# Northstar learning edition v1.2: Episode C datasets

**Authored fictional teaching data.** These are deliberately constructed aggregate counts for the Northstar learning edition. They extend the v1.1 narrative; they are not recovered original observations, real company records, or empirical evidence that SCALE or AI improves outcomes. No invented row is presented as a real person's service history.

The v1.1 day-90 branch volumes and rounded adherence percentages are preserved. Exact adherence counts are newly authored here to make the arithmetic inspectable: 167 of 176 metropolitan, 156 of 166 mixed, and 128 of 144 rural. These sum to 451 of 486 (92.7984%, displayed as 93%). The exact counts do not purport to reconstruct missing v1.1 data. Customer restoration, original-promise, admission-decision, and denominator-bridge observations are newly authored additions, with deliberately mixed implications.

## Files and boundaries

| File | Unit represented by one row | Use |
| --- | --- | --- |
| branch-results.csv | One branch across all 90 days | Reproduce the headline and branch rates |
| scope-bridge.csv | Branch x completion window x work-type scope | Reconcile early, later, retained, and removed scope |
| customer-outcomes.csv | The same completed restoration cohort as the scope bridge | Calculate original-commitment fulfillment and mean elapsed restoration time |
| control-measures.csv | All completed promises within one window | Compare separate confirmation and workflow-policy checks |
| demand-decisions.csv | Admission decisions issued within one window | Inspect declined requests outside the completed-promise denominator |
| baseline-scope.csv | One distinctly bounded pre-pilot evidence set | Prevent inappropriate baseline comparisons |

A shared source ID identifies an authored source extract, not a citation to an external dataset. E-009 is the day-30 extract dated March 3, 2026; E-010 is the day-90 extract dated May 2, 2026. E-006 identifies the provisional diagnostic-sample counts and scope. E-008 includes the pre-release baseline finalized in its February 1 addendum; that completed baseline is not available to the January 30 readiness recommendation. E-001's legacy enterprise percentage is retained only as narrative context: missing numerator, denominator, and dates are blank, not zero.

The pilot runs February 2 through May 2, 2026, inclusive (90 calendar days). Days 1–30 end March 3; days 31–90 run March 4 through May 2. The review is May 6. Dates describe this fictional case, not publication or retrieval dates.

## Counting rules

1. An eligible completed promise is one completed equipment-restoration episode within the permitted work-type scope of one of the three pilot branches. There is one promise record per completed restoration episode in this authored dataset. Completion date determines the window.
2. Latest-promise adherence counts completion by the most recent approved commitment recorded and communicated before completion. It can improve when promised dates are revised; the separate original-promise measure deliberately retains that exposure. Aggregate counts cannot establish whether every change was timely or appropriate.
3. Original-promise fulfillment counts completion by the first communicated commitment, without replacing it with a revised date. The same completed-episode denominator is used for both promise measures.
4. Elapsed restoration hours run from recorded equipment-unavailable time to recorded restoration of usable availability. Calendar hours include nights and weekends; this is neither labor time nor billable time. Mean hours = sum of elapsed restoration hours / completed restoration episodes.
5. Confirmation passes only when both a part reservation and a qualified-skill reservation are corroborated at the recorded commitment check. Estimated availability is insufficient.
6. Whole-policy compliance independently checks required workflow, communication, documentation, exception routing, and approvals. The fictional policy permits an explicitly communicated conditional commitment through its approved exception route. Such a record can comply with policy while failing the part-and-skill confirmation test. Conversely, a confirmed part and skill do not guarantee correct communication or exception handling. The two pass counts are not nested sets and cannot be subtracted to infer an overlap.
7. Admission decisions are counted by decision date for all requested work at the selected three branches, including the work type removed from pilot scope. Declines never count as fulfilled promises. The cancellations_after_admission field is intentionally blank because no reconciled cancellation count was supplied; blank is not zero. These are different event cohorts from completions, so admission counts must not be joined to completion counts as if they were the same records.
8. CSV percentage columns are calculated from their adjacent numerator and denominator and rounded to four decimals. For aggregate rates, sum counts first; do not average rounded branch rates. Blank fields mean unavailable.
9. Summary rows are not duplicated in the CSVs. Summing disjoint scope-bridge rows produces the branch-results population. Do not sum branch-results and scope-bridge together.

