Stage Gate Process¶
Core Idea¶
A stage-gate process is a designed sequence in which a long-horizon commitment is partitioned into stages separated by go/no-go decision gates, with each gate requiring evidence accumulated in the prior stage and licensing escalated resource commitment for the next. The structural commitment is a triple: a sequence of stages, each producing a body of evidence about whether the underlying bet is working; gates at the stage boundaries at which an evidence-conditional go/no-go decision is taken; and escalating commitment, where each subsequent stage costs more, locks in more, and demands more decisive evidence to justify. The pattern expects most candidates to die at gates; the funnel is a feature, not a bug.
What stage-gate makes visible is the separation of evidence generation from commitment escalation. The actor never has to commit the full project budget on initial-stage evidence, because the gate structure permits incremental commitment proportional to accumulated evidence. The total budget is decomposed into a sum of stage-conditional commitments rather than a single up-front bet, and the project decision is decomposed into a sequence of gate decisions. The intervention space is correspondingly rich: gate criteria, kill criteria, per-stage evidence requirements, advance-commitment levels, gate-keeper composition, and the option to revert a candidate to an earlier stage rather than killing it outright. The pattern's force depends on two structural facts that distinguish a real gate from a check-in: each gate has explicit kill criteria, not just go criteria, and the gate-keepers are independent of the stage executors. An instance lacking either is gate-shaped but missing the structural force.
How would you explain it like I'm…
Checkpoint Gates
Go Or Stop Gates
Evidence Gated Funnel
Structural Signature¶
the long-horizon commitment to be partitioned — the ordered sequence of evidence-producing stages — the gate as evidence-conditional go/no-go boundary — the escalating-commitment ladder — the kill-criterion alongside the go-criterion — the executor/adjudicator independence invariant — the attrition funnel as expected output
A configuration exhibits the stage-gate pattern when each of the following holds:
- A partitionable commitment. There is a single high-uncertainty, long-horizon bet whose total cost can be decomposed into a sequence of smaller, conditionally-incurred commitments rather than a lump up-front outlay.
- An ordered sequence of stages. Each stage is a bounded segment of work whose product is a defined body of evidence about whether the underlying bet is working. Stages are ordered, and evidence accumulates across them.
- Gates at the stage boundaries. At each boundary an explicit decision is taken, conditioned on the evidence produced so far, that resolves to advance, terminate, or (optionally) revert. A gate is a decision moment, not a check-in.
- Escalating commitment. Each successive stage costs more, locks in more, and demands stronger evidence to justify; the commitment authorised by a gate-pass must be matched to the strength of the evidence behind it.
- Paired go and kill criteria. A genuine gate carries explicit thresholds for termination, not only for advancement; "no-go" is a designed, normal outcome. Absent kill criteria, the gate degenerates into a rubber-stamp.
- Adjudicator–executor independence. The party deciding whether to advance is structurally distinct from the party whose work is being judged; without this separation the go/no-go decision loses its honesty.
- An attrition funnel. Most candidates are expected to die at gates; high attrition is the signature of a well-tuned selection mechanism, not a string of failures.
Composed, these turn one irreducible project decision into a chain of evidence-conditional gate decisions, and a single bet into a sum of staged, revocable commitments whose funnel of kills is the intended product.
