2. Channel quarantine concepts

Posttest plausibility review rarely produces a single verdict of "pass" or "fail." More often it surfaces a channel whose data looks suspicious but not conclusively wrong — a pressure sensor that drifts only at high temperature, a thermocouple whose readings diverge from its redundant partner during a specific test segment, or a strain gauge whose noise floor rises after a load step. Channel quarantine is the mechanism that lets you act on that suspicion without either ignoring it or prematurely condemning the channel. It is a controlled, reversible isolation of a suspect measurement channel from downstream consumption, applied while the evidence is still being assembled.

The concept borrows from safety and quality engineering: when a component's integrity is in question, you do not keep using it as if nothing happened, and you do not necessarily scrap it. You isolate it, tag it, and route around it until a disposition is reached. For test data, "routing around it" means downstream analysis, control loops, alarm logic, and reporting treat the channel as unavailable or flagged rather than as valid.

Isolation Principles

Quarantine is defined by four principles that distinguish it from simple flagging or deletion.

Non-destructive isolation. The raw samples are never modified, overwritten, or deleted. Quarantine is a state applied to the channel's usability, not to its content. This preserves the evidence needed for root-cause analysis and allows the quarantine to be lifted without data loss. In practice this means the acquisition record keeps every sample, and a separate quarantine register records the channel, the time window, the reason, and the disposition status.

Scope control. Quarantine applies to a defined scope: a channel, a time window within a test, or a specific test run. A pressure channel that misbehaves only during a thermal soak should not be quarantined for the entire campaign. Narrow scope keeps the cost of quarantine proportional to the evidence and avoids discarding good data from unaffected segments.

Downstream propagation. A quarantined channel must be visibly excluded wherever it would otherwise be consumed. If a derived quantity — say, a mass-flow calculation that divides by a quarantined density channel — depends on the quarantined input, the derived quantity inherits the quarantine. Propagation is the part most often neglected: a channel marked bad in the review tool but still feeding a report is effectively not quarantined at all.

Reversibility with audit trail. Every quarantine action and every release is logged with who, when, why, and on what evidence. Reversibility matters because quarantine is a hypothesis, not a conviction; the audit trail matters because a released channel must be defensible later.

Quarantine Triggers

Triggers fall into three families, and the review workflow should treat them differently because their evidentiary weight differs.

Automated plausibility triggers. These are rule-based conditions evaluated during or immediately after acquisition: range violations, rate-of-change limits, flatline detection, excessive noise, or disagreement beyond a tolerance between redundant channels. For example, a rule might quarantine a channel when its sample-to-sample deviation exceeds six times the rolling median absolute deviation for more than 50 consecutive samples. Automated triggers are fast and consistent but prone to false positives on legitimate transients, so they typically open a quarantine in a provisional state pending review.

Reviewer-initiated triggers. During posttest review, an analyst may quarantine a channel based on visual inspection, cross-channel correlation anomalies, or comparison against a reference run. These triggers carry more context but are less reproducible, which is why the reason code and evidence reference must be recorded explicitly.

Upstream or external triggers. A calibration certificate that has expired, a sensor swap mid-test, a known instrumentation fault reported by the test conductor, or a data-acquisition dropout all justify quarantine regardless of what the samples look like. These are the easiest to defend and should be applied promptly.

The table below summarizes how each trigger family should be handled in the workflow.

Integration with the Posttest Review Workflow

Quarantine is not a side activity; it is a state that the review workflow must carry end to end. The diagram below shows how a channel moves between states as review evidence accumulates.

The workflow integration has three practical consequences. First, quarantine must be applied before derived quantities are computed, not after. If a mass-flow calculation runs on a quarantined density channel and the result is then flagged, you have already contaminated the derived record. The correct order is: evaluate triggers, set quarantine state, then compute derived quantities with quarantined inputs excluded or marked.

Second, the review report must distinguish between "channel passed review" and "channel was quarantined and later released." These are different evidentiary claims. A released channel has a documented reason for its earlier suspicion; a never-quarantined channel does not. Collapsing them loses information that a later auditor may need.

Third, quarantine state must be visible in the review dashboard alongside the plausibility findings that motivated it. A reviewer looking at a flagged channel should see, in one place, the trigger, the quarantine state, the scope, and the pending disposition. This prevents the common failure mode where a channel is quarantined in one tool and silently consumed in another.

A Worked Example

Consider a 40-minute thermal-vacuum test with a redundant pair of pressure transducers, PT-101 and PT-102, sampled at 10 Hz. During review, the analyst notices that PT-102 diverges from PT-101 by more than 0.8% of reading between t = 18:00 and t = 26:00, while agreeing within 0.1% elsewhere. The divergence coincides with a chamber temperature ramp.

The analyst applies a reviewer-initiated quarantine to PT-102 for the window 18:00–26:00, with reason code REDUNDANCY_DIVERGENCE and a reference to the comparison plot. Because the test's leak-rate calculation uses PT-102 as its primary input during that window, the derived leak-rate channel inherits the quarantine for the same window. The analyst then checks the PT-102 raw signal for a step or dropout; finding none, and noting that the divergence tracks temperature, the analyst flags a probable thermal sensitivity issue and routes the channel to Dispositioned with a recommendation to recalibrate PT-102 over temperature before the next campaign. PT-101 remains valid throughout, so the test's primary conclusion is unaffected, but the report explicitly notes that leak-rate results for 18:00–26:00 rest on a single transducer.

This example illustrates the full loop: a trigger, a scoped quarantine, downstream propagation, an evidence-based disposition, and a documented residual limitation. None of these steps required deleting data or halting the review, which is precisely the value of quarantine as a workflow primitive.

In summary, channel quarantine is a scoped, non-destructive, reversible isolation state with mandatory downstream propagation and an audit trail. Its triggers come from automated rules, reviewer judgment, and upstream events, each with different evidentiary weight. Integrated correctly into the posttest review workflow, it lets analysts act decisively on suspicion while preserving both the data and the reasoning needed to resolve it.