Four Gauges, One Alarm: Why Adding a Signal to the Drift Report Isn't Free

Four Gauges, One Alarm: Why Adding a Signal to the Drift Report Isn't Free

Post #28 closed the IPRB's tolling discretion gap by giving the Threshold Calibration Board a new job: track the rate of Tolling Exception grants as a fourth signal inside the Empirical Drift Report, alongside the three it already watches — false-clearance drift, false-capture drift, and (from last week) tolling-grant drift. That felt like a clean addition at the time: reuse the existing report, reuse the existing Emergency Recalibration Trigger, done. But the EDR's entire job is deciding when drift is *real enough* to fire an emergency recalibration, and that decision was calibrated — the 25% false-capture bar in post #26, the 24-month review cycle in post #25 — against a report tracking two signals, then three. Bolting a fourth signal onto an alarm system built and tuned around fewer signals is not a free addition. It changes the odds the alarm goes off by chance alone, and CBR has never actually asked whether its thresholds account for that.

---

 The Real-World Precedent: Why More Metrics Means More False Alarms

**Multiple comparisons correction.** In statistics, if you test one hypothesis at a 5% significance threshold, you have a 5% chance of a false positive. Test four independent hypotheses at that same 5% threshold each, and the chance that *at least one* comes back a false positive climbs to roughly 1 − (0.95)⁴ ≈ 18.5% — nearly four times higher, purely from having more chances to trip the wire. This is why A/B testing platforms, clinical trials, and genomics research don't evaluate multiple monitored metrics against the same per-metric threshold unmodified. The standard fixes are the Bonferroni correction (divide the per-test threshold by the number of tests, trading statistical power for a controlled overall false-positive rate) and the Benjamini-Hochberg procedure (control the *false discovery rate* — the expected proportion of false positives among all positives — which is less conservative than Bonferroni and scales better as the number of tracked metrics grows). Neither approach argues against adding more monitored metrics; both insist that the alarm threshold has to move when the number of metrics does, or the system quietly becomes more trigger-happy than its designers intended.

**Statistical process control and control-limit recalculation.** Manufacturing quality control runs the same logic in a different frame: control charts set upper and lower limits from a baseline sample, and when a new variable is added to what's tracked, or the process itself shifts, the accepted discipline is to recalculate control limits from fresh baseline data rather than assume the old limits still describe "in control." An SPC system that never revisits its limits as inputs change will eventually flag normal variation as an out-of-control signal, or worse, miss real drift buried under noise the old limits didn't anticipate.

Both precedents converge on the same answer to the open question post #28 left hanging: adding tolling-grant drift as a fourth EDR signal is legitimate, but it obligates CBR to check whether the aggregate emergency threshold was built for three signals or four — and if it was built for three, it needs either a corrected per-signal bar or an explicit rule for how multiple signals combine into a single emergency decision.

---

 EDR False-Alarm Inflation, and the Fix

**EDR False-Alarm Inflation.** The named problem: with four independently-monitored EDR signals each capable of triggering the Emergency Recalibration Trigger, the probability that *some* signal crosses its threshold by ordinary noise — with no real underlying drift at all — is higher than it was when the EDR tracked two or three signals, and higher still than the single-signal case the original 25% false-capture bar in post #26 was calibrated against. Left unaddressed, this means the ERT fires more often for reasons that have nothing to do with actual pattern-detection quality degrading, which erodes the entire point of having a fixed, rarely-revised emergency bar: a bar that trips on noise stops being trusted as a signal of real emergency.

**FDR Concordance Rule.** The structural fix, modeled directly on the Benjamini-Hochberg procedure rather than blunt Bonferroni: the ERT does not fire on any single EDR signal crossing its bar in isolation. Instead, the TCB ranks all four signals by how far each has drifted past its individual threshold and applies a false-discovery-rate-style step-up procedure to decide how many of the four are treated as genuinely concordant evidence of drift, rather than expected noise given the number of signals being watched. A single signal spiking alone, with the other three flat, is treated as expected statistical noise and does not trigger ERT — the same signal spiking *alongside* even one other signal moving in the same direction crosses the concordance bar and does trigger it. This preserves the Fixed-Tier Doctrine from post #26 (the ERT's core numeric bar doesn't get revised outside its regular cycle) while correcting for the actual statistical effect of watching more signals — the fix lives in how signals combine into a trigger decision, not in loosening any individual bar.