What It Is Not¶
- Not a
decisiontaken at one moment. Adecisionresolves an alternative at a point; stage-gate is a sequence of evidence-conditional decisions with escalating commitment between them, so the unit of analysis is the chain, not the gate. - Not
escalation_of_commitment. That pathology is the failure of a stage-gate — sunk cost biasing the gate toward "go"; the prime's kill criteria and adjudicator independence exist precisely to resist it, so the two are antagonists, not synonyms. - Not
precedent_stare_decisis. Precedent binds a present decision to a past one by analogy; a gate conditions advancement on freshly produced evidence about this bet, with no claim that prior cases govern. - Not
prioritization. Prioritisation ranks candidates by value at one time; stage-gate sequences and kills them across time as evidence accumulates, and high attrition is its intended output, not a ranking artefact. - Not
local_autonomy_tiered_escalation. That routes a decision upward by severity; a gate is a forward checkpoint that the same candidate must clear to proceed, independent of any authority hierarchy. - Not mere
monitoring. Monitoring observes without a designed branch point; a gate is a decision moment with paired go/kill criteria — a check-in that cannot terminate the candidate is gate-shaped but lacks the structural force. - Common misclassification. Calling any phased plan with review meetings "stage-gate." If the reviews lack explicit kill criteria and the reviewers are the executors themselves, the gates are rubber-stamps; the diagnostic is whether any candidate has actually been killed at a late gate.
Broad Use¶
- New product development (canonical): staged sequences of discovery, scoping, business case, development, testing, and launch, separated by idea screens and go-to-development, go-to-testing, and go-to-launch gates.
- Pharmaceutical development: Phase 0 through Phase IV, with evidence packages required between phases and heavy attrition between early phases and approval.
- Military acquisition: milestone decision points (technology development, engineering and manufacturing, production) gated by review boards with evidence packages.
- Venture capital: pre-seed through later rounds, each committing more capital, requiring more evidence, and applying explicit gate criteria.
- Software releases: alpha, beta, release candidate, and general availability, each with explicit exit criteria and an escalation in user exposure.
- Academic credentialing: coursework, qualifying exam, proposal defence, dissertation defence, and publication, each gate licensing escalated commitment of advisor and committee resources.
- Real estate development: site identification, feasibility, entitlement, design, construction, and lease-up, each escalating capital and requiring evidence (zoning, permits, pre-leasing).
- Aerospace systems engineering: the design-review sequence, each gate requiring evidence packages and licensing escalated commitment.
In each substrate the same triple recurs — sequenced stages producing evidence, gates resolving evidence into go/no-go, and escalating commitment — independently rediscovered because the underlying decision-economics recur.
Clarity¶
Naming a process as stage-gate exposes three things non-staged commitment hides. The advance commitment is not all-or-nothing at project start; it is sliced into incremental commitments, each licensed by accumulated evidence. The option to kill is built into the structure — gates are decision moments, not check-ins, and "no-go" is a normal outcome rather than a failure of the stage. And the evidence requirements are stage-specific: an early stage needs feasibility or safety evidence, a late stage needs efficacy evidence at scale, and mismatching evidence to stage — demanding scale-level proof at the first gate — is a common pathology.
The pattern also makes attrition explicit and acceptable. A funnel in which nine of ten candidates die at an intermediate gate is not broken; it is working as designed. This reframing is itself clarifying: it converts what looks, under a non-staged frame, like a string of failures into the expected operation of a well-designed selection mechanism, and it redirects attention from "why do so many die?" to "are the gates killing the right candidates at the right stages?"
Manages Complexity¶
Stage-gate compresses an open-ended, high-uncertainty commitment problem into a tractable sequence of design choices: choose the stage scope, define the stage exit criteria, define the gate evidence package, and define the commitment escalation per gate-pass. The decision problem becomes a sequence of gate decisions rather than a single irreducible project decision, and the budget becomes a sum of conditional commitments rather than a lump bet.
It also organises role specialisation. Stage executors and gate adjudicators are different roles — and should be, since gate-keepers who also execute stages carry a conflict of interest — and evidence assemblers differ from evaluators. The gate structure thereby provides natural handoff points and review rituals, distributing a complex commitment across clearly separated responsibilities. This separation is not incidental decoration: it is what keeps the go/no-go decision honest, because the person deciding whether to advance is not the person whose work is being judged.
