1# Hostile referee report — WP 25 v1.0 (self-review, 2026-10-27)23Written as a skeptical referee for a top general-interest or computational-economics journal would, before4submission. Items marked **[open]** are not fixed in v1.0 and are disclosed in the paper's limitations.56## Summary verdict7Interesting and unusually honest paper: the central "efficiency benchmark" is a negative result that undercuts the8obvious policy story, and the authors derive rather than assert the externality. The theory is tight but9second-order; the quantitative sections rest on illustrative parameters; the agent laboratory is small. Major revision.1011## Major comments12131. **Is Proposition 3 just Angeletos–Pavan (2007) / Colombo–Femminis–Pavan (2014) relabelled?** The use-of-information14 efficiency under `U_xX + U_XX = 0` is theirs. The acquisition of a *correlated* signal by a measure-zero agent15 differs from private-signal acquisition (the "knowing what others know" term). The paper should (and does, in the16 proof) show the identity explicitly rather than cite; the referee would still like a one-paragraph argument for why17 the correlation externality nets out (answer: the agent's private valuation of the signal already includes its18 covariance with the aggregate, and with `U_X ≡ 0` the aggregate's second-order effect on others is pure transfer).19 **Addressed in Section 4.4; could be sharpened.**202. **Second-order approximation.** All welfare statements are to second order around a symmetric steady state. With21 large adoption swings (D from 0 to 1) and strongly strategic substitutes (alpha = −1.33) the approximation may be22 poor. **[open]** No global check in the non-linear economy is provided for the portfolio mapping (the non-linear23 steady state is used, but the game is LQ).243. **Which decision is delegated is a modelling choice.** The paper admits the savings channel is dead and switches to25 portfolio choice. A referee will ask whether firm input choices (misallocation) would give a different ranking of26 channels. **[open]** Flagged as the first extension.274. **Information parameters are made up.** sig_theta_R = 0.06, sig_H_R = 0.10, sig_A_R = 0.04 drive the 3.3 % of28 consumption number and the 7.7 % wage-noise s.d. The paper sweeps them and calls them illustrative; a referee will29 still object to any headline number. **Mitigation:** headline numbers are presented as orders of magnitude; the30 qualitative results (PV decreasing in D, D²/M scaling, zero wedge) are parameter-free.315. **Model B decision-quality returns are reduced-form.** The mapping from the Model A information friction to the32 return term (m_H, sigma_rH, sigma_rA) is not derived. **[open]**; the paper says so.336. **Concentrated ownership is bolted on.** The owner class is outside the household block (a point mass). A referee34 may ask for an explicit two-type household economy. Defensible as Proposition 6's two-class economy, but the owner35 class does not consume or save optimally. **[open]**367. **Model C's market maker is ad hoc.** Depth, tick and inventory limits are parameters; the H4 amplification37 depends on them and on the call-vs-CDA design. The paper is explicit that this is the finding (execution matters),38 but a referee will want the amplification mapped to Proposition 7's `lambda_r h D (1−m)/m` quantitatively.39 **[open]**: only qualitative correspondence is given.408. **The stress test needed engineering.** Dividend persistence 0.95, preloaded leverage, auto-repay off, −10 s.d.41 shock, thin ladder: five choices to produce a spiral. The paper discloses them in the table notes, but the referee42 will read this as "amplification can be produced when you set things up for it". The honest framing — the43 transient model error alone does not cascade; it needs persistence, leverage already in place and thin depth —44 should be the headline of that subsection. **Partly addressed.**459. **LLM laboratory is small** (8 agents × 24 periods × 3 seeds in the quick design) and uses 4-bit 7B/3B models.46 Results about model identity vs prompt are suggestive. The claim "LLM agents trade more under a shared signal" must47 be verified against the final cells.csv before submission. **[check before submission]**4810. **Effective M.** Spiro 2026-type cross-vendor correlation (~0.77) implies the effective M is near 1 even with49 many vendors; the paper treats M as illustrative but could calibrate an effective M from reported error50 correlations. **[open]**5111. **Calibration deviations** (beta 0.96/0.95 vs ledger 0.99; alpha 0.33 vs 0.354; rho_e/sigma_e vs Flodén–Lindé)52 are documented but not justified beyond convention. Minor.5312. **Code-paper consistency.** All numbers are macros generated from results; figures are generated by scripts; the54 git revision is printed. The referee could not find invented numbers. Good.5556## Minor comments57- Notation: alpha (strategic) vs alpha_K (capital share) fixed in v1.0; nu is used for both Frisch and the earnings58 innovation in docs/model_specification.md (paper uses eta for the innovation... check).59- The literature section is long (2,100 words); fine for a working paper.60- The architecture figure is dense.61- The bibliography (144 entries) is verified; the Holmström–Milgrom DOI contains an underscore (aer.bst does not print62 DOIs, so no compile issue).6364## Would I recommend publication after revision?65Yes, conditional on (2) a global check of the LQ welfare statements, (7) a quantitative mapping of the lab66amplification to the multiplier, and (9) verified LLM numbers. The efficiency benchmark and the D²/M loading are the67publishable core; the compute-market crowding-out is a clean negative result; the distributional results are68expected but cleanly decomposed.69