Poster 2 · ObliGate research · public data · legacy era (pre-2016 declarations) · data vintage 2026-08-26

When Applicants Fight Back and Win, Does the Money Follow?
Clawback, Contest, and Observable Restoration in FEMA Public Assistance

Arin · ObliGate Compliance Inc. · August 2026 (v1.5, vintage-corrected) · Data: OpenFEMA Grant Award Activities (1,220,585 versions, pulled 2026-08-26) × Second Appeals Tracker (July 2026 snapshot, 1,685 appeals) · code public
Disclosure: the author is affiliated with ObliGate Compliance Inc., which builds documentation software for FEMA Public Assistance applicants. This is an observational study of decision implementation; it offers no advice on whether to appeal. Selection into appealing and into granting is unmodeled.
85.3% vs 13.6%
post-decision restored obligations: granted appeals (81/95) vs denied controls (44/324)
71.7 pts
the falsification gap — OR 36.8, p = 3.2e-39, exact 95% CIs disjoint
$69,071
median restored dollars per winning mover project (n = 80; $85,622 in the contested-dollar subset)
118 days
median winner lag to new obligations; denied-arm movers: 531 days and larger — the amendment signature
Bar chart: post-decision positive obligation versions appear on 85.3 percent of granted or partially granted projects versus 13.6 percent of denied controls, with disjoint 95 percent confidence intervals
Figure 1. The falsification control: if post-decision positives were ordinary amendments, denied projects would show them at similar rates. They sit at a ~14% noise floor; winning projects at 85.3%. The claim rests on the 71.7-point gap, not the raw rate.
Bar chart of match rates by disaster declaration era: 76.6 percent for pre-2016 declarations, 19.3 percent for 2016 to 2019, 1.7 percent for 2020 and later
Figure 2. The honest scope, updated: the completed dataset restores 2016–2019 records (the earlier snapshot's cliff there was an artifact); the 2020+ floor persists because the tracker cites Grants Portal GMP numbers that do not resolve against this dataset's project numbering. Only 1.3% of references point to disasters absent from the data.

Why this is open

Oversight measures appeal decisions, never money movement: GAO-18-143 measured timeliness; DHS OIG capping reports quantify recommended deobligations but never post-appeal restoration; FEMA's 2021 regulatory impact analysis counted appeal volumes with no dollars. Medicare RAC appeals — the closest analog — document overturn rates, not restoration microdata. To our knowledge, no study in any federal grant program links clawback records to appeal records and measures observable restoration and its timing.

Linkage: canonicalized (disaster, applicant, project) keys; bundled references split (2,441 refs from 1,673 appeals); era-diagnosed match rates (Fig. 2); appeal set held at the July snapshot for the vintage recompute (419/419 pairs recovered, zero status drift); all code public.

Findings (legacy era)

Appeals largely contested money already pulled back: 55.7% of matched appealed projects had a deobligation before the appeal was received.

Winning was visible in the money trail: restored obligations appeared on 85.3% of winning projects vs 13.6% of denied controls (Fig. 1); winners: median restored $69,071 per mover project (n = 80), new obligations a median 118 days (~3.9 months) after the decision; denied-arm movers were later (531 days) and larger ($164,480) — the amendment signature.

Restoration was typically partial: among winners with pre-appeal deobligated dollars (n = 43), the median restored-to-contested ratio was 0.63 (95% CI 0.30–0.89; IQR 0.07–0.97) — 23% ~fully restored, 49% partial, 21% near zero, 7% above 1.5× (new scope, not restoration). Denied controls (n = 192): median 0.00; 92% near zero. "Contested" is proxied by pre-appeal deobligated dollars (the tracker records no amount-in-dispute).

Limitations (load-bearing)

  1. Legacy-era scope (pre-2016 declarations); the 2020+ linkage floor is a key-numbering gap; modern-era generalization unknown.
  2. "Post-decision positive version" is consistent-with-implementation, not proven line-item restoration; the control rules out amendment noise, not selection; arm exchangeability is an assumption.
  3. No causal claim — who appeals and who wins is unmodeled by design.
  4. Project-worksheet unit; late decisions had less time to show restoration; OpenFEMA is raw, restated-over-time administrative data (every number vintage-stamped); we observe obligation events, not disbursements.
Methods note & code archived: doi.org/10.17605/OSF.IO/2NK5V (v1.2 staged). To our knowledge, the first linked-records measurement of the clawback–contest–restoration pipeline in a federal grant program. Reproducible from public data: join_feasibility.py · amount_matching.py + vintage reconciliation scripts. Sources: OpenFEMA Public Assistance Grant Award Activities (v2), pulled 26 August 2026; Second Appeals Tracker (July 2026 snapshot). Companion: Poster 1 (national deobligation baseline). Contact: ObliGate Compliance Inc.