Skip to main content
When you analyze a ticket, Waterline runs a pipeline that combines semantic search across your codebase with LLM-powered scoring — then produces a deterministic result from consistent thresholds.

​
The pipeline

1

Fetch the ticket

Waterline retrieves the ticket summary, description, and acceptance criteria from Jira, GitHub Issues, or Asana.
2

Search the codebase

The ticket is used to search your indexed codebase semantically. Waterline finds the functions and classes most likely to implement the ticket’s requirements.
3

Score relevance

Each candidate symbol is scored for how directly it addresses the ticket. Noise is filtered out before the next step.
4

Extract and map criteria

Waterline extracts the discrete acceptance criteria from the ticket description, then maps each piece of code evidence to the criteria it supports.
5

Aggregate the score

Each criterion is classified as SATISFIED, PARTIAL, or UNSATISFIED based on the strength of evidence. The overall score is computed from consistent, fixed thresholds — not a variable LLM output.

​
Results are cached

Waterline caches analysis results. When a cached result is available, it returns in under a second. After each push, Waterline pre-computes code-to-ticket alignments so the next analysis is faster — but a full re-analysis only runs when you request it.

​
What you get back

Each analysis returns:
  • Overall progress % — the headline score
  • Per-criterion breakdown — SATISFIED / PARTIAL / UNSATISFIED for each requirement, with a confidence level
  • Evidence list — the specific functions and classes that support each criterion
  • Uncertainty level — LOW, MEDIUM, or HIGH, indicating how confident Waterline is in the overall result

​
Result example