Abstract Reasoning¶
The frame supports reasoning about commitment under uncertainty, attrition, and option value. Information-commitment matching: at each gate the commitment for the next stage should be matched to the strength of evidence from the prior stage, and over-committing on weak evidence — pouring late-stage budget into a candidate with a weak earlier signal — is a classic failure. Kill-criterion design: gates need explicit thresholds below which a candidate is killed; without them, gates degenerate into rubber-stamps. Option value: the staged structure is equivalent to a sequence of real options, and the value of preserving optionality grows with downstream uncertainty. Stage-skipping and reverting: strong early evidence can support skipping a stage, and a late-stage failure can support reverting rather than killing, so the graph is not strictly linear. And a characteristic catalogue of failure modes falls out of the structure: bureaucratised gates that become compliance checks, evidence-quality drift where stages produce evidence shaped to pass rather than to inform, kill-avoidance where no candidate ever fails, gate-explosion that produces paralysis, and the urgency-override that skips a gate without evidence.
Knowledge Transfer¶
The roles map across substrates: the stage is a bounded segment of work producing a defined evidence package; the gate is the evidence-conditional go/no-go at the boundary; the escalation is the increase in committed capital, exposure, or lock-in that each gate-pass authorises; the go and kill criteria are the conditions for advance and termination; the gate-keeper is the decision-maker structurally independent of the executors; and the attrition profile and optionality value are the portfolio-level quantities the structure exposes. The intervention vocabulary ports surgically, in part because the pattern has been independently codified in multiple high-stakes domains, each refining a different aspect — one domain best at evidence packages, another at round economics, another at gate-keeper independence, another at gate-process design.
Concrete transfers are well attested. Gate-criteria design ports from venture-capital term sheets to regulatory evidence packages to acquisition milestone reviews, each substrate having learned conventions for adequate per-stage evidence. Kill-criterion discipline ports as a single diagnostic question — "what would kill this candidate?" — applicable to any staged investment. Information-commitment matching ports as a portfolio rule: commit proportional to evidence quality. Gate-keeper independence ports as an organisational-design rule: the people who decide to advance should not be the people whose work is judged. Attrition-as-design ports as a portfolio frame in which high attrition is the signature of a well-designed funnel. And real- option valuation ports from finance into staged R&D budgeting, connecting formal valuation theory to organisational practice. A single worked instance shows the substance: an oncology drug advancing through microdosing, single-arm safety, mid-stage efficacy, and large-scale efficacy gates, each producing evidence that licenses the next-stage commitment, each with explicit go and kill criteria, the total commitment sliced rather than committed up front, and roughly one in ten Phase-I candidates reaching approval — by design, not by failure. The same structure, with the same intervention vocabulary, governs a venture portfolio, an aerospace design review, and a doctoral programme.
Examples¶
Formal/abstract¶
Sequential hypothesis testing as a multi-stage decision problem instantiates the stage-gate structure formally. Consider a candidate treatment whose true effect \(\theta\) is unknown. The partitionable commitment is the full trial-and-approval budget; the ordered stages are a sequence of experiments of increasing size \(n_1 < n_2 < n_3\), each producing an evidence statistic \(T_k\) (an estimated effect with shrinking confidence interval). At each gate, a decision rule compares \(T_k\) against two thresholds: a go boundary above which the candidate advances and an explicit kill boundary below which it is terminated — the classical Wald sequential probability ratio test structure, where between the boundaries you keep sampling and outside them you stop. The escalating commitment is exact: each stage's sample size, and hence cost, grows, so the evidence bar to license the next stage rises with it. The attrition funnel is the prior-weighted fraction of candidates whose \(T_k\) falls below the kill boundary at each gate — and the design optimises expected total cost by killing weak candidates cheaply and early. The adjudicator independence invariant maps to a pre-registered, blinded analysis: whoever ran the experiment cannot move the boundary after seeing \(T_k\). This shows the intervention space directly — set kill thresholds to control false-advance rate, set stage sizes to match commitment to evidence quality, and read the attrition profile as a tuning diagnostic rather than a failure log.
Mapped back: The sequential-test gates, go/kill boundaries, escalating sample-size commitment, and pre-registration independence are the prime's stages, paired criteria, escalation ladder, and adjudicator-independence invariant rendered as a formal decision rule.
