New thesis: The procurement cost of judicial enforcement decomposes into three distinct channels under the under-the-gun umbrella: (i) demand fragmentation (C1), the dominant channel — smaller orders that forgo bulk discounts; (ii) reduced bidder participation (C2) — at identical order size, urgent tenders attract 4.2% fewer bidders; (iii) supplier composition shift (C3) — the supplier mix changes under urgency, with no within-firm markup.
Headline figure replaced: the v6 coefficient plot is replaced by a two-panel cascade waterfall (fig_three_channel_cascade.pdf) decomposing the urgency premium and the UTG gap across four canonical specifications, attributing each pp drop to a named channel.
Within-firm markup hypothesis directly refuted: the within firm-buyer-item triple regression (8,232 triples observed in both regimes) yields a negative urgency coefficient (−2.67%, p < 0.01). Same firm, same buyer, same item, charges less under urgency, not more. The 52% supplier-FE attenuation in v6 was therefore selection (composition shift), not markup.
Welfare bound replaces v6 point estimate: $16M lower bound (5.4% × $300M) → $74-88M upper bound (UTG-pct × \(250-300M sanction-exposed). Publishable range: **\)16-88M/yr** on the per-unit-price margin.
Quantified policy counterfactuals: admin-channel expansion at 50% target = ~\(44M/yr recovery; framework-agreements at 40% feasibility = ~\)36M/yr recovery (each addresses a distinct channel).
Identification defended candidly: new paragraph in §Empirical Strategy on scientific-committee selection bias direction (theoretically ambiguous, not "conservative" as v6 claimed).
Dose-response demoted: the v6 saturating-dose-response paragraph drops from main text to footnote (mid and high bins indistinguishable, z = 0.60).
Title finalized: "Bitter Pills: Judicial Enforcement and the Cost of Public Procurement in Brazil" (replaces working title "Bitter Pills to Swallow: The Enforcement Costs of Health Litigation").
Macroized manuscript: every numerical claim, sample size, table input and figure path resolves through manuscript/paper/values.tex. Each analysis script (v4: 00, 01, 03, 04, 06, 08, 11; v6: 20, 30, 31, 32, 33, 34) emits its block at end of run; manuscript prose is regenerable from code with zero hand-editing. 173 unique macros referenced, 11 auto-generated blocks, 0 TBD leaks across 4 PDFs (main, submission, OnlineAppendix, cover_letter).
Drift bugs closed by macroization:
qty-on-admin coef cited in prose was 0.787 (item+year, no PBU FE) but the paper reports preferred = item+year+PBU everywhere else. Preferred coef is 1.203 (administrative orders are roughly 3.3× the size of litigated orders).
Dose-response prose claimed a monotonic-then-plateau pattern that the data did not support; refreshed run gives 6.8% / 12.4% / 10.5% with mid-vs-hi \(z = 0.60\) (overlapping ICs). Reframed as a saturating dose-response, with the indistinguishability test moved to a footnote.
UTG range "23–30%" was a loose claim. Now computed as min..max sweep across 4 specs (col 1 item only → col 4 item+ym+PBU = 23.2%–30.4%). Total effect paragraph rewritten to expose the sweep.
New §A.7 Regex Classifier Validation in Online Appendix; \BPregexValidationStatus placeholder is promoted to a quantified F1 sentence once validation_sample.csv is hand-labeled.
Cross-ref bugs fixed:
tab_placebo.tex and tab_supplier_fe.tex were emitted by etable without \caption/\label, so \ref{tab:placebo} and \ref{tab:supplier_fe} rendered as Table ??. Script 20 now wraps in \begin{table} + \caption + \label.
OnlineAppendix referenced sec:institutional from the main paper without xr-hyper, rendering Section ??. Added \externaldocument{main}.
Target: Journal of Public Economics short paper format.
Body: Compressed from 5,263 to 3,800 words.
Literature review: Collapsed from 5 paragraphs to 1.
Heterogeneity: Moved entirely to Online Appendix.
New empirical exercises:
Placebo test (items never litigated): coefficient \(-0.020\) (SE = 0.032, n.s.) for negotiated prices, \(-0.031\) (SE = 0.045, n.s.) for reference prices. Falsification passes.
Supplier fixed effects (2,202 firms): 52% attenuation of urgency coefficient (0.051 to 0.025). With firm FE + quantity controls, residual falls to 0.011 (n.s.). Decomposes premium into roughly equal demand-side and supply-side components.
UTG Panel B: Explicitly framed as quantity channel --- sanctions distort the entire procurement environment.
Conclusion: Sharpened to two-pronged reform (expand admin mechanism + remove court-order signal from tender notices).
Maps: Enhanced maps with SIRGAS 2000 projection and quintile breaks for litigated, administrative, and ordinary purchases. New 3-panel comparison map.
Regressions: Administrative vs ordinary — 40 models (8 outcomes x 4 FE specs + LPM variants). Key finding: admin purchases carry lower reference prices than ordinary (-9.1%).
Manuscript: Restructured to 24-page body (under 32pp target) with 5 tables/figures inline; all other results in appendix. Added admin vs ordinary section.
MkDocs: New pages for admin vs ordinary results, maps comparison, and changelog.