Skip to content

Two-Pass Generation and Evaluation

Mode-separation protocol — instantiates Divergence-Convergence Cycle Orchestration

Splits one working session into two ordered passes — generate with all judgment suspended, then switch modes and evaluate — so criticism never strangles ideas at the moment they're being formed.

The smallest possible instance of the whole archetype is a rule you can drop onto a single meeting: do it in two passes, never one. Two-Pass Generation and Evaluation separates the generative and evaluative modes in time — the first pass produces options under a contract that suspends all judgment, and only after an explicit switch does the second pass bring criteria to bear and cut. Its defining move is the deferral: in a mixed session the senior person's early frown kills a risky idea before it is even fully said, so the protocol forbids evaluation until generation is closed. It structures nothing about what to explore and gathers no evidence; it does exactly one thing, which is to keep criticism from strangling ideas at the instant they are most fragile. That minimalism is the point — it is the cheapest way to honour the archetype's core invariant that the two modes stay distinguishable.

Example

A magazine runs its weekly story-pitch meeting as two passes. In the first, everyone throws out story ideas and the single enforced rule is that no one evaluates — no "we did that last year," no editor's raised eyebrow, no feasibility objections. Half-formed and risky pitches get to reach the table intact, because the contract protects them. Then the editor marks the switch out loud — "the list is closed; we're now filtering" — and the second pass begins. Now the same room applies the criteria it agreed in advance: audience fit, reporting feasibility, freshness. The mechanism is nothing more than that wall between the two passes, but it changes the output: the ideas that would have died under a premature frown survive long enough to be judged on their merits rather than on their first impression.

How it works

  • Open with a suspend-judgment contract. The first pass is declared generation-only; criticism, ranking, and feasibility objections are explicitly out of bounds until it closes.
  • Mark the switch. A single, explicit transition closes generation and opens evaluation, so no one is ever half-generating and half-judging.
  • Evaluate in the second pass against pre-set criteria. Only after the switch do the options meet the criteria — ideally ones agreed before the first pass, so they aren't bent around whatever favourite emerged.
  • Keep it portable. The protocol carries no apparatus; it is a rule imposed on an existing session, document, or conversation, which is why it can go anywhere.

Tuning parameters

The dials that adapt this protocol to a specific session:

  • Separation strength — how firmly judgment is suspended in the first pass. Strict silence-the-critic protects novelty best but frustrates people itching to evaluate; a softer version is easier to run but leaks criticism back in.
  • Switch explicitness — how ceremonially the mode change is marked; a clear, announced switch is what stops the two passes from bleeding together.
  • Same-people versus split passes — whether the generators and evaluators are one group or deliberately different people, the latter trading familiarity for independence.
  • First-pass stopping rule — whether generation ends on a count, on saturation, or on the facilitator's call before the switch.
  • Second-pass strictness — how hard the evaluative cut is, from a gentle shortlist to a decisive down-to-one.

When it helps, and when it misleads

Its strength is protection of novelty at almost no cost: it is the lowest-overhead way to stop evaluation from suppressing generation, and because it needs no special roles, tools, or time budget, it can be imposed on nearly any working session. It is the right tool when the risk is simply that good ideas die too young.

Its limits follow from its minimalism. The protocol guards only the boundary between the two modes within one activity; it does nothing to ensure the breadth it protects is meaningful, so a team can two-pass its way to a wall of cosmetic variants and feel disciplined about it.[1] It gathers no evidence and preserves no reserves. Its characteristic failure is a fake contract — the suspend-judgment pass in which the senior voice still signals disapproval by tone, so the separation is theatre — or a second pass whose criteria are invented after a favourite has appeared. The discipline that keeps it honest is to enforce the contract for real, mark the switch unmistakably, and fix the evaluation criteria before the first pass can produce a front-runner.

How it implements the components

Two-Pass Generation and Evaluation fills only the mode-boundary components — the interface between divergence and convergence — and nothing else:

  • divergence_phase_contract — the first pass's suspend-all-judgment rule is the contract; it is what the protocol exists to enforce.
  • phase_transition_gate — the explicit switch from the generation pass to the evaluation pass is the lightweight gate between the two modes.

It does not structure meaningful breadth across dimensions — that is Parallel Concept Sprints — nor does it bound the work in time, which is Timeboxed Ideation and Downselection. Its gate is a single cognitive mode-switch, not the heavyweight go/kill investment checkpoint of Stage-Gate Concept Review.

References

[1] Separating idea generation from idea evaluation — deferring judgment so that criticism does not run concurrently with ideation — is the foundational rule of brainstorming as set out by Alex Osborn. Two-Pass Generation and Evaluation is that rule reduced to its portable minimum: one contract, one switch. ↩