13 checks matched their expected outcomes. Rejected inputs and interrupted transactions are intentional acceptance cases.
Open the interactive reference report / Machine-readable record
{
"tests": 10000,
"input_bytes": 1051391,
"ingest_seconds": 0.130844,
"analysis_seconds": 0.084038,
"database_bytes": 4542464,
"python": "3.12.14",
"sqlite": "3.53.1"
}Timings describe one local run on the available runtime. They are not measured production CI performance.
Final fixture classification matches an independent expected table
Five synthetic builds use known fixture statuses and explicit full/partial modes.
Expected: {"new failure": 2, "repeated failure": 1, "failure without baseline": 0, "retry pass": 1, "recovered": 1, "pass": 14, "skipped": 1, "not reported": 1}
Observed: {"new failure": 2, "repeated failure": 1, "failure without baseline": 0, "retry pass": 1, "recovered": 1, "pass": 14, "skipped": 1, "not reported": 1}Matched expected outcome
Earlier attempts remain inspectable after partial retry
Original artifacts are retained as immutable BLOBs; twelve ingested job attempts.
Expected: 12 Observed: 12
Matched expected outcome
Repeating all twelve imports adds no observations
Repeat exact basename/content/mode identities against the committed database.
Expected: [true, 102] Observed: [true, 102]
Matched expected outcome
A different artifact cannot overwrite an existing attempt
Attempt key is build/job/attempt; complete artifact fingerprint and mode must agree.
Expected: [true, true] Observed: [true, true]
Matched expected outcome
Malformed XML has no database side effects
Parse and validate every artifact before beginning the ingestion transaction.
Expected: [true, true] Observed: [true, true]
Matched expected outcome
An actual file above 10 MiB is rejected by the CLI
Bounded file read and combined-attempt size check, using the production limit.
Expected: [2, true] Observed: [2, true]
Matched expected outcome
Abrupt exit after a case insert leaves no partial build or artifact
A subprocess calls os._exit(86) inside the transaction, before COMMIT. SQLite recovers the journal on reopen; this is not a physical power-loss test.
Expected: [86, true] Observed: [86, true]
Matched expected outcome
The interrupted identity can be imported once after restart
Normal ingestion after recovery and read-only analysis.
Expected: [true, 6] Observed: [true, 6]
Matched expected outcome
Two concurrent importers commit one attempt and return one duplicate
Two real Python processes write the same database; BEGIN IMMEDIATE serializes comparison and insertion.
Expected: [1, 1, 1] Observed: [1, 1, 1]
Matched expected outcome
Every embedded XML download reproduces its committed SHA-256
Parse the exact generated HTML JSON payload, decode all artifact bytes and recompute digests.
Expected: true Observed: true
Matched expected outcome
Unchanged committed history produces identical HTML
No wall-clock timestamp, random IDs or network state enters report generation.
Expected: "f433e32970570adba9892b42daff89a84ae02446b976e039b5f6b0ff2be24ee7" Observed: "f433e32970570adba9892b42daff89a84ae02446b976e039b5f6b0ff2be24ee7"
Matched expected outcome
Check returns a failing job status for final failures
The CLI independently reports final failure/error count, without making XML ingestion itself a release gate.
Expected: [1, 3] Observed: [1, 3]
Matched expected outcome
Ten thousand synthetic results retain identities and ten failures
One generated XML file, actual local SQLite ingest and full analysis. Timing is an observation, not a service-level commitment.
Expected: [10000, 10] Observed: [10000, 10]
Matched expected outcome
Boundaries
Synthetic local fixtures, not production CI data.
No provider API connector, automatic branch/commit comparison, root-cause diagnosis, or release certification.
Abrupt process exit tests transaction recovery, not physical power-loss durability.
Per-attempt resource limits do not impose an unlimited-history database or browser capacity guarantee.