Skip to content

Micro-Experiment Sequence

Experimental sequence — instantiates Sense-Act Loop Coupling

Chains many small, cheap tests into a branching series, carrying each result forward so every next experiment is chosen from what the previous ones revealed.

A Micro-Experiment Sequence chains many small, cheap tests into a branching series in which each result is carried forward and reframes the next test — so the sequence as a whole learns something no single experiment could. Its unit is not one probe but the chain: the value lives in how results accumulate in a running trace and how each outcome forks which experiment runs next. It runs the loop at a deliberate cadence, updating its model of what is now possible after each result settles. Crucially, it does not design or bound the individual probe — that job belongs to the Active Probe Protocol it consumes for each step. Its own contribution is the sequencing, the memory across steps, and the branching.

Example

A two-person startup cannot tell why sign-ups stall on their landing page. Instead of one big redesign, they run a sequence of micro-experiments, each cheap and reversible. Experiment 1: rewrite the headline; the result — click-through is flat, but scroll depth rises. Logged, that result reframes the next question: the problem isn't the hook, it's the ask. Experiment 2, chosen because of experiment 1: move the sign-up form above the fold; the result — sign-ups climb 20%, but people drop off at the email field. Experiment 3, branched from that: make the email field optional; the result — completion jumps.

Each experiment was selected from the accumulated trace of the ones before it. Run in a different order, or with the results forgotten between steps, the sequence would have learned far less — it would have been flailing that happened to include some A/B tests. What makes it a mechanism is that the chain has memory and the branches are chosen from it: a path through a design space too large to map up front, bought one cheap result at a time.

How it works

  • Delegate each step's design to a bounded probe. Every experiment is an Active Probe Protocol — the sequence assumes the single probe is already scoped and safe, and concerns itself with what comes after.
  • Accumulate results in a running trace. The sequence has memory: each outcome is recorded so later choices can draw on all the earlier ones.
  • Branch, don't batch. Each result reroutes which experiment runs next; the sequence is a tree grown from outcomes, not a fixed list run to completion.
  • Update what's now possible, at a settled cadence. After each result the model of what is attractive, blocked, or newly reachable updates — paced so interpretation stabilizes before the next step fires.

Tuning parameters

  • Branching policy — how aggressively each result reroutes the plan versus running a pre-set batch; adaptive branching learns faster but is harder to audit.
  • Step cadence — how long each result is allowed to settle before the next experiment; too fast and interpretation never stabilizes (thrash), too slow and momentum dies.
  • Memory horizon — how many past results inform the next choice; a longer horizon catches slow patterns but risks over-fitting to history.
  • Reversibility budget — how cheap each step is to undo; keeping steps disposable is what lets the sequence branch freely.

When it helps, and when it misleads

Its strength is accumulating a path through a space too large to map in advance — cheaply, reversibly, with each result buying the next question. It shines exactly where a single large experiment would be too blunt and pure trial-and-error too aimless.

Its failure mode is subtle: an undisciplined branching sequence with a rich trace is a garden of forking paths — with enough experiments and enough freedom to reinterpret after the fact, you will eventually "find" a result that is noise dressed as signal.[1] The classic misuse is HARKing across the sequence — deciding what you were really testing only after you see which branch looked good. The guarding discipline is to pre-commit each step's hypothesis and success bar before it runs, and to keep the trace honest so the branching stays auditable rather than retrofitted.

How it implements the components

Micro-Experiment Sequence fills the chaining-and-memory side of the archetype's machinery — the part that turns isolated probes into cumulative learning:

  • state_memory_trace — its core: the running record of results that gives the sequence memory across steps.
  • changed_observation_state — each experiment's result is a first-class observation that the next step consumes.
  • affordance_update — after each result, the model of what is now possible, attractive, or blocked updates.
  • loop_cadence — the deliberate rhythm at which experiments run and interpretations settle between them.

A Micro-Experiment Sequence chains and remembers; it does not implement epistemic_action_probe, safety_and_contamination_boundary, or perceptual_constraint_map — designing and bounding each single probe is its nearest twin, Active Probe Protocol, which this sequence consumes for every step.

Editorial Notes

Form Classification

Form family: Experiment, Test & Rehearsal

Rationale: Micro-Experiment Sequence operates as a bounded trial, probe, simulation, or rehearsal that generates evidence from performance because it chains many small, cheap tests into a branching series, carrying each result forward so every next experiment is chosen from what the previous ones revealed.

Independent corroboration: The frozen evidence defines Micro-Experiment Sequence as 'Chains many small, cheap tests into a branching series, carrying each result forward so every next experiment is chosen from what the previous ones revealed', so its operative form is Experiment, Test & Rehearsal.

Review outcome: Independent reviewer agreement; high confidence.

Origin Attribution

Primary origin: Statistics & Experimental Design

Origin pattern: Cross-disciplinary synthesis

Present-day reach: Multi-domain

Rationale: Choosing each small experiment from prior results is sequential or adaptive experimental design. Lean innovation and management practice made cheap branching tests an operating routine, but the evidence-acquisition logic is statistical.

Related originating lineages:

Review resolution: NIST's sequential Bayesian experiment-design program explicitly selects new experiments using information from prior measurements. That is the mechanism's defining logic; innovation and management are retained for the low-cost learning cadence. The alternates are retained only as formative or independently established origins, not because the mechanism can be applied there. origin_mode=cross_disciplinary_synthesis states the provenance relationship; domain_reach=multi_domain separately records breadth because independent established uses occur in several fields. confidence=high reflects the strength and specificity of the evidence; encyclopedia_synthesis=false because the entry generalizes an established mechanism without inventing a new composite.

Review outcome: Researched adjudication after independent review; high confidence.

Sources consulted:

References

[1] Gelman, A., & Loken, E. "The Garden of Forking Paths: Why Multiple Comparisons Can Be a Problem, Even When There Is No ‘Fishing Expedition’ or ‘p-Hacking’ and the Research Hypothesis Was Posited Ahead of Time". Unpublished manuscript, Columbia University (2013). Shows how data-contingent analytical choices can make pure noise appear statistically significant. registry