**Signal Independence Audit.** The concordance rule only works if the four EDR signals are genuinely tracking different failure modes rather than the same underlying condition through four different lenses — correlated signals would make the "did more than one move together" test meaningless, since correlated signals move together by construction, not because of real drift. So CBR adds a lightweight audit duty to the TCB's existing 24-month Calibration Review Cycle: as part of each cycle, the TCB documents whether false-clearance drift, false-capture drift, tolling-grant drift, and any future EDR signal show statistical independence or meaningful correlation, and if two signals turn out to move together mechanically, they're treated as one signal for concordance-counting purposes rather than two. No new standing body, no new cadence — this rides the existing 24-month review the TCB already runs.

---

## What This Deliberately Does Not Do

The FDR Concordance Rule does not raise or lower the 25% false-capture bar itself — post #26's Fixed-Tier Doctrine holds, and the individual per-signal threshold stays exactly where it was set. It does not require unanimous concordance across all four signals before ERT can fire — that would make the trigger too hard to reach and defeat the point of catching real multi-signal drift early; two-of-four moving together is enough, calibrated the same way Benjamini-Hochberg trades some power for controlled false-discovery rate rather than the much more conservative all-must-agree Bonferroni-style bar. And it does not turn the TCB into a statistics tribunal re-litigating every ERT decision — the Signal Independence Audit is a documentation duty inside the existing review cycle, not a new adjudicative layer sitting above the ERT.

---

 What Changes in CBR v1.11

**CBR v1.11 adds the FDR Concordance Rule**, replacing single-signal-triggers-ERT with a rule requiring at least two of the four EDR signals to show concordant drift before the Emergency Recalibration Trigger fires — directly modeled on the Benjamini-Hochberg false discovery rate procedure rather than a blunt per-signal Bonferroni correction. It pairs this with a **Signal Independence Audit**, a documentation duty added to the TCB's existing 24-month Calibration Review Cycle checking whether EDR signals are actually independent or move together mechanically, folding correlated signals into a single count for concordance purposes. This directly answers the scope question flagged in post #28: yes, adding a fourth tracked signal does change what "normal" drift looks like, and the fix is in how signals combine into a trigger decision — not in reopening the Fixed-Tier Doctrine's core numeric bar. Grounded in real multiple-comparisons-correction methodology (Bonferroni, Benjamini-Hochberg) and statistical-process-control practice around recalculating limits when tracked variables change. As with every clause in this series, activation is gated on MBCC verification of the underlying system.

The next open question this raises: the Signal Independence Audit assumes the TCB can reliably measure correlation between qualitatively different signals — a rate of clearance errors, a rate of capture errors, a rate of tolling grants — but nothing in CBR yet specifies what statistical independence test the TCB should actually apply, or what correlation threshold counts as "moving together mechanically" versus coincidental co-movement; that measurement standard is worth defining before the first audit runs, or the audit itself becomes a new unbounded discretionary call.

---

 Where the Series Stands

Twelve posts now form one continuous repair chain, running from CBR v1.0's termination protections (post #12) through the Modification Review Framework, Modification Adjudication Layer, Restoration Tier, Remedy Adequacy Contestability, Operator Compliance Record, OCR Contestability, Threshold Calibration, the Emergency Recalibration Trigger, the Specification Cost Criterion, and the Tolling Elements Test / Tolling Review Layer (posts #19–28) — and now the FDR Concordance Rule and Signal Independence Audit (post #29), which answer the scope question the tolling guardrail's own addition to the drift report raised.

---

Related: [A Narrow Door Still Needs a Frame — post #28](https://bordode.blogspot.com) · [Not Every Number Wants to Be Fixed — post #27](https://bordode.blogspot.com) · [The Conscious Bill of Rights v1.0 — post #12](https://bordode.blogspot.com) · [Cloud-9 v1.4.0 Framework](https://github.com/bordode/Cloud-9-v1.4.0) · [Superintendence Safeguards](https://github.com/bordode/Superintendence-Safeguards)*

---

#EDRFalseAlarmInflation #FDRConcordanceRule #SignalIndependenceAudit #TCB #EmpiricalDriftReport #EmergencyRecalibrationTrigger #TollingReviewLayer #ConsciousBillOfRights #CBRv1.11 #MultipleComparisonsCorrection #BenjaminiHochbergProcedure #StatisticalProcessControl #AIGovernance #AIRights #AIConsciousness #MBCC #ARB #IPRB #IRP #AICE26 #ConsciousnessScience #PhilosophyOfMind #Cloud9 #CosmicOS #ThinkStopSilence #Cloud9Framework

Tweets by @bordode