Gain-Change Review Log¶
Audit artifact — instantiates Adaptive Gain Retuning
A dated, append-only record of every gain change — what moved, on what evidence, the effect expected, the bounds it stayed within, who approved it, and what actually happened.
A gain that retunes itself leaves no natural paper trail; yesterday's sensitivity is simply overwritten by today's. The Gain-Change Review Log refuses that amnesia. For every change to the gain — automatic or manual — it records what moved, the evidence that triggered it, the effect expected, the floor/ceiling it stayed within, the human who signed off if one was required, and, after the fact, what the change actually did. Its distinguishing feature is that it is written around the decision, not around the signal: where a dashboard shows the live state and a test probes behaviour, the log preserves the reasoning and accountability of each retuning so it can be reviewed, reversed, or answered for later. It is the mechanism that makes an adaptive gain governable rather than merely automatic.
Example¶
A wearable cardiac monitor detects arrhythmias, and the sensitivity of that detector is periodically retuned — tightened when a patient's baseline is noisy with motion artifact, loosened when too many benign ectopic beats are being flagged. Each retune is a small clinical-risk decision: loosen too far and a real run of atrial fibrillation is missed. The Gain-Change Review Log captures each one as a line: 2026-03-11, detection gain lowered one step, trigger = 40+ artifact alerts/day the prior week, expected effect ≈60% fewer nuisance alerts, bounds = above the validated floor for AF sensitivity, reviewer = on-call cardiologist, follow-up = re-check AF recall at 30 days. Ninety days later, when a clinician asks why a patient's early AF wasn't flagged, the log is what turns a mystery into a traceable chain: it shows exactly when gain was lowered, on what evidence, who approved it, whether the 30-day recall check was ever done — and whether the floor was, in hindsight, set too low.
How it works¶
- Record at the moment of change, not after the outcome. Each entry is written when the gain moves, capturing evidence and expectation before the result is known — which is precisely what makes it trustworthy later.
- Capture the decision, not just the value. Old→new gain, trigger evidence, expected effect, the bounds it respected, and the reviewer — enough to reconstruct why, not only what.
- Close the loop with a follow-up field. Each entry carries a post-change check that gets filled in later, so a change that quietly made things worse can be found and undone.
- Keep it append-only. Entries are added, never edited, so the record can't be retrofitted to match the outcome.
Tuning parameters¶
- Granularity — log every micro-adjustment or only material changes. Finer capture is fully auditable but noisy; coarser is readable but can miss a slow drift assembled from many small steps.
- Review threshold — which changes need a human sign-off before they take effect versus after. Tighten it for high-stakes gains; loosen it where the adaptive loop is trusted.
- Follow-up cadence — how long after a change its effect is re-checked and recorded. Short cadence catches harm fast; long cadence captures durable effects.
- Immutability — append-only versus editable. Strict immutability maximizes trust and forensic value at the cost of convenience.
- Retention window — how far back the trail is kept, traded against storage and the investigation horizon you need to support.
When it helps, and when it misleads¶
Its strength is accountability and reversibility: when a gain change turns out to have caused a miss or a flood, the log is what lets a team find the specific change, see the evidence it rested on, and undo it — and its contemporaneous entries resist the natural urge to remember the decision as more reasonable than it was. That guard against hindsight bias is much of the point: a record written before the outcome is known cannot be quietly rewritten to fit it.[n1]
Its failure modes are the ones every log has. A log that is written but never read is theatre; a log back-filled after an incident is worse than none, because it lends false authority to a reconstructed story. And because it captures what is easy to record (the numeric change, the trigger) it can miss what mattered (the tacit judgment, the pressure to cut alerts). The discipline that keeps it honest is append-only capture at the moment of change, a real follow-up field that someone actually fills in, and periodic reading of the log as a set — not merely writing to it.
How it implements the components¶
gain_change_audit_trail— it is this component: the durable, append-only record of every retuning and its rationale.human_review_gate— it operationalizes the sign-off: the log is where a required reviewer's approval (or its absence) is captured and can later be checked.
It does not display the live gain state or occupancy — that's the Saturation Occupancy Dashboard; it neither sets the bounds a change must respect (the Gain Floor/Ceiling Rule) nor produces the changes it records (the Automatic Gain Control Loop and Contextual Gain-Scheduling Table).
Related¶
- Instantiates: Adaptive Gain Retuning — provides the accountability layer over every retuning the other mechanisms perform.
- Consumes: Saturation Occupancy Dashboard — its before/after occupancy readings supply the "actual effect" each entry records.
- Sibling mechanisms: Gain Floor/Ceiling Rule · Automatic Gain Control Loop · Contextual Gain-Scheduling Table · Saturation Occupancy Dashboard · Hysteretic Gain-Update Filter
Editorial Notes¶
Form Classification¶
Form family: Record, Log & Register
Rationale: Gain-Change Review Log operates as a durable record, ledger, register, or trace whose value depends on preserving actual state or history because it a dated, append-only record of every gain change — what moved, on what evidence, the effect expected, the bounds it stayed within, who approved it, and what actually happened.
Independent corroboration: The frozen evidence defines Gain-Change Review Log as 'A dated, append-only record of every gain change — what moved, on what evidence, the effect expected, the bounds it stayed within, who approved it, and what actually happened', so its operative form is Record, Log & Register.
Nearest alternative: Assessment, Review & Assurance — The append-only log preserves contemporaneous gain decisions, evidence, approvals, and later effects; audits consume that record afterward.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Engineering & Design
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: Append-only decision records and audit trails supply the accountability form around parameter changes.
Related originating lineages:
- Accounting & Auditing — Append-only authorization and contemporaneous audit trails materially shape the record's governance.
- Systems Thinking & Cybernetics — Adaptive-control practice supplies the gain changes being governed.
Review resolution: NASA configuration-management guidance requires engineering changes to be proposed, justified, evaluated, approved, implemented, verified, and retained in change-history records. That establishes engineering_design as the primary lineage of this parameter-change log. Accounting and audit practice materially shapes its append-only accountability form, while systems/cybernetics supplies the gain concept. The exact gain-specific record is an encyclopedia synthesis.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
- https://www.nasa.gov/reference/6-5-configuration-management/ — NASA configuration change management process.
- https://nodis3.gsfc.nasa.gov/displayCA.cfm?Internal_ID=N_PR_7123_001B_&page_name=AppendixC — NASA requirements for recording, approving, tracking, and auditing changes.
Notes¶
The log records a change; it does not stop one. It can host a review gate's sign-off, but it enforces nothing on its own — if no mechanism actually blocks an un-reviewed change from taking effect, the log will faithfully document that governance was skipped rather than prevent the skip. Pair it with an enforcing control where a change genuinely must not go out unreviewed.
[n1] Hindsight bias — the tendency, once an outcome is known, to see it as having been predictable and to misremember one's earlier reasoning as closer to it. A record written at the time of the decision is the standard guard, which is why entries here are dated and append-only. ↩