Maintainer Tooling Vocabulary

Precise language matters because developer tooling can validate the shape of a decision without approving the decision itself. Use these terms in command help, diagnostics, reviews, and automation so a passing maintenance check is not mistaken for security, policy, or performance acceptance.

Decision and Authority

flowchart LR
    authority["review authority"]
    record["governed record"]
    validator["typed validation"]
    derivation["derived arguments<br/>or comparison"]
    evidence["maintenance evidence"]
    decision["maintainer decision"]

    authority --> record --> validator
    validator --> derivation --> evidence --> decision
    validator --> evidence

Term	Meaning	Does not mean
review authority	person or owning process accountable for accepting the underlying risk, exception, lane assignment, or baseline	the validation command
governed input	reviewed repository data whose content changes a maintenance decision	any configuration file read by a script
governance record	one structured exception, deviation, roster entry, or baseline row inside a governed input	an informal comment or copied command argument
validation	checking implemented shape, required fields, expiry, identity, and relationship rules	approving the underlying exception or scientific claim
derivation	deterministic output computed from a governed source, such as audit ignore arguments	a second source of policy truth
maintenance decision	the action a maintainer or automation takes after considering validation and evidence	command success by itself

Security allowlist validation establishes that advisory records are attributable and time-bounded. It does not establish that accepting the advisory risk is correct. Deviation validation establishes that local exceptions remain linked to review; it does not alter shared standards.

Commands and Consumers

Term	Meaning
maintainer command	private binary subcommand implementing one repository-maintenance contract
command contract	governed inputs, parsing rules, effects, outputs, diagnostics, and exit behavior of one command
automation consumer	Make or CI logic that invokes a command and relies on its output or status
workspace root	explicit or current directory from which repository-relative inputs and outputs are resolved
package guardrail	integration proof that the developer-tooling package satisfies repository shape policy
repairable diagnostic	failure output that identifies the invalid record and rule closely enough for a maintainer to correct it

The public bijux gnss command tree is an operator interface, not a maintainer command. Product workflows, reports, and scientific validation remain outside this vocabulary even when Make invokes both binaries.

Test-Lane Language

flowchart LR
    tests["test inventory"]
    roster["curated slow roster"]
    generator["lane-expression<br/>generator"]
    slow["slow lane"]
    fast["fast lane<br/>excluding slow"]
    guard["suite-selection<br/>integration proof"]

    tests --> roster --> generator
    generator --> slow
    generator --> fast
    roster --> guard
    generator --> guard

Term	Meaning	Important distinction
slow-test roster	sorted, unique reviewed names assigned to governed slow selection	not a list generated by measured duration
slow-name namespace	tests selected through the maintained slow__ naming convention	every named test is also recorded in the roster for review
lane expression	generated nextest filter selecting slow tests or excluding them from fast execution	not the test result itself
suite-selection proof	integration test that roster entries resolve and feed coherent fast and slow expressions	does not measure duration or prove science
scientific test owner	product package responsible for the scenario and assertions	developer tooling owns only lane coherence

Call a test “slow-rostered” only when it is present in the governed roster. Call it “slow” based on observed execution only when the measurement context is also stated.

Benchmark Evidence

Term	Meaning	Review caution
product benchmark	benchmark implementation owned by receiver or navigation	developer tooling does not define measured semantics
curated benchmark set	package and benchmark names invoked by comparison tooling	absence from the set means no measurement
raw benchmark log	captured standard output from the current benchmark invocation	may include environment-sensitive measurements
current snapshot	normalized benchmark-name and duration rows parsed from the current run	not a maintained acceptance reference
maintained baseline	reviewed comparison snapshot intended to represent an accepted reference context	meaningful only with compatible environment and provenance
threshold ratio	configured multiplier used to classify slower current measurements	not a scientific accuracy budget
regression finding	matched current value exceeds its maintained baseline value by the threshold	requires investigation; environment may contribute
strict mode	command mode that fails when regression findings exist	cannot enforce comparison when no baseline exists

Do not say “benchmarks passed” when the command merely ran. State whether a baseline existed, which names matched, which threshold applied, and whether strict mode converted findings into failure.

Output and Evidence

Use “maintenance evidence” for output supporting repository review. Qualify it by lifecycle:

• generated evidence belongs to the current execution and may be disposable
• normalized current evidence is structured for comparison but is not automatically accepted
• maintained evidence is intentionally reviewed and versioned
• standard output may be a machine contract, a human report, or both; identify which before changing it

Avoid using “artifact” alone in this package. Product crates use artifact for versioned GNSS runtime records, while developer tooling emits benchmark and governance evidence. Name the concrete output instead: benchmark log, current snapshot, baseline, derived arguments, or diagnostic.

Failure Language

• invalid means a governed input violates an implemented rule
• missing means a required input cannot be found at the resolved workspace root
• expired means a date-shaped governance record is earlier than the current day used by the command
• unsupported means the workflow does not recognize the requested meaning
• execution failure means a file, date, parser, subprocess, or write operation failed before a valid decision could be produced
• regression finding means comparison completed and exceeded a threshold; it is not synonymous with command failure outside strict mode

Use the command inventory, governance inputs, output contract, and benchmark contract for the concrete surfaces behind these terms.

Language is accurate when a reader can distinguish authority from validation, policy from derivation, product evidence from maintenance evidence, lane selection from test results, and regression findings from command failure.