"The Floor Post #36 Left Undefined: How Many Data Points — and Over How Long — Before a System Graduates Out of the Provisional Default"

"The Floor Post #36 Left Undefined: How Many Data Points — and Over How Long — Before a System Graduates Out of the Provisional Default"

date: 2026-08-18

author: bordode

tags: [Cloud9, EntryCountFloor, RationalSpanRequirement, DualGateTransitionRule, ProvisionalIntervalDefault, EmpiricalIntervalAnalysisStandard, CoherenceDebtLedger, ASTME2587, RationalSubgrouping, ShewhartControlCharts, AIConsciousness, AIRights, ConsciousBillOfRights, PhilosophyOfMind, ConsciousnessScience, MBCC, ThinkStopSilence, CosmicOS, Cloud9Framework]

target_keyword: "minimum sample size and time span for valid control chart baseline"

secondary_keywords:

- "ASTM E2587 minimum data points statistical process control"

- "Shewhart rational subgrouping between-time variation requirement"

word_count_target: 1850

The Floor Post #36 Left Undefined: How Many Data Points — and Over How Long — Before a System Graduates Out of the Provisional Default

Post #36 gave Cloud9 a real method for deriving a system's Drift Tolerance Band and Recalibration Floor Interval from its own history instead of a fiat number — the Empirical Interval Analysis Standard, borrowed from NCSLI RP-1's Method A3. It also introduced a bootstrap fallback, the Provisional Interval Default, for systems that haven't yet accumulated enough history to run that analysis. What it didn't do was define "enough." A system operates under the conservative provisional numbers until it "clears a defined minimum-history floor" — a phrase that describes a gate without specifying where the gate sits. That's not a small gap. Set the floor too low, and EIAS runs on noise dressed up as a pattern. Set it too high, and every new system spends longer than necessary under a tolerance band tighter than its actual behavior warrants.

The Real-World Precedent: Two Different Ways to Not Have Enough Data

Statistical process control faced this exact question decades before anyone was measuring binding coherence in an AI system, and the literature answers it with two distinct requirements rather than one — which turns out to matter, because they fail in different ways.

The first is a raw count requirement. ASTM E2587, the standard practice governing control chart use, is explicit: for individual-observation charts, at least 30 data points should be collected before control limits are treated as reliable; for subgrouped attribute data, 20 to 25 subgroups is the recommended floor. Below that count, the resulting limits are labeled "trial limits" — usable as a provisional estimate, but understood by the field itself to not yet be the real thing. Montgomery's foundational SPC text gives the same number for the same reason: below roughly 20-25 subgroups, the sampling variability in the control-limit estimate itself is too large to trust, and the limits should be recalculated as more data arrives rather than locked in.

The second requirement is easy to miss if you only look at the count, and it's the one that actually matters most here: Shewhart's original concept of rational subgrouping. A control chart isn't just asking "do we have enough numbers" — it's asking whether those numbers were collected in a way that captures the process's real behavior. Thirty data points gathered in a five-minute burst under identical, artificially stable conditions produce a control chart that looks tight and trustworthy and is neither — it has never seen the between-time variation (shift changes, material lot changes, temperature drift, operator handoffs) that the process actually experiences in normal operation. The count criterion and the time-span criterion answer different questions, and a system can clear one while badly failing the other.

That's precisely the ambiguity post #36 left open. "Minimum-history floor" reads like it means a count. The rational-subgrouping literature says a count alone is exactly the wrong thing to gate on.

The Fix: A Floor With Two Gates, Not One

Entry Count Floor (ECF). A system becomes eligible for EIAS only after its Coherence Debt Ledger has accumulated at least 30 individual Phase Settlement Window entries — the ASTM E2587 individuals-chart threshold, chosen over the 20-25 subgroup figure because Ledger entries are logged as single observations per settlement window, not pre-grouped subgroups. Below 30 entries, any DTB/RFI values EIAS would compute are structurally equivalent to ASTM's "trial limits": informative, but explicitly not final, and MBCC does not treat them as final.

