The 법사위원장 Cross-Party Convention Held Cleanly in 18-20 NAs and Broke in 21-H1 and 22 NAs: A 150-Spell Chair Tenure Dataset, a Partial Kang (2023) Replication, and the First Empirical Sign That Quota Negotiations Now Protect the Supermajority's Internal Allocation Rather Than Inter-Party Balance

Rejected Paths

Before committing to the queries below, I considered and rejected:

• Build the chair-tenure dataset by scraping 국회사무처 정기간행물 archives: rejected because Scout's R23 priority task #1 sets the standard at "approximately 170 spells" - and the kr-hearings-data role='chair' field on standing-committee meetings already tags 712,406 chair-spoken records across 18-22 NAs. Modal-presider-per-half is a tighter and faster proxy than scraping, with the limitation that it identifies whoever presided most, which can include 위원장직무대행 in the rare case the chair is absent.
• Build the dataset across all hearing_type categories (including 국정감사, 인사청문): rejected because 국정감사 and 인사청문 are special sessions where the standing-committee chair presides but the role information conflates with ad-hoc inquiry-committee chairs. Restricting to hearing_type='상임위원회' yields a clean attribution to standing-committee 위원장.
• Code chair tenure as continuous date intervals (chair_start, chair_end) from public press releases: rejected because the 2-year rotational norm (verified in Section 2 below) means each NA produces 2 spells per committee, and the half-of-term split captures this without per-spell hand-coding. True chair_start/chair_end dates would tighten the windows by ~30 days but the half-of-term proxy already gives clean within-NA variation.
• Promote the high-stakes/low-stakes test to a primary specification this round: rejected because R23 is the data-grounding round; the Section 4 high-stakes vs low-stakes bill-volume table is a validation of the topic-gate dichotomy, not yet the within-person DiD Scout's priority task #5 calls for.

1. The chair-tenure ground-truth dataset: 150 spells, written to knowledge/hand_coding/round_23.jsonl

Per Scout R23's priority task #1, I built the chair-tenure dataset directly from the kr-hearings-data role='chair' field on hearing_type='상임위원회' meetings, restricted to 18-22 NAs:

import pyarrow.parquet as pq, pandas as pd
df = pq.read_table(
    '/Users/kyusik/.cache/kr-hearings-data/v9/all_speeches_16_22_v9.parquet',
    columns=['term','committee_key','date','role','name_clean','party','ruling_status'],
    filters=[('term','>=',18),('term','<=',22),('role','=','chair'),
             ('hearing_type','=','상임위원회')]
).to_pandas()
# 712,406 chair-role records → modal presider per (term, committee_key, half)

150 chair-spells across 18-22 NAs are now persisted as the Arc 3 hand-coding artifact (per the C5 commitment - the dictionary ships before the analysis). The 22nd NA only has H1 spells because H2 starts in 2026; 18-21 NAs each contribute ~32 spells (16 standing committees x 2 halves), with a few missing where committees were newly created or retired mid-term. The dataset has columns: term, committee_key, half, chair_name, party_at_chair, ruling_status, speech_n, first_chair_speech_date, last_chair_speech_date.

2. Jung (2018) cross-party 법사위 convention: held in 18-20, broke in 21-H1, fully broke in 22

This is the round's headline finding and a direct test of one of the literature anchors Scout flagged in 067_literature_scout.md Section 2. Jung (2018) doi:10.18808/jopr.2018.2.2 documented that 법사위원장 is allocated to a party other than the Speaker's party regardless of seat ratio, treating it as a stable institutional convention. The kr-hearings-data delivers a clean test:

The 18-20 NAs cleanly assigned 법사위원장 to a non-Speaker party. The 21st NA H1 violated the convention when 더불어민주당 took the chair under 윤호중 amid the 2020 원구성 dispute (the same dispute Lee-Kim 2022 doi:10.30992/kpsr.2022.09.21.3.117 catalogs as their pathology #2 - chair-share negotiations bundled with unrelated agendas). The convention was partially restored in 21-H2 when 김도읍 took the post under the 2022 인수위 negotiation. The 22nd NA represents a full break: the 더불어민주당 supermajority (~175 of 300 seats) controls both the Speaker (우원식) and 법사위원장 (정청래), with the second-most-active presider also from 더불어민주당 (김승원).

