's verdict is not "the electrolyte work failed" but "another cycle of this kind isn't worth the bench time." It stops the incremental line and frees the rig for the more informative experiment — the opportunity the grind was quietly consuming.
How it works¶
- Name the next cycle. The specific experiment, pilot, or refinement pass whose value is in question — not the research program as a whole.
- Estimate expected information gain. Ask how much the cycle would narrow the live uncertainty or discriminate between hypotheses, treating information — not a hoped-for positive result — as the benefit.
- Carry the uncertainty band. Weight that expected gain by how noisy and confounded the read will be, so a cycle that would produce only ambiguous data scores low.
- Compare to the best alternative experiment. Set the cycle against the most informative other use of the same equipment and time.
- Apply the gate. Run it, redesign it to yield more information, or stop the line — the threshold names the condition and the resulting action.
Tuning parameters¶
- Information bar — how much expected learning a cycle must promise to proceed. A high bar prevents low-yield grinding but risks stopping before a rare, decisive result.
- Cycle size — how much work each increment bundles. Small cycles gate often and cheaply; large ones commit big blocks of time on one gate decision.
- Uncertainty tolerance — how much measurement noise is acceptable before a cycle counts as uninformative, trading rigor against throughput.
- Alternative breadth — whether the opportunity-cost comparison spans the whole lab's queue or just the current line, balancing honesty against review overhead.
When it helps, and when it misleads¶
Its strength is stopping the low-information grind — the extra cycle run out of thoroughness or momentum rather than genuine need — and redirecting scarce experimental capacity toward the questions that would actually move. It is a practical reading of the value of information: a study is worth running only when the expected value of what it teaches exceeds its cost.[n1]
The characteristic failure is premature stopping before a threshold or nonlinear effect that only later cycles would reveal — the decisive result that lay just past where the gate cut — and its inverse, metric gaming, where a team reframes a low-information cycle as "confirmatory" to keep a favored line alive. The guarding discipline is to pre-specify what information each cycle is meant to produce and to hold the uncertainty band in view, so the gate stops on genuinely exhausted information rather than on a bad-luck run or a rhetorical relabel.
How it implements the components¶
marginal_benefit_estimate— recasts marginal benefit as expected information gain: how much the next cycle would narrow uncertainty, not whether it would succeed.uncertainty_band— noisy, confounded, lagged reads are made explicit and discount the expected gain, so ambiguous cycles score low.opportunity_cost_review— the cycle is judged against the most informative alternative experiment competing for the same rig time.stop_threshold— the gate condition and its consequence: run, redesign for more information, or stop the line.
It does not model the physical risk and reversible pull-back of escalation (risk_or_side_effect_estimate, reversibility_plan) — that is Training Volume Limit; nor does it redirect the freed equipment budget once a line is stopped (saved_resource_disposition) — that disposition is Budget Stop Rule.
Related¶
- Instantiates: Marginal Stop Rule — supplies the information-valued gate on whether the next experimental cycle is worth its cost and displaced alternative.
- Sibling mechanisms: Project Kill Criteria · Budget Stop Rule · Ad Spend Cap · Training Volume Limit · Treatment Escalation Limit · Repair-vs-Replace Decision · Sunset Review with Reauthorization
Editorial Notes¶
Form Classification¶
Form family: Decision, Gate & Allocation
Rationale: Research Continuation Gate operates as a case-specific gate, selection, routing, prioritization, or resource disposition because it a review gate that decides whether another experiment, pilot, or refinement cycle is worth running.
Independent corroboration: The frozen evidence defines Research Continuation Gate as 'A review gate that decides whether another experiment, pilot, or refinement cycle is worth running', so its operative form is Decision, Gate & Allocation.
Nearest alternative: Assessment, Review & Assurance — Research Continuation Gate includes features of a bounded evaluation of existing evidence or work that produces a finding or disposition, but its defining operation is a case-specific gate, selection, routing, prioritization, or resource disposition.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Deciding whether another experimental cycle has sufficient expected information value is rooted in research design.
Related originating lineages:
- Organizational & Management Science — Stage-gate portfolio management materially contributes resource and continuation decisions.
- Philosophy — Epistemology of inquiry informs whether remaining uncertainty warrants more investigation.
Review resolution: Both blind reviewers agree that statistics_experimental_design is the primary historical origin. Explicit reconciliation of alternate origin disagreement adopts reviewer_a's evidence: Deciding whether another experimental cycle has sufficient expected information value is rooted in research design. The selected record uses alternates=organizational_management, philosophy, origin_mode=cross_disciplinary_synthesis, and domain_reach=multi_domain; the other review proposed alternates=data_science, mathematics, origin_mode=cross_disciplinary_synthesis, and domain_reach=multi_domain. The selected combination better preserves the mechanism-specific formative lineages and calibrated scope; broader present-day use is not treated as proof of additional historical origin.
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¶
The gate values a cycle even when the expected result is negative, provided a clean negative would teach a lot — ruling a hypothesis out is information. This is why "will it work?" is the wrong question at the gate and "what will it settle?" is the right one.
[n1] Value of information — a decision-theoretic quantity capturing how much reducing uncertainty is worth before acting; an experiment is justified when its expected information value exceeds its cost. It is the natural currency for a gate deciding whether more study is worthwhile. ↩