Receiver Evidence

Receiver quality is not one end-to-end success condition. It is a chain of claims about search, lock, continuity, measurements, handoffs, and emitted evidence. Judge each claim at the boundary where it is created, then add only the downstream proof needed to show that its meaning survives.

Match the Claim to the Proof

Claim	Primary evidence	What that evidence does not prove
An acquisition is correct	hypothesis, code phase, Doppler, ambiguity, uncertainty, and truth comparison	sustained tracking lock
A channel tracks correctly	epoch errors, state transitions, continuity, fades, and reacquisition	observation or position accuracy
An observation is usable	measurement errors, quality decisions, residuals, covariance, and exclusions	estimator correctness
A receiver run is reproducible	controlled inputs, ordered outputs, exact discrete state, and bounded numeric equality	physical realism of the input
A navigation handoff is sound	typed observations, feature behavior, validation report, and refusal evidence	ownership or correctness of navigation science
A run artifact is trustworthy	field semantics, units, thresholds, observed values, and serialization checks	repository persistence or history behavior
A change is faster	comparable benchmark or profile evidence	unchanged receiver correctness

The test strategy maps these claims to unit, stage, integration, truth-table, golden, and feature-gated test families. The invariant catalog describes properties that must hold across individual scenarios.

Build Confidence by Boundary

flowchart LR
    input["declared samples,<br/>config, and conditions"]
    acquisition["acquisition<br/>evidence"]
    tracking["tracking state<br/>and continuity"]
    observations["measurement epochs<br/>and quality"]
    handoff["optional navigation<br/>handoff"]
    artifacts["typed receiver<br/>artifacts"]

    input --> acquisition --> tracking --> observations --> handoff
    acquisition --> artifacts
    tracking --> artifacts
    observations --> artifacts
    handoff --> artifacts

A later green result cannot erase an earlier unexplained failure. For example, a position inside a broad budget does not justify an acquisition ranking change, and a stable tracking epoch does not justify silently dropping an observation. Use change validation to choose the first boundary that must be checked and the next boundary that could expose a regression.

Separate Repeatability from Truth

Deterministic receiver execution makes failures reproducible. It does not show that a synthetic propagation model, noise profile, oscillator drift, or reference value represents the intended physical case.

flowchart TD
    claim["receiver claim"]
    repeatable{"same controlled input<br/>gives stable evidence?"}
    reference{"independent truth,<br/>invariant, or reference?"}
    budget{"named tolerance and<br/>coverage satisfied?"}
    supported["bounded claim supported"]
    reject["claim remains unsupported"]

    claim --> repeatable
    repeatable -- no --> reject
    repeatable -- yes --> reference
    reference -- no --> reject
    reference -- yes --> budget
    budget -- no --> reject
    budget -- yes --> supported

Read determinism and purity for exact and toleranced replay contracts. Read validation budgets before changing a threshold; loosening a budget because a test failed is not a scientific explanation.

Bound the Current Evidence

The suite contains broad synthetic, truth-table, golden, and integration coverage, but its conclusions remain scenario-bound. It does not establish equal coverage for every constellation, band, interference environment, oscillator, antenna, hardware front end, or field condition. Feature-gated navigation and RTK evidence also applies only when those paths are built and exercised.

The known limitations states the practical bounds. Use the risk register when a change increases coupling, widens a public contract, or depends on evidence the repository does not yet have. Use definition of done and the review checklist to close the review without converting missing proof into an implicit claim.

Stop on These Signals

• an expected output was regenerated from the implementation it is meant to judge
• a broad integration pass is offered instead of evidence for the changed stage
• a numeric tolerance hides changed ordering, state, identity, or refusal
• performance results omit workload, build, environment, or uncertainty
• navigation output is used to claim receiver correctness without inspecting the receiver handoff
• synthetic success is described as universal field performance
• an artifact records a pass without the applied threshold and observed value

Quality review is complete when the claim, owning boundary, independent evidence, tolerance, coverage, and remaining limitation can all be named without relying on the test name alone.