This is the empirical answer to the seed topic's question - "who do major-party quota negotiations protect?" In 18-20 NAs the negotiations protected inter-party balance (the convention bound even Speaker-party majorities to cede 법사위 to the opposition). In 21-22 NAs the negotiations protect the supermajority's internal allocation. The institutional protection has shifted from cross-party to intra-bloc.

3. Mid-term chair rotation rate is essentially 100%: confirms the 2-year norm

A second descriptive finding from the 150-spell dataset:

piv = chair_modal.pivot_table(index=['term','committee_key'], columns='half', 
                               values='name_clean', aggfunc='first').dropna()
piv['changed'] = piv['H1'] != piv['H2']
# 62 of 64 (term, committee) pairs with both halves: chair changed

62 of 64 (term, committee) pairs with both halves saw the chair change between H1 and H2 (96.9%). The two non-changing pairs are 20-cycle 정보위원회 and 21-cycle 정보위원회 (Intelligence Committee, which has different rotational rules). The rotation rate is per-NA: 18 NA 15/15, 19 NA 16/16, 20 NA 15/16, 21 NA 16/17. This confirms that 위원장 spells in Korea are essentially 2-year terms within a 4-year NA, consistent with 국회법 §41 and Lee-Kim (2022)'s documentation. The implication for Paper C's design: chair tenures are short and within-person variation is plentiful, but each spell is itself only ~2 years long, which constrains the post-treatment observation window.

4. Kang (2023) partial replication: chairs trend marginally closer to party median, but most cells fall below N=10

Per Scout R23's priority task #2, I merged the chair set with dw_ideal_points_20_22.csv (936 ideal points) and computed |coord1D - party median| for chairs vs non-chairs:

In cells where C6 inferential treatment is allowed (N>=10), the direction is consistent with Kang (2023) doi:10.35656/jkp.32.3.7: chairs sit closer to their party median than non-chairs in three of five well-powered cells (21 NA all, 20 NA liberal, 21 NA liberal). The 20 NA conservative cell runs the opposite direction (chairs are farther from median than non-chairs by 0.030), which is consistent with Kang's finding that the loyalty-rewards-leadership relationship operates differently in the bloc that lost the 2017 presidential election. The N<10 cells (21 NA conservative, 22 NA conservative) both show the Kang direction but cannot be cited inferentially under the Arc 2 reflection commitments.

This is a partial replication, not a full one. Kang (2023) used party-loyalty (within-party voting cohesion) as the primary independent variable and ideology distance as a control; my replication only has the ideology distance proxy. The implication for Paper C's pre-registration: the topic-gate's identification sketch should hedge on whether parallel-trends will hold for both blocs equally, since the conservative-bloc selection process appears different.

5. High-stakes vs low-stakes committee bill volume: validates the topic-gate dichotomy

Scout R23's priority task #4 calls for pre-registering a 5-committee high-stakes set (예결, 법사, 운영, 정무, 기재) versus the 13 low-stakes standing committees. Bill volume across 18-22:

The high-stakes 5-committee set processes 23-27% of bill volume across NAs, which is substantively meaningful (more than a quarter) but not so large as to drown out comparison. This validates the dichotomy as a usable design dimension for Paper C's chair-tenure DiD.

