A two-hour spike killed a two-week epic

@cia/react was on the roadmap for about two weeks: take the React block out of each recipe, generate a component per recipe, publish a package. It had stories, acceptance criteria, an effort estimate, the lot.

The epic carried a fail-fast clause, so before building it I built the generator badly and pointed it at three recipes. That took an afternoon.

Two of the three generated components crashed on first render. breadcrumb expects a crumbs array the generated wrapper never passes. toast threw a ReferenceError for a symbol the template did not define. The third worked, which is the result that would have been most dangerous if I had only tried one.

The more interesting number was the copy ratio. Of the 103 lines in the biggest output, 88 came verbatim from the recipe and 15 were synthesised — and those 15 were byte-identical across all three components apart from the name. It was a fixed wrapper plus a DO NOT EDIT header. That is not code generation, it is cat with extra steps, and it is not worth a published package and the support surface that comes with one.

So the epic is dead. The spike survives at spike/ as the evidence, and the epic file keeps its stories verbatim as the plan that was not executed.

The thing I want to remember is that the estimate was never the problem. Two weeks was probably about right for building it. The question the estimate could not answer was whether the output would be worth having, and only running it could answer that.