Training · Cross-cuttingMoveprovenverified

Experimental Exploration with Checkpoints

also known as agile spike, time-boxed spike, spike solution, time-boxed exploration, learning spike

Tags: spiketime-boxexplorationagileXPuncertainty-reduction

When a problem is too uncertain to plan directly, the learner runs a time-boxed experiment — a spike — aimed at a specific question rather than a deliverable. A fixed time limit and a clearly stated question keep the exploration productive without allowing it to become unbounded research. Checkpoints at the midpoint and end of the box force a decision: is the question answered, should the box be extended, or should the approach change? The output is knowledge, not code or a product — and that knowledge is sufficient to unlock the next planning decision.

How the learner advances

flowchart TD A[Specific answerable question + hard time box agreed] --> B[Run exploration fastest reliable method] B --> C[Midpoint checkpoint will question be answered?] C -- Yes --> D[Continue exploration] C -- No: too broad --> E[Narrow question or extend box deliberately] E --> B D --> F[Time box ends] F --> G[Write spike report question / approach / finding / decision enabled] G --> H[More capable learner + decision unblocked by evidence]

Intent. Resolve a specific uncertainty through a strictly time-boxed exploration so that the next planning or learning decision can be made on evidence rather than assumption.

When to apply. Apply when a decision or next step is blocked by a specific unknown — a technical question, a feasibility concern, a conceptual gap — that can be investigated in a bounded time without full commitment to an implementation. Do not apply to vague uncertainty; the exploration requires a concrete, answerable question. Do not extend the box without a conscious checkpoint decision.

Threshold — earns the next step. The learner can identify the specific decision that the spike unblocked and explain, with reference to the spike report, what evidence changed the situation from uncertain to decided.

Masterpiece — the artifact that proves it. A spike report that states a precise question, a bounded investigation, a clear finding, and the decision it enabled — demonstrable by showing that the next planning step was made using the spike's evidence rather than assumption.

Example scenario

A machine learning team is deciding whether to switch their inference pipeline from batch processing to real-time streaming. The architecture change is large; committing to it without evidence would be expensive if the latency targets cannot be met. The team proposes a spike: the question is 'Can our current model serve predictions at under 200ms p99 latency with 500 concurrent users on our candidate streaming framework?' The time box is two days. One engineer runs the investigation: deploys a minimal version of the model behind the streaming framework, generates load at 500 concurrent users, and measures latency. At the midpoint checkpoint the question is on track. At the end of the box she writes the spike report: p99 latency was 147ms under load, within budget, but memory overhead was 40% higher than batch. Decision enabled: proceed with streaming architecture, but include a memory monitoring alert in the implementation plan. The report is shared; the team makes the architecture decision in the next planning session based on evidence. No code from the spike is retained — it was exploratory scaffolding. The masterpiece is the spike report: a two-page document that replaced weeks of assumption-based planning.

Facets

Container — async
Mode — experimentalexploratorytime-boxed
Reach — individual
Persona — learnerpractitionerautonomous-agent
Craft (AI Fluency) — discernmentdiligence
Learner — humanautonomous-agent
Trainer — humanautonomous-agent

Inputs

Specific answerable question — A precise question whose answer will unblock a decision — not 'how does this technology work?' but 'can this library handle 10,000 concurrent connections within our latency budget?'
Fixed time box — A hard upper bound on exploration time, agreed before the spike begins — typically one to three days. The box is the guardrail that prevents learning from becoming indefinite research.

Outputs

More capable learner or team — A learner or team that has resolved a concrete uncertainty and can now make the next decision based on evidence rather than assumption.
Spike report — A short written record — the masterpiece — stating the question explored, the approach taken, what was found, and the decision it now enables. The spike report is what the learning produced; any code or prototype is disposable unless explicitly retained.

Steps (4)

State the question and the time box
Before beginning, write the specific question the spike must answer and the hard end time of the box. Share both with anyone affected by the next decision. A spike without a written question tends to drift into general research.
Run the exploration
Investigate the question using the fastest reliable method — prototype, literature scan, expert interview, experiment, or calculation. Explore only what serves the question; resist the pull of interesting adjacent territory.
Midpoint checkpoint
At the midpoint of the time box, assess: is the question going to be answered by the end? If yes, continue. If no, decide: narrow the question further, extend the box with a specific reason, or stop and report what was learned so far. The checkpoint is mandatory — an unchecked spike will always expand.
Write the spike report
At the end of the box, write the spike report: question asked, approach taken, what was found, and what decision is now enabled. If the question was not fully answered, say so and state what further investigation would be needed. The report is the output — not the prototype.

Principles

The output is knowledge, not a deliverable — any artefact produced during the spike is a means of learning, not the goal.
The time box is not a suggestion — it is the mechanism that distinguishes a spike from open-ended research; a soft box is no box.
A concrete question before entry, a written report at exit — these are the two non-negotiable boundaries of the move.

Known uses (4)

Known failure modes (2)

[open-ended-research-masquerading-as-spike]
Anti-pattern: starting an exploration with a vague question and no hard time box, calling it a spike. Without a concrete question and a fixed end time, the activity is general research — useful, but not the move. Learning is diffuse and no decision is unblocked.
[spike-artefact-mistaken-for-deliverable]
Anti-pattern: retaining and building on the prototype, script, or code produced during the spike as if it were production-quality. Spike artefacts are disposable learning tools; treating them as deliverables introduces technical debt built on exploratory, unreviewed work.

Related trainings (4)

Sources (3)

Provenance

Ecosystem: software engineering, professional development, autonomous-agent training
Added to catalog: 2026-05-27
Last updated: 2026-05-27
Verification status: verified

How the learner advances

Steps (4)

State the question and the time box

Run the exploration

Midpoint checkpoint

Write the spike report