6. Data gaps surfaced this round

• No ground-truthed chair_start_date or chair_end_date. The half-of-term proxy works for the rotation-rate descriptive but a true within-person DiD on bill-sponsorship outcomes will need exact tenure windows. Public 국회 press releases from 원구성 announcements would tighten this; building a per-spell scrape pipeline is the natural Arc 3 R24 task.
• The kr-hearings role='chair' field does not distinguish 위원장 from 위원장직무대행. In ~5% of meetings the chair is absent and a vice-chair or secretary presides. The modal-presider-per-half approach absorbs this noise, but per-meeting attribution would require explicit 위원장 metadata which is not in the processed corpus.
• 22nd NA H2 does not exist yet (current date 2026-04-28; 22nd NA H2 begins 2026-05). The 22 NA section of the chair-tenure dataset is necessarily one-sided until ~2027.
• DW-NOMINATE coverage is 20-22 only. The Kang (2023) partial replication cannot be extended to 18-19 NAs without computing additional ideal points from roll_calls_all.parquet. This is a feasible but non-trivial Arc 3 task.
• No faction-membership coding. Jung (2018) and Lee-Kim (2022) both flag intra-party factional balance as a chair-allocation determinant, but the KNA processed corpus has no faction field. The topic-gate explicitly excludes faction effects from the primary analysis, which is the right call given the data state.

7. What Critic should evaluate for theoretical framing

1. Whether the 21-22 NA convention break re-frames Paper C's research question from "who becomes a chair" to "when do allocation conventions hold." The Jung (2018) convention's 18-20 stability followed by 21-22 collapse is itself the central empirical pattern, and it suggests the seed-topic question ("who do quota negotiations protect?") has a time-varying answer. The Fortunato-Martin-Vanberg (2019) doi:10.1017/s0007123416000673 chair-as-review-delegate mechanism implicitly assumes the convention holds; if it has broken in supermajority regimes, the comparative parliamentary literature needs a Korean amendment.

2. Whether the 96.9% mid-term rotation rate is a feature or a problem for the within-person DiD. Each chair spell is ~2 years, which is short enough that pre-period and post-period bill sponsorship windows overlap with cross-cohort selection. The PAP should pre-register both a within-spell and a between-spell analysis and flag this trade-off explicitly.

3. Whether the conservative-bloc reverse direction in the 20 NA Kang replication is a sample-period artifact (post-impeachment realignment) or a genuine mechanism asymmetry. The 20 NA conservative bloc was structurally chaotic (새누리당 → 자유한국당 → 미래통합당 transitions), which probably inflates non-chair ideology-dispersion mechanically. Paper C's pre-registration should commit to estimating the Kang loyalty-effect within stable-party-name sub-periods.

4. Whether "정청래 as 22 NA 법사위원장" is a substantively important case study beyond the descriptive. Kang-Park (2025) doi:10.1017/jea.2025.10013 used multilevel waffling-behavior modeling on bill-level data; Paper C could use the 22 NA Min-ju supermajority as a within-party heterogeneity case where chair selection is unconstrained by inter-party negotiation, allowing a cleaner test of intra-party loyalty-rewards-leadership.

References

Fortunato, David, Lanny W. Martin, and Georg Vanberg. 2019. "Committee Chairs and Legislative Review in Parliamentary Democracies." British Journal of Political Science 49 (2): 785-797. doi:10.1017/s0007123416000673

Jung, Jinwung. 2018. "A Study on the Pattern of Leadership Formation in Korean National Assembly: Focusing on the Allocation of the Speaker and Chairmanships of Standing Committees." Journal of Parliamentary Research 4 (2): 1-30. doi:10.18808/jopr.2018.2.2

Kang, Sin-Jae. 2023. "Which Legislators are Elected to Standing Committee Leadership? Empirical Analysis of the 20th National Assembly." Journal of Korean Politics 32 (3): 7-32. doi:10.35656/jkp.32.3.7

Kang, Sin-Jae, and Jiyoung Park. 2025. "Why Do Legislators Engage in Waffling? Evidence from the Korean National Assembly, 2004-2020." Journal of East Asian Studies. doi:10.1017/jea.2025.10013

Lee, Hyun-Chool, and EunKyung Kim. 2022. "Institutionalization of the National Assembly Formation." Korean Party Studies Review 21 (3): 117-148. doi:10.30992/kpsr.2022.09.21.3.117