Coldstart has not been released. This section describes the method and development build at the source revision named above; there is no release you can install today.
The CSV-export plan looked sound until the team inspected the data. The existing query returns rows shaped for the screen: labels are formatted, notes are shortened, and the customer label is not a stable ID. Following the plan unchanged would build the wrong export.
Coldstart treats a plan as the best current route, not a promise to ignore new evidence. It asks who owns the decision first, then whether the resulting work belongs in the current session.
The three pages
| # | Concept | What it answers |
|---|---|---|
| 6.1 | Divergence | Should this surprise pause for a person, change the current route, or wait in an owned queue? |
| 6.2 | The open-work queue | How does valid work survive without expanding this session? |
| 6.3 | Safe subtraction | When may residue from the rejected route be removed? |
The ordinary way to change course
Without a formal method, a small team can stop, discuss the surprise, update a checklist, and keep a separate list of follow-up work. That is often enough. The trouble starts when three different things all become “while we are here” work:
- a product choice that needs the person steering the project;
- a technical correction already inside the approved outcome; and
- useful work that belongs to a later data-migration session.
Mixing them can hide a decision or make one export session absorb several jobs.
One export, three dispositions
The row-shape discovery first raises a product question: should the export reproduce the screen, or use approved raw fields and stable IDs? That meaning belongs to a person, so the dependent work pauses at a checkpoint.
After the person chooses raw fields, changing the current read route is an adaptation. It stays inside the approved export and requires no second product decision.
Migrating historical customer IDs is real work, but it has separate data risk and checks. It enters the open-work queue, a durable list of unfinished actions with an owning location. The export session stays bounded.
| Disposition | What it means | What it does not mean |
|---|---|---|
| Checkpoint | A person must answer before dependent work continues | The whole project must stop |
| Adapt | The current route changes inside the approved outcome | Anything discovered is now in scope |
| Queue | Valid work gets an owner for later | It will complete automatically |
Changing the plan is not the last step
The revised plan should tell one current story. Version control already preserves the old wording, so abandoned alternatives do not need to remain beside the live route.
Code left by the rejected route is a separate question. The temporary adapter that turned display rows back into export values may be dead, but a changed plan does not prove that. References, entry points, dynamic use, and ownership still decide whether it can leave.
The method has a nearby limit: it helps once somebody notices the contradiction. It cannot guarantee that every hidden product choice, external consumer, or risky assumption will be recognized. And sorting a surprise into the right disposition is not the same as the project turning out better for it; what the three dispositions establish is where the work went, not whether it was worth doing.
Next
Start at the switch itself: divergence.