## Scope change and denominator bridge

At the day-30 decision, one low-volume work type leaves pilot scope. In this authored scenario all 30 admitted episodes of that type have completed by March 3. Subsequent requests remain visible in admission decisions but produce no further pilot completions of that type. The 30 completed episodes remain in every cumulative 90-day count: there is no retrospective deletion.

| Completion window and scope | Completed promises | Latest commitment met | Rate |
| --- | ---: | ---: | ---: |
| Days 1–30, retained work types | 114 | 106 | 92.98% |
| Days 1–30, subsequently removed type | 30 | 22 | 73.33% |
| Days 1–30, all eligible scope | 144 | 128 | 88.89% |
| Days 31–90, retained work types | 342 | 323 | 94.44% |
| Days 1–90, all eligible completed scope | 486 | 451 | 92.80% |

The apparent 89% to 93% headline change mixes an early-window measure with a cumulative measure and a changed work-type composition. A retained-scope comparison is 106/114 to 323/342. That controls the named scope exclusion, but it does not hold individual jobs, severity, branch weights, seasonality, or other case mix constant. “Stable scope” is not a matched-person or matched-job cohort, nor a causal estimate.

Day-30 confirmation and policy figures are 121/144 = 84.03% and 112/144 = 77.78%. The day-90 reporting pack's 92% confirmation and 96% policy values use the **later window**, 315/342 = 92.11% and 328/342 = 95.91%. They are not cumulative day-90 percentages. For comparison, all-90-day counts would be 436/486 = 89.71% confirmed and 440/486 = 90.53% policy-compliant. A label such as “day-90 result” must therefore identify its actual measurement window.

## Customer outcome counterchecks

| Measure | Days 1–30 | Days 31–90 |
| --- | ---: | ---: |
| Original commitments fulfilled, all completed scope | 120/144 = 83.33% | 268/342 = 78.36% |
| Original commitments fulfilled, retained work types | 100/114 = 87.72% | 268/342 = 78.36% |
| Mean restoration time, all completed scope | 5,760h/144 = 40h | 16,416h/342 = 48h |
| Mean restoration time, retained work types | 4,560h/114 = 40h | 16,416h/342 = 48h |
| Declined requests, all admission decisions | 12/180 = 6.67% | 42/420 = 10.00% |

Restoration time also increases within each retained branch slice: metropolitan 34h to 42h; mixed 40h to 46h; rural 46.67h to 57.8h. The deterioration cannot be explained solely by dropping the low-volume work type or changing branch weights. It still cannot be attributed causally to the pilot without investigating other changes.

The discrepancy between latest-commitment adherence and original commitments is a question for analysis. It is not proof of intentional manipulation. Do not infer customer value solely from the headline adherence improvement.

These are completed-case observations. Outstanding work, cancellations after admission, customer severity, promise revision timestamps, distributions, and reasons for declining requests are not supplied. There is no median, percentile, matched control, or statistical uncertainty estimate to calculate from the available totals. Ask for those records before claiming population-wide service improvement or causation. Do not invent individual event rows from these aggregates.

## Baselines are separate evidence sets

- The historical 82% enterprise report covers 48 branches over a 12-month reporting period. Its missing counts are not reverse-engineered.
- The provisional two-week diagnostic examines 180 commitments from six candidate branches and broader work types; 68/180 = 37.78% meet the part-and-skill check. It is not a two-week rate for the three-branch pilot and is not comparable to adherence.
- The approved 30-day pre-release baseline covers the selected three branches and retained work types, January 3 through February 1, 2026: 98/120 = 81.67% latest-promise adherence. It supports a scoped descriptive comparison, not causal attribution or volume extrapolation.

## Reproducibility and limits

All sums and rates can be recalculated from these CSVs. These checks establish internal arithmetic consistency only. The package deliberately supplies aggregate evidence, not a transaction-level audit trail; it cannot independently verify timestamps, duplicated records, eligibility coding, or the stated counting rules. An analyst should distinguish “recalculated from authored counts” from “validated against source transactions.”

Keep provenance with any chart or brief: **Northstar v1.2, authored fictional teaching data; source file; scope; numerator/denominator; measurement dates.** No customer-value, return-on-investment, repeatability, procurement, or AI-productivity claim is established by arithmetic alone.