Applied/industry¶
Venture-capital financing rounds run the stage-gate pattern at portfolio scale. The partitionable commitment is the total capital a startup would need to reach profitability; instead of writing one cheque, investors slice it into pre-seed, seed, Series A, B, and C stages, each a bounded segment of company-building that produces evidence — a working prototype, then early users, then revenue retention, then unit economics, then scalable growth. Each gate is a financing decision conditioned on the prior stage's metrics, and the escalation is steep: a Series B round commits far more capital and demands far stronger evidence than a seed round. The kill criterion is the down-round-or-no-round outcome, a designed and normal event — most seed-funded companies never raise a Series A, and that high attrition funnel is the signature of a working selection mechanism, not a string of disasters. The adjudicator–executor independence invariant appears as the investment committee and new lead investors who are structurally distinct from the founders being judged. The pattern diagnoses two classic pathologies: over- committing on weak evidence (a hot seed market pouring growth-stage money into companies with no retention signal) and kill-avoidance (insiders bridging a failing company round after round rather than letting the gate do its work). A parallel applied instance is software release management: alpha, beta, release candidate, and general availability are stages with explicit exit criteria, the escalation is user-exposure rather than capital, and a feature killed at beta is the gate operating as designed.
Mapped back: Financing rounds and release tiers are stages; the funding decision and the GA cutover are gates with paired go/kill criteria; the rising cheque size and widening user exposure are the escalation ladder; and the high fraction of startups and features that die between stages is the intended attrition funnel.
Structural Tensions¶
T1 — Evidence Generation versus Commitment Escalation (scopal). The structure separates producing evidence from committing resources, but the two cannot be fully decoupled: late-stage evidence is often only generable by making the late-stage commitment (you learn whether a drug scales by running the scale trial). The prime's clean "commit proportional to evidence" rule stops being the whole story precisely when evidence is endogenous to commitment. Failure mode: gating on evidence that the prior stage structurally cannot produce, freezing genuinely promising candidates at a gate that demands proof only the next stage could yield. Diagnostic: ask whether the gate's evidence requirement is producible within the prior stage's budget, or whether it presupposes the very commitment it gates.
T2 — Kill Discipline versus Sunk-Cost Drift (temporal). Kill criteria are set ex ante, but they are applied after resources, reputations, and careers have been invested — and the accumulated stake biases the adjudicator toward "go." The gate's honesty decays over time precisely as the candidate progresses, because each surviving stage raises the cost of killing. Failure mode: kill-avoidance, in which no candidate ever fails a late gate because terminating it would indict prior gate decisions — the funnel silently flattens into a conveyor. Diagnostic: check whether any candidate has actually been killed at the late gates, or only at the cheap early ones; an all-pass late funnel is the signature.
T3 — Adjudicator Independence versus Adjudicator Competence (coupling). The prime demands that gate-keepers be structurally distinct from executors to keep the decision honest — but the people most competent to judge the evidence are often the executors themselves, who hold the domain knowledge. Independence and competence pull in opposite directions. Failure mode: an independent gate populated by adjudicators who cannot actually read the evidence, so the gate degenerates into a procedural rubber-stamp that looks honest while rubber- stamping. Diagnostic: ask whether the gate-keepers could articulate the specific technical reason a candidate should be killed, or only whether the paperwork is complete.
T4 — Funnel Attrition versus Pipeline Starvation (scalar, local vs global). High attrition is locally healthy — the signature of a well-tuned selection mechanism — but globally, a too-aggressive funnel can starve the portfolio of survivors, killing candidates that would have succeeded given more stages. The gate that optimises per-candidate evidence quality can degrade portfolio yield. Failure mode: tuning kill thresholds to maximise the purity of survivors while the absolute number of approved outputs collapses below what the enterprise needs. Diagnostic: monitor not just the kill rate but the end-of-funnel throughput; rising attrition with falling final yield signals over-selection, not discipline.
