Contamination Monitoring Log¶
Monitoring log — instantiates Control-Condition Specification
A running record kept during execution that captures every instance of crossover, spillover, and drift so the tested contrast can be reported as what actually happened, not what was planned.
A control condition can be defined perfectly on paper and still erode in the field. Control participants borrow the treatment materials, treated and untreated people talk, a site quietly starts offering the intervention to everyone, or "usual care" drifts over the study's twelve months. Contamination Monitoring Log is the live ledger that catches this while the study is running: each time the wall between arms is breached, someone records who, where, when, and how much. It does not prevent contamination and it does not define the arms — it observes and dates the leaks, so that at analysis time the team knows how much of the intended contrast survived and can bound the conclusion to the comparison that was actually delivered.
Example¶
A district is running a two-year trial of a new middle-school math tutoring program across fourteen schools. Treatment schools get the program; control schools continue as usual. But teachers move between schools, attend the same professional-development days, and share resources in a shared drive. Contamination is not hypothetical.
The evaluation team stands up a Contamination Monitoring Log before the first tutoring session. Whenever a control-school teacher reports having downloaded the program's worksheets, a coach notes a treatment school lending materials, or a control student transfers into a treatment classroom, it goes in the log with a date, a source, and a coarse severity rating. Termly, the team tallies it. By month eighteen the log shows that three control schools have adopted noticeable pieces of the program after a viral staff-meeting demo. That entry does not get erased — it gets carried into the report. The published finding reads not "program versus no program" but "program versus a partially-contaminated business-as-usual, with three control sites showing documented spillover," and the effect estimate is flagged as likely conservative because the control absorbed some of the treatment.[n1]
How it works¶
The log is a structured, append-only instrument built around anticipated breach channels rather than open-ended note-taking:
- Pre-enumerate the channels. Before launch, list the realistic ways the arms can bleed together — crossover of units, sharing of materials, provider migration, shared training, control access to the treatment, drift in the comparator's content.
- Capture at the point of observation. Field staff, coaches, and site liaisons log each incident when they see it, with date, location, channel, and a rough magnitude, rather than reconstructing it later.
- Roll up on a cadence. Periodic tallies turn scattered incidents into a contamination profile per arm and per site, so the size and timing of leakage is visible during the study, not just after.
- Feed the interpretation. The final tally becomes an appendix to the results, converting "the contrast was clean" from an assumption into a documented, quantified claim with named exceptions.
Tuning parameters¶
- Channel granularity — how finely breach types are pre-coded. Finer coding captures subtle drift but burdens the people doing the logging.
- Reporting threshold — how small an incident is worth an entry. A low threshold catches slow drift; a high one avoids drowning in noise but can miss cumulative leakage.
- Severity scale — whether incidents are flagged binary or rated by dose. Dose ratings support a graded sensitivity analysis but demand judgment the field staff may not have.
- Audit intensity — how actively the team probes for unreported contamination versus relying on volunteered reports. More probing finds hidden leaks; it also strains site relationships.
When it helps, and when it misleads¶
Its strength is honesty under real-world conditions: it converts the fragile assumption that the two arms stayed separate into an evidence trail, and it lets a null or shrunken effect be correctly read as "the contrast decayed" rather than "the treatment failed." It is also what tells reviewers exactly how much to trust the headline comparison.
It misleads through under-capture. A log only records what people notice and are willing to report, and contamination is precisely the thing participants and sites have reason to hide, so a quiet log can create false confidence in a contrast that actually collapsed. It is also purely descriptive: recording a breach is not the same as fixing it, and a team can mistake diligent logging for control integrity. The guarding discipline is to pair the log with active auditing, to treat a suspiciously empty log as a red flag rather than a clean bill, and to pre-commit to how logged contamination will adjust the analysis so the record cannot be quietly ignored when the numbers are disappointing.
How it implements the components¶
control_condition_integrity_guardrail— it operationalizes the integrity guardrail as continuous surveillance, turning "protect the comparison" into a dated record of every observed breach.contrast_bounded_reporting_record— its running tally becomes the appendix that bounds the reported effect to the contrast actually delivered, naming where and how much the arms leaked.
It does not define what the control arm should contain (control_condition_definition) — that is Control Arm Protocol — and it does not check delivery against a specification at a point in time (treatment_control_contrast_map, comparator_equivalence_boundary), which is the job of its nearest twin, Control Condition Fidelity Checklist; the log records breaches as they occur over the whole study, the checklist verifies conformance at a moment.
Related¶
- Instantiates: Control-Condition Specification — it preserves the interpretability of the contrast during execution by documenting how much of it survived.
- Consumes: Control Arm Protocol supplies the specified arms whose separation the log then watches for breaches.
- Sibling mechanisms: Control Condition Fidelity Checklist · Control Arm Protocol · Attention Control Script · Placebo or Sham Procedure · Standard-Care Comparator Specification · Usual-Care Inventory Form · Waitlist Control Schedule · External Control Justification Memo
Editorial Notes¶
Form Classification¶
Form family: Record, Log & Register
Rationale: Field observers append each actual crossover, spillover, or drift incident with time, place, channel, and magnitude and preserve rollups by arm and site, so its operative form is an accumulating contamination record.
Nearest alternative: Monitoring, Sensing & Alerting — Cadenced rollups make leakage visible during the study, but the mechanism's evidentiary value depends on the durable incident history rather than continuous sensing alone.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Controlled-trial methodology cohered prospective logging of crossover, spillover, and treatment drift so the delivered contrast can be distinguished from the planned arms.
Related originating lineages:
- Medicine & Healthcare — Clinical-trial operations institutionalized monitoring of treatment contamination during execution.
Review resolution: Experimental-design logging of treatment contamination is primary, with clinical trials as a genuine institutional lineage rather than merely a later use.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Diffusion (or imitation) of treatments is one of Cook and Campbell's classic threats to internal validity: when the control group gains access to elements of the treatment, the difference between arms narrows and the effect is biased toward the null. Documenting it is what lets the team argue the true effect is at least as large as the contaminated estimate. ↩