Simulation as Alibi — Audit & Supporting Materials

Supporting materials for:

Tang, S., & Lin, Z. (2026). Simulation as Alibi — How the Social Order Gets Outsourced to LLM Agents. Ninth AAAI/ACM Conference on Artificial Intelligence, Ethics and Society (AIES-26), camera-ready version.

This repository provides the coding data, codebook, audit materials, and analysis scripts behind the paper's systematic coding analysis (Stream 1b), together with documentation of the other data streams.

---

Repository structure

simulation-as-alibi/
├── README.md                      # this file
├── data/
│   ├── stream1b_systematic_coding/
│   │   ├── coding_data_142.json   # 142-paper coding data (EI, DV, DP, RT, evidence quotes)
│   │   └── cross_tabulation_ei_dv.csv
│   ├── stream1a_cda/              # CDA corpus: qualitative stream (no data published)
│   ├── stream2_platforms/         # 5 platforms: case audit + Artificial Societies snapshot
│   └── stream3_governance/        # 6 governance frameworks: list + official sources
├── codebook/
│   └── codebook_stream1b_v6.md    # DV coding rules (emergence-explanation-sentence anchoring)
├── analysis/
│   └── statistics_summary.json    # all statistics reported in the paper
└── scripts/
    └── reproduce_statistics.py    # reproduce every number in Table 2 & Section 4.2

Quick start

python3 scripts/reproduce_statistics.py

This reproduces, from data/stream1b_systematic_coding/coding_data_142.json:

Statistic	Paper reports	Script reproduces
χ² (df=4)	16.62	16.622
p	0.002	0.002
Cramér's V	0.242	0.242
r(EI, DV)	-0.141 (p = 0.094)	-0.141 (p = 0.092)
r(EI, RT)	0.194 (p = 0.021)	0.194 (p = 0.019)
Absent concentration	14/17 (82%) in High EI	14/17 (82%)
arXiv share	~80%	80%

What the paper claims, and where it is backed here

1. 142-paper systematic coding corpus (Stream 1b) — coding_data_142.json. Each record contains: Paper ID, title, venue, year, and the four coded dimensions used in the paper — emergence intensity (EI), designer visibility (DV), deployment proximity (DP), regulatory translatability (RT) — plus the verbatim emergence-explanation sentence (English) and the coding rationale (recorded in the coders' working language, Chinese; see the Stream 1b README).
2. DV coding is anchored in the emergence-explanation sentence — codebook/codebook_stream1b_v6.md (Section 3.3 of the paper).
3. Table 2 cross-tabulation and all statistics in Section 4.2 — reproduce_statistics.py, cross_tabulation_ei_dv.csv, statistics_summary.json.
4. The threshold signature — significant χ² alongside a non-significant ordinal correlation — is confirmed by the script (THRESHOLD CONFIRMED).
5. ~80% arXiv preprint share — computed from the Venue field (Section 8 limitation discussion).
6. Stream 2 (5 platforms) and Stream 3 (6 governance frameworks) — documented in data/stream2_platforms/README.md and data/stream3_governance/README.md, including the full case audit (case_audit.md) and official source links.
7. Stream 1a (CDA, 20 papers) — data/stream1a_cda/README.md. Stream 1a is the qualitative component of the study; consistent with the paper, no per-paper CDA corpus or cross-tabulation is published, and no quantitative statistic in the paper depends on this stream.

Coding protocol summary (DV)

The DV code is assigned by locating the paper's emergence explanation sentence — the sentence in which the paper explains why an emergent phenomenon occurs — and asking whether the designer or design choices appear in that sentence as a causal subject:

• DV = Full: the emergence explanation attributes the outcome to specific design choices (e.g., "removing memory prevents the emergence of stable cooperation").
• DV = Partial: the architecture is disclosed in the methods but the emergence explanation does not return to those design choices.
• DV = Absent: emergence is narrated as spontaneous, self-organized, or population-level (e.g., "norms emerged naturally from interactions").

Risk, governance, and ethical discussions do not count as evidence of designer visibility.

Notes on scope

• The quantitative analysis (Stream 1b) covers 142 papers spanning the full publication spectrum (peer-reviewed venues and arXiv preprints alike), because the central debates over emergent behavior are currently unfolding in the preprint literature; ~80% of the corpus are preprints (see the paper's limitation discussion).
• Case studies (Section 6) are document-based pattern inference, as stated in the paper.

License & contact

Materials are released for transparency and audit. Please cite the AIES-26 paper. For questions: jamietang@ruc.edu.cn.