## 1. Artifact boundary and type **Type:** KPI definition. **Boundary:** The calculation of 30-day repeat attendance from the library's workshop records (workshop date, anonymous participant ID, attendance status) for free adult-learning workshops only. **Material unknowns:** How "eligible" is flagged in the records, the exact attendance-status values, whether anonymous IDs persist across workshops, and whether a workshop identifier exists. ## 2. The artifact (proposed KPI definition) **KPI name:** 30-Day Repeat Attendance Rate | Element | Definition | Basis | |---|---|---| | Measured construct | Share of new eligible attendees who attend a later eligible workshop within 30 days of their first eligible attendance | Supplied | | Eligible workshop | A workshop record the library classifies as a free adult-learning workshop. The classification field is not supplied. | Supplied (concept); field is an unknown | | Attendance | A record whose status means "attended". Registered, cancelled, and no-show do not count. Actual status labels are unknown. | Proposed | | First eligible attendance | Earliest workshop date, per participant ID, among attended eligible records in the data | Proposed | | Cohort | Participants whose first eligible attendance falls in the reporting period | Proposed | | Maturity rule | A participant enters the cohort only if first date + 30 days ≤ data extract date. Later participants are reported as "not yet matured". | Proposed | | Returning participant | A cohort member with at least one attended eligible record dated 1 to 30 days after their first date (day 0 = first date) | Supplied ("within 30 days"); day counting is proposed | | Same-day sessions | A second session on the first date does not count as a return | Proposed | | Duplicates | Multiple records for one participant on one date count as one attendance | Proposed | | **Formula** | **Returning participants ÷ Cohort participants × 100** | | | Unit | Percentage of participants (distinct anonymous IDs) | | | Numerator | Distinct IDs in the cohort meeting the return rule | | | Denominator | Distinct IDs in the matured cohort | | | Time window | 30 calendar days after first eligible attendance | Supplied | | Source fields | participant_id, workshop_date, attendance_status, eligibility indicator (field unknown) | Supplied except eligibility | | Missing data | Records with a null ID, date, or status are excluded and counted in a reported "excluded records" figure | Proposed | | Zero denominator | Report "Undefined (no matured first-time attendees)". Never report 0%. | Proposed | | Target / threshold | Not supplied | | | Measured result | None available | | **Named limitations:** - The rate measures repeat attendance only, not learning, satisfaction, or retention. No proxy substitution is made. - If IDs are not persistent, returns are undercounted. If IDs are shared, they are overcounted. - "First" attendance means first in the available records. A participant who attended before the data begins is treated as new unless the data covers a lookback period. **Synthetic example** (invented for checking the rule; not client data). Sample assumptions: reporting period is first attendances from 1 to 15 June 2026, and extract date is 30 September 2026. | ID | Records (attended unless noted) | Cohort? | Returning? | |---|---|---|---| | P1 | 2 Jun; 20 Jun | Yes | Yes (day 18) | | P2 | 3 Jun; 3 Jul | Yes | Yes (day 30) | | P3 | 3 Jun; 4 Jul | Yes | No (day 31) | | P4 | 5 Jun; 12 Jun status "no-show" | Yes | No | | P5 | 8 Jun, two sessions same day | Yes | No | | P6 | 20 May; 10 Jun | No (first date before period) | n/a | | P7 | 10 Jun; 25 Jun at a workshop flagged ineligible | Yes | No | Sample calculation: 2 ÷ 6 × 100 = 33.3%. This figure illustrates the formula only. ## 3. Verification record | Check | Method | Expected condition | Observed result | Status | |---|---|---|---|---| | Day-30 boundary inclusive | Manual sample calculation | P2 (day 30) counts, P3 (day 31) does not | P2 counted, P3 not | PASS | | Same-day session excluded | Manual sample calculation | P5 not returning | P5 not returning | PASS | | Non-attended status excluded | Manual sample calculation | P4 not returning | P4 not returning | PASS | | Ineligible later workshop excluded | Manual sample calculation | P7 not returning | P7 not returning | PASS | | Earlier first attendance leaves cohort | Manual sample calculation | P6 excluded from denominator | P6 excluded | PASS | | Formula arithmetic | Manual sample calculation | 2 ÷ 6 = 33.3% | 33.3% | PASS | | Zero-denominator rule | Manual inspection of rule text | Empty cohort yields "Undefined", no division | Rule specifies this | PASS | | Run on real library records | Query on client data | Rate computed with excluded-record count | No client data or query tool available | NOT RUN | | Eligibility field mapping | Compare rule to client schema | An identifiable eligibility field or rule exists | Field and rule not supplied | BLOCKED | | ID persistence across workshops | Review of ID assignment method | Same person keeps one ID | ID assignment method not supplied | BLOCKED | All PASS results come from manual inspection of synthetic data. No code or tool was executed, and nothing has been checked against real records. ## 4. Design trace and limits **Retained transfer (Direct):** Cohort follow-up from epidemiology and product retention analysis. An index event (first eligible attendance) is paired with a fixed follow-up window, and only participants with a fully observed window are counted. - **Concrete change:** The index-event rule, the maturity rule, and the exclusion of "not yet matured" participants. - **Breakdown condition:** Unlike continuous exposure, workshops are scheduled. If few eligible workshops run within 30 days of someone's first visit, the rate falls without any change in engagement. The KPI does not adjust for this. **Other lenses:** Personal, Symbolic, and Fantasy analogies produced no change that the conventional cohort definition didn't already cover, so they were discarded. **Unresolved constraints:** The eligibility rule, the status labels, ID persistence, and any lookback period for identifying true first attendance are unknown. Until they are supplied, the definition is structurally consistent but not validated against client data.