A Lightweight Code Version of the Context Engine

At some point, this stops being a concept.

If you’re using this system consistently, you’ll start to notice something:

you’re already doing the same operations over and over.

• summarising inputs
• updating context
• cleaning contradictions
• running tasks with structured state

This part is about making that explicit.

Not with a full app.

Just enough code logic to show:

this can be systemised without becoming a project.

---

The two core operations

Everything you’ve done in this series reduces to two actions:

1. Update context
2. Run a task with context

That’s it.

So the simplest possible system is just two functions.

---

1) Updating context (the real engine)

This is your rolling context logic in code form.

function updateContext(oldContext, newData) {
  const summary = summarise(newData)

  const merged = mergeAndClean(oldContext, summary)

  return merged
}

That’s the entire idea.

But what matters is what those inner steps actually do.

---

summarise(newData)

This is not “shorten text.”

It is:

• extract goals
• extract constraints
• extract decisions
• extract open questions
• ignore everything else

---

mergeAndClean(oldContext, summary)

This is where most people fail.

This function must:

• overwrite outdated truths
• preserve still-valid constraints
• append new decisions (not duplicates)
• remove contradictions
• prune low-value noise

Conceptually:

function mergeAndClean(oldContext, summary) {
  return {
    goals: overwrite(oldContext.goals, summary.goals),
    constraints: overwrite(oldContext.constraints, summary.constraints),
    preferences: overwrite(oldContext.preferences, summary.preferences),

    decisions: appendUnique(oldContext.decisions, summary.decisions),

    openLoops: overwrite(oldContext.openLoops, summary.openLoops),

    // optional: prune anything stale or duplicated
  }
}