V1 record — describes retired software; superseded for V2 by srd.md, constitution 2.0.0 and ADR-0027.

ADR-0005: Sound speed is derived at the point of use, not stored

Status: Accepted Date: 26 August 2026 Requirements: SRD FR-02, FR-16, FR-24, FR-37; SRD §2.2 Raised by: feature 007's specification, where the SRD does not assign the derivation

Context

FR-24 has the monitor compute the residual between measured and forecast sound speed, and the SRD is emphatic that it is sound speed and not temperature. FR-02 has the environment generator produce temperature, salinity and pressure, "from which sound speed is derived". SRD §2.2 lists sound speed computation among the genuinely bespoke logic inside the boundary.

What the SRD does not say is where the derivation happens. Feature 007's specification surfaced the gap: if the simulated sensors publish sound speed as a fourth datastream and the ingest client stores it, then the monitor reads it; if they publish only the three measured quantities, the monitor derives it. Two specifications were about to make opposite assumptions.

Decision

Simulated sensors publish temperature, salinity and pressure only, as three SensorThings Datastreams. Sound speed is never published, never stored, and is not a datastream.

Sound speed is derived at the point of use, by a single shared implementation in libs/harness_core, called by the monitor (FR-24) and by telemetry (FR-37).

Consequences

• There is one implementation of the sound speed equation in drogna, and it is inside the boundary where SRD §2.2 puts it. Both consumers get the same numbers by construction rather than by agreement.
• The observation store holds only measured quantities. A derived value stored beside its inputs is a second source of truth that can disagree with them after a change to the equation, and there would be no way to tell which was right.
• The equation is versioned with the code, not with the data. Recomputing a residual after changing the equation gives a different answer for the same stored observations, which is correct: the residual is a statement about the current model of sound speed, not a historical fact.
• The environment generator derives sound speed too, for the field it writes, and must call the same implementation. A generator with its own copy of the equation would make the recovery error of AT-03 partly an artefact of the disagreement.
• Sensors carry no bespoke logic, which is what a simulated sensor ought to look like.
• The chosen equation and its validity range are a documented modelling decision, and the harness's numerics are deliberately fake (SRD §1.1) — the point is the seam, not the oceanography. Whichever formulation is used, it is stated in docs/algorithms/, not left implicit in a function body.