Principles
The principles
Every decision in rqm answers to these ten principles. When a new rule seems to need an exception to one of them, the rule is the thing that changes.
- Lifecycle is not verification.
statusis written by a person, and verification is computed. Nothing in a YAML file can assert that a requirement is verified. - Evidence is not a claim. A passing run produces a run record. Only an explicit
baseline updatechanges the baseline. - Exact or nothing. There is no fuzzy matching, no suffix heuristic, no reciprocal link that rqm adds for you and no partial credit for an ambiguous case. A test earns credit for a requirement only through the static mapping (annotations and selectors), joined to a report case by exact comparison against the closed set of declared test identities. The one extension is the prefix rule for parameterized tests, which looks only at declared ids. Report content can say which case ran and what happened to it, and never grants credit.
- Failures are input, not absence. A report that cannot be parsed, a baseline that cannot be read, a case that cannot be tied to a test and a suite that did not run are four different, named outcomes. None of them is silently “no problem found”.
- Hashing is versioned and tested against fixed digests. The algorithm has an id, and golden digests pin its output. When the canonical form changes, the id changes, and old records become stale instead of being reinterpreted.
- A cache is an optimisation, never an authority. rqm has no cache today. A cache added later is keyed by content hash only.
- Determinism is scoped, not claimed everywhere. The commands that only read (
list,show,check,coverage,trace,testsandaffected) give the same bytes for the same input. The commands that run tests and record evidence carry run ids, times and durations, and make no such claim. - Everything displayed is a function of something stored. No display rule exists that cannot be derived from the stored facts, and
verifiedhas exactly one rule that produces it. - Execution is argv, always. A suite is a program and a list of arguments, started without a shell.
- Nothing depends on how YAML is written. rqm writes YAML with a serializer and defines correctness as “the strict reader parses it back to the same model”, never as a particular layout of bytes.