T5 — Sequential Gates versus Irreversible Lead Time (temporal/coupling). The staged structure assumes commitment can be deferred until evidence licenses it — but some commitments have lead times longer than the stage that would justify them, forcing pre-commitment before the gate. Where long-lead procurement, hiring, or capacity must begin before its gate, the gate becomes a formality ratifying a decision already made. Failure mode: the urgency-override, where a gate is skipped or pre-empted "because we had to order the long-lead item anyway," quietly converting a real gate into theatre. Diagnostic: identify any stage whose enabling commitments must begin before the prior gate closes — those gates are structurally compromised.
T6 — Evidence Fidelity versus Gameable Metrics (measurement). Gates condition on evidence, but evidence is produced by the executors who benefit from passing — so the measured evidence drifts toward what passes the gate rather than what informs the decision. The prime's reliance on evidence-conditional gating presumes the evidence is honest; when the metric becomes a target, it ceases to measure. Failure mode: evidence-quality drift, where stages optimise the gate statistic (enrolment numbers, demo polish, vanity metrics) while the underlying bet's true signal is unmeasured. Diagnostic: ask whether the gate criterion has remained a leading indicator of downstream success, or whether candidates that sailed through later failed in the field — a sign the metric has decoupled from reality.
Structural–Framed Character¶
Stage-gate sits on the framed side of the structural–framed spectrum. Underneath there is a genuine relational skeleton — a long bet decomposed into a sequence of evidence-conditional branch points with escalating commitment — but the prime is bound to organisational decision-making tightly enough that invoking it imports a managerial frame rather than merely recognising a pattern already wired into a physical system.
Two diagnostics drive the grade hardest. Institutional origin and human-practice-bound both read fully framed: every native instance — FDA Phase 0–IV packages, DAB milestone reviews, venture financing rounds, software release tiers, doctoral qualifying exams — presupposes an institution that codifies stages, convenes gate-keepers, and authorises budget. The load-bearing structural facts the Core Idea insists on, paired kill criteria and adjudicator–executor independence, are not features of any physical substrate; they are governance arrangements that only exist where there are roles, parties, and decision rights. There is no thermostat or chemical network that "kills a candidate at a gate." The remaining diagnostics sit at the middle: the vocabulary travels only partly — "stage," "gate," "go/no-go," "kill criterion," "funnel" carry a project-management home lexicon that each domain must adopt rather than retell freely — and invoking the prime half-imports an evaluative posture (high attrition is good, kill-avoidance is bad), so it is not value-neutral until you specify what it does.
The honest concession is that the relational core is real, and the formal Wald sequential-test example shows the gate/escalation/attrition triple can be written as a pure decision rule. But that formal rendering is the exception that proves the framing: it must strip away the institutional machinery (committees, evidence packages, role separation) that constitutes stage-gate in every working instance. With institutional origin and human-practice-bound both pinned high and the rest at the middle, the aggregate sits firmly framed, matching the assigned grade.
Substrate Independence¶
Stage-gate is a strongly substrate-independent prime within its band — composite 4 / 5 on the substrate-independence scale. Its domain breadth is wide (4 / 5): the same triple of evidence-producing stages, evidence-conditional gates, and an escalating-commitment ladder was independently rediscovered in pharmaceutical development (Phase 0–IV with attrition between phases), military acquisition (milestone decision boards), venture capital (pre-seed through Series C financing rounds), software release management (alpha, beta, release candidate, general availability), academic credentialing (qualifying exam, proposal defence, dissertation defence), and real-estate development (feasibility, entitlement, construction, lease-up). The structural abstraction is real but not maximal (3 / 5): the relational skeleton — slice a long bet into revocable, evidence-licensed increments — is medium-neutral, yet the load-bearing facts the Core Idea insists on (paired kill criteria, adjudicator–executor independence) are governance arrangements that presuppose roles, parties, and decision rights, so no thermostat or chemical network "kills a candidate at a gate." That ceiling is what holds the composite below the top: every working instance is a human project. What lifts it to a 4 is exceptionally strong transfer evidence (5 / 5) — the pattern has been formally codified in domain after domain, each refining a different facet (evidence packages, round economics, gate-keeper independence), with the Wald sequential-test rule showing the gate/escalation/attrition triple as a pure decision rule and real-option valuation porting from finance into staged R&D budgeting.