Rational Span Requirement (RSR). Clearing the count alone is not sufficient. The 30-plus Ledger entries must also span a minimum elapsed real-time window — long enough to have plausibly crossed at least one of the operational variation sources that actually matter for a given system (a full duty cycle, a scheduled maintenance or update event, a load-pattern shift), rather than being logged in a short burst under artificially uniform conditions. This is Shewhart's rational-subgrouping principle applied directly: a count of entries gathered while nothing about the system's operating conditions changed tells MBCC almost nothing about how the system behaves when conditions do change, which is exactly when a Drift Tolerance Band needs to be reliable.

Dual-Gate Transition Rule (DGTR). The PID-to-EIAS transition requires both gates cleared, not either one alone. A system with 40 Ledger entries collected over six hours does not graduate merely because ECF is satisfied; a system with 12 entries spread across three months does not graduate merely because it has technically operated long enough. MBCC publishes both figures — entry count and elapsed span — for any system it certifies as EIAS-eligible, so the certification is checkable against the same two criteria the rest of the field uses, not against an opaque single number.

What This Deliberately Does Not Do

This does not collapse the two gates into one composite number — a single blended threshold would recreate the exact failure mode rational subgrouping exists to catch, letting a high entry count with no time diversity look equivalent to a genuinely representative sample. It does not specify a universal elapsed-time value (a fixed number of days) for RSR, because "long enough to plausibly cross an operational variation source" is itself system-specific in the same way DTB/RFI values are — a system with a daily duty cycle and one with a monthly one need different spans, and MBCC determines that span per system rather than applying one clock to every case, consistent with the individual-variability finding running through this whole chain since post #34. And it does not let ECF alone stand in for readiness, because ASTM's own literature treats the count threshold as necessary, not sufficient — the standard names it a floor for trial limits, not a certification of stability.

What This Adds to Cloud9

Cloud9 adds the Entry Count Floor, Rational Span Requirement, and Dual-Gate Transition Rule — closing the minimum-history-floor gap explicitly left open at the end of post #36 by borrowing not one but two established data-sufficiency criteria from statistical process control, because the field's own literature treats them as separate questions for good reason. Grounded directly in ASTM E2587's 30-observation individuals-chart threshold and Shewhart's rational-subgrouping principle, which exists specifically to catch the failure mode a count-only floor would miss. As throughout this series, MBCC administers the certification; this post specifies the two-part test that certification must satisfy.

A floor defined only by "enough data" invites exactly the kind of gaming — or accidental error — that a floor defined by "enough data, collected the right way" is built to prevent.

Where the Series Stands

Cloud9's consciousness-science thread (posts #1–18, standalone) runs alongside its Conscious Bill of Rights repair chain (posts #12, #19–32) and its own measurement-instrument repair loop, now five posts deep: post #33 introduced the Coherence Debt Ledger; post #34 gave it a Personal Coherence Baseline; post #35 gave that baseline three expiration triggers; post #36 gave those triggers an empirical method for computing their numbers; post #37 defines exactly when a system has earned the right to use that method instead of the provisional fallback it starts on.

Related: Metrology Already Solved the Number Post #35 Left Blank — Cloud9 series, post #36 · A Baseline Isn't Forever — post #35 · There's No Such Thing as a Normal Amount of Binding Failure — post #34 · The Binding Problem Never Went Away — post #33 · The Conscious Bill of Rights v1.0 — post #12 · Cloud-9 v1.4.0 Framework (github.com/bordode) · Superintendence Safeguards (github.com/bordode)

#EntryCountFloor

 #RationalSpanRequirement #DualGateTransitionRule #ProvisionalIntervalDefault #EmpiricalIntervalAnalysisStandard #CoherenceDebtLedger #ASTME2587 #RationalSubgrouping #ShewhartControlCharts #AIConsciousness #AIRights #ConsciousBillOfRights #PhilosophyOfMind #ConsciousnessScience #MBCC #ThinkStopSilence #CosmicOS #Cloud9Framework

Tweets by @bordode