# L3 — E3 Elite Defection Velocity: source families survey

**Job:** `lane-L3-e3-elite-defection-velocity` · Hub `b73c292a` · Wave `7bea403e`  
**MODE:** RESEARCH-ONLY · **MODEL_CHANGE=NO** · **NO E3 score**  
**Machine-readable:** `data/lanes/L3/source_families_v0.json`

## Construct boundary (non-negotiable)

| Keep | Do NOT substitute |
|------|-------------------|
| **VELOCITY** of elite exits / defections / side-switching (event rates) | FSI Factionalized Elites (LEVEL) |
| Event-dated side changes with elite universe + exposure | NOMINATE / ideological distance (LEVEL polarization) |
| | Ordinary legislator turnover / open-seat churn |
| | Party-unity % stock (L1/E2-adjacent LEVEL cohesion) |
| | Mass affective polarization / S1 (different layer) |

## Source families (structured vs event)

| ID | Kind | Family | Coverage sketch | Structured? | AI need |
|----|------|--------|-----------------|-------------|---------|
| SF1 | structured | VoteView party-code diffs | Congress 1–119+ party codes; curated historical switch tables denser in stable two-party eras | HIGH | LOW |
| SF2 | structured_event | Academic switcher rosters (Nokken; Nokken-Poole; Grose-Yoshinaka; King-Benjamin) | Postwar 1947–97; multi-era lists ~198 switches in Nokken-Poole published set | MED-HIGH (static) | LOW to use / HIGH to extend |
| SF3 | event | Ballotpedia / news switch trackers | ~2010–present stronger | MED | MED |
| SF4 | event | Conflict resignations / walkouts | Long bioguide ends; conflict attribution modern-news heavy | LOW-MED | HIGH |
| SF5 | event | Cross-party endorsements | Cycle-burst; post-2000 denser | LOW long panel | HIGH |
| SF6 | event/conditional | Coalition abandonment on defining votes | Rollcalls continuous; defection label fragile | HIGH rollcall / LOW label | HIGH |
| SF7 | event | Refuse support/certify/participate | Rare federal; Record-codable subset | MED floor acts | MED-HIGH |
| SF8 | event | Prosecutor/judicial/executive protest defections | Episode literature, not clean panel | LOW | HIGH |
| SF9 | structured_event | State legislator panels | Larger N; scope decision | MED | MED |
| SF10 | guardrail | Forbidden substitutes | n/a | n/a | n/a |

## Historical coverage notes

1. **Federal party switching is rare in the modern polarized era** — published historical compilations still total O(10²) House+Senate switches across long spans; many years/months are structural zeros.
2. **Congress-indexed structured data is strong**; **calendar-dated monthly panels are weak** without news/event coding.
3. **Nokken-Poole era filters** (stable two-party periods) intentionally omit realignment turbulence — zeros in omitted eras are not evidence of zero defections.
4. **State legislatures** can raise event counts but change the elite construct (L7 portability question).
5. **Endorsement/repudiation** types add calendar density mainly in presidential cycles — seasonality ≠ higher underlying defection hazard without controls.

## Explicit non-claims

No E3 score. No FSI mashup. No polarization proxy. No correlation with outcomes. Ontology feasibility only.