- Composite substrate independence — 4 / 5
- Domain breadth — 4 / 5
- Structural abstraction — 3 / 5
- Transfer evidence — 5 / 5
Relationships to Other Primes¶
Parents (1) — more general patterns this builds on
-
Stage Gate Process is a kind of, typical Sequencing
Stage-gate is a sequence of precedence-ordered, evidence-conditional stages with escalating commitment — a specialization of sequencing (deliberate ordering of steps under precedence). The gate-decision chain is the load-bearing structure; sequencing is the genus.
Children (1) — more specific cases that build on this
-
Decision decompose Stage Gate Process
Each gate is a
decision(resolve advance/terminate/revert); the prime is the structured sequence of gate decisions. The file: a decision is the 'building block' of which stage-gate is a sequence.
Path to root: Stage Gate Process → Sequencing → Dependency
Neighborhood in Abstraction Space¶
Stage Gate Process sits among the more crowded primes in the catalog (6th percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.
Family — Staged Processes & Drift (32 primes)
Nearest neighbors
- Funnel Analysis — 0.80
- Evidence-Latency Window — 0.76
- Optimal Stopping Rule — 0.76
- Decision — 0.75
- Decision Cycle Subordination — 0.74
Computed from structural-signature embeddings · 2026-06-14
Not to Be Confused With¶
The sharpest confusion is with escalation_of_commitment. The two share
the vocabulary of staged investment and sunk cost, but they are structural
opposites in their stance toward continuation. escalation_of_commitment is
the pathology by which prior investment biases a decision-maker toward
throwing good resources after bad — the more committed, the harder to quit.
Stage-gate is the designed defence against exactly that bias: it inserts
gates with explicit kill criteria and adjudicators independent of the
executors, so that continuation must be re-earned on fresh evidence rather
than inherited from past spending. The relationship is therefore adversarial,
not equivalent — and the clean diagnostic is that a stage-gate process which
exhibits escalation of commitment (no candidate ever killed at a late gate)
is a stage-gate that has failed at its core purpose. A practitioner who
conflates them will treat the funnel's high attrition as the disease when it
is the cure.
A second confusion is with prioritization. Both winnow a population of
candidates, but prioritisation operates at a single time-slice — it ranks
available options by value or urgency and allocates scarce resource to the
top of the list. Stage-gate operates across time and commitment levels: a
candidate is not ranked against peers but advanced or killed against an
absolute evidence threshold appropriate to its stage, with the commitment
ladder rising at each pass. A prioritised list can be re-ranked freely as
values shift; a stage-gate candidate that fails a kill criterion is removed,
not demoted. The distinction matters because the interventions differ — to
fix a prioritisation failure you change the ranking function, whereas to fix
a stage-gate failure you change the gate criteria, the kill thresholds, or
the adjudicator's independence.
Finally, stage-gate is distinct from decision simpliciter. A decision
is the atomic act of resolving among alternatives at a point; it is the
building block of which a stage-gate process is a structured sequence. The
prime's contribution is everything that surrounds the individual gate
decisions — the ordered stages that produce the evidence each decision
conditions on, the escalating commitment that makes later decisions costlier,
and the attrition funnel that is the sequence's intended product. Treating a
stage-gate as a single big decision collapses precisely the structure
(decomposition into revocable, evidence-licensed increments) that gives the
prime its force, and forfeits the option value the staged form preserves.
Solution Archetypes¶
No catalogued solution archetypes reference this prime yet.