Signal Evidence Risks

Signal defects are amplified downstream. Incorrect metadata can alter search space, a one-chip error can corrupt every replica, and a small phase drift can surface only after long execution. The risk register therefore focuses on claims that can look plausible in a short or receiver-level test while the reusable signal contract is wrong.

How Weak Evidence Propagates

flowchart LR
    fact["catalog, code, sample,<br/>or DSP defect"]
    helper["reusable signal helper"]
    receiver["receiver behavior"]
    navigation["measurement and<br/>navigation outcome"]
    artifact["reported evidence"]

    fact --> helper --> receiver --> navigation --> artifact

A downstream failure can reveal the defect, but the correction still needs signal-layer proof at the first wrong boundary.

Active Risk Register

Risk	Warning sign	Required evidence	Claim limit
Catalog identity drift	Carrier, code rate, component role, secondary code, or wavelength changes without a named authority.	Registry consistency, exact metadata assertions, and the relevant specification or independent reference.	Correct metadata does not prove acquisition or tracking.
Reference circularity	Expected chips or periods are produced by the implementation being tested.	Checked-in or independently generated reference data with documented constellation, component, chip order, and provenance.	Self-consistency is not signal truth.
Short-window confidence	One period or frame passes while chunked or long-duration execution is untested.	Period-boundary, chunk-equivalence, and long-duration phase evidence.	A correct initial frame does not prove continuity.
Numeric tolerance masks discrete error	A broad epsilon permits the wrong chip, component, format, sign, or ordering.	Exact assertions for discrete meaning; derived tolerances only for continuous quantities.	Numerical proximity cannot repair identity.
Spectrum assumption drift	A plotted or computed shape omits modulation, sampling, normalization, or front-end bandwidth.	Formula or reference comparison with every physical assumption named.	A spectral helper does not prove receiver sensitivity.
Raw-IQ reinterpretation	Sample rate, intermediate frequency, byte order, signedness, or quantization is inferred from runtime configuration.	Metadata round trip plus boundary and conversion cases for each supported container.	Signal owns sample meaning, not dataset discovery or frame scheduling.
Validation becomes policy	Compatibility issues are promoted into lock, navigation, or operator verdicts.	Structured issue-kind properties and a clear consumer decision boundary.	Alignment evidence is not a receiver failure by itself.
Public surface outruns proof	A helper or table is exported because one caller needs convenient access.	Shared semantic argument, public guardrail, and domain-specific proof.	Reachability does not establish stable meaning.
Receiver evidence substitutes for signal proof	A receiver test passes, but no reference, property, or continuity test exercises the changed helper.	Focused signal proof followed by the first affected receiver boundary when needed.	Receiver success can corroborate but not define signal truth.

The test evidence guide describes the active test families. The known limitations keeps receiver, navigation, storage, and command claims outside this package.

Decide Whether the Claim Is Supported

flowchart TD
    claim["proposed signal claim"]
    authority{"named specification,<br/>formula, or reference?"}
    property{"exact or numeric property<br/>matches the contract?"}
    duration{"required boundary and<br/>duration exercised?"}
    consumer{"public or downstream<br/>meaning changed?"}
    supported["bounded signal claim<br/>is reviewable"]
    narrow["add evidence or<br/>narrow the claim"]

    claim --> authority
    authority -- no --> narrow
    authority -- yes --> property
    property -- no --> narrow
    property -- yes --> duration
    duration -- no --> narrow
    duration -- yes --> consumer
    consumer -- yes --> supported
    consumer -- no --> supported

For a public or downstream change, “supported” also requires compatibility review. The diagram does not imply that consumer evidence may be skipped; it shows that consumer evidence comes after the signal property is established.

Block the Change When

• the authority for a changed signal fact is unnamed
• expected data is regenerated solely because the implementation changed
• a one-frame test is used to claim chunk-stable or long-duration behavior
• a tolerance hides changed identity, sequence, format, sign, or ordering
• spectrum evidence omits its physical assumptions
• raw capture facts are guessed from receiver settings
• signal validation starts deciding receiver or navigation quality
• public exposure is justified only by one caller’s import convenience

Record Residual Risk

When complete evidence is unavailable, state the unsupported constellation, component, sample format, duration, rate, bandwidth, or consumer. Do not use “other signals should behave similarly” as proof. A family-specific reference does not automatically cover another modulation or secondary-code structure.

Signal quality review is complete when the physical authority, exact or toleranced property, duration, chunk behavior, public impact, downstream boundary, and remaining coverage gap are all explicit.