The resolution source is the openFDA drug enforcement endpoint. The openFDA drug enforcement documentation shows that the endpoint is https://api.fda.gov/drug/enforcement.json, that report_date is a searchable field, and that example queries can combine date ranges with recall classification filters; the API response format includes meta.results.total, which is the field this question resolves to. FDA’s Enforcement Reports page also says recalls are posted once classified, the report is weekly, and firms often make public recall announcements before FDA completes classification. The endpoint response I could inspect showed last_updated of 2026-06-24, so as of Sunday, July 5, 2026 this is a forecast only a little more than a month ahead and mostly about what part of the already-visible recall pipeline lands inside the July 6-August 11 reporting window. (open.fda.gov)

That timing point is the key forecasting lever. Because FDA says classification commonly follows the firm’s announcement, the best predictor is the set of recent drug recall announcements that look severe enough to plausibly become Class I during the target window. FDA’s drug recalls page shows a cluster of recent drug recall announcements in late April through mid-June 2026, including B. Braun on April 28, Sun Pharma/Aisling on May 13, Haleon Gas-X on June 4, BD on June 8, and Beekeeper’s Naturals on June 12. (fda.gov)

Several of those look like strong Class I candidates. B. Braun’s Lactated Ringer’s notice says particulate matter infused intravenously could cause pulmonary emboli, organ damage, permanent impairment, or death, and it lists two recalled lots for NDC 0264-7750-07. BD’s ChloraPrep/FREPP notice says contamination may lead to serious systemic infection, sepsis, illness, and death; the public notice lists two NDCs/lots and adds that other lots are being recalled in a customer letter. Beekeeper’s Naturals says its saline nasal spray may contain Aspergillus spp. and that there is a reasonable probability of serious and life-threatening infections in vulnerable users. Sun Pharma recalled one batch of doxorubicin liposome injection because of potential glass particles. My read is that these four events alone create a meaningful near-term Class I pipeline for July-August reporting. (fda.gov)

The main downward adjustment is that not every recent drug recall should be expected to resolve as Class I. Haleon’s Gas-X notice describes potential chemical contamination that may result in nausea, vomiting, abdominal pain, and diarrhea, which reads more like a likely Class II than a Class I case. Also, the question counts enforcement records rather than distinct recall events, so multiple package configurations can inflate totals, but some announcements are still only one lot or one batch and may yield only a single enforcement record. (fda.gov)

Putting that together, I center the distribution around about 11 total records, with a median around 10. My rough decomposition is: (1) already-announced serious recalls likely classified into the window contribute something like 7-10 records in expectation; (2) a smaller contribution comes from additional classifications/newly announced high-severity recalls that arrive during the window itself; and (3) there is a real right tail because one contamination/sterility event with multiple package identifiers can create many enforcement records instead of just one. The biggest uncertainty is timing, not whether serious recalls exist: FDA explicitly notes that public announcements often precede classification, so some visible candidates can slip just before July 6 or just after August 11. That is why I used a right-skewed mixture rather than a tight single-rate count model. (fda.gov)

The resolution source is the openFDA drug enforcement endpoint. openFDA says this dataset comes from FDA’s Recall Enterprise System, is updated weekly, and the API response includes meta.results.total; the enforcement-report field definition states that report_date is the weekly Enforcement Report date for the recall. (open.fda.gov)

That weekly cadence matters a lot here. FDA says recalls are included in the Enforcement Report once they are classified, and FDA’s drug recalls guidance notes that public notices often happen before FDA completes classification and posts the recall to the report. FDA’s Recalls and Shortages Branch also says drug recall postings to the Enforcement Report are published every Wednesday. So, if that cadence persists, the window from July 6 through August 11, 2026 likely captures five weekly report dates: July 8, July 15, July 22, July 29, and August 5, 2026. The five-date count is my calendar inference from FDA’s weekly Wednesday publication schedule plus the report_date definition. (fda.gov)

For a top-down base rate, FDA’s Office of Compliance annual report says that in 2025 FDA classified 321 drug recall events totaling 755 recalled drug products. That is about 14.52 recalled products per week across all classes. (fda.gov) If I spread that evenly across five weekly reports, that is roughly 72.6 all-class recalled-product records in a window shaped like this one. (fda.gov)

The main uncertainty is the Class I share. My sources did not give a clean text value for the Class I proportion of drug recall records, so I inferred it. Recent FDA drug-recall announcements were frequent enough to make a single-digit Class I total over five weekly reports plausible: the drug recalls page listed ten announcement rows between March 24 and June 12, 2026, including multiple contamination, particulate, glass-particle, and mispackaging cases. But FDA also explicitly says that not all recalls are announced on FDA.gov or in the news media, so public announcements are an incomplete lower-bound signal, not the full enforcement count. (fda.gov)

I therefore used a midpoint assumption that about 10% of recalled drug product records in this type of five-report window will be Class I, with a plausible band of roughly 8% to 12%. Applied to the 2025 all-class base rate, that gives about 5.81 to 8.71 Class I records over five weekly reports, with a midpoint around 7.26. (fda.gov) I nudged the final mean slightly to 7.2 rather than exactly 7.26 because some recent public announcement rows are not perfect one-for-one predictors of eventual openFDA Class I drug enforcement records, and because classification timing can slide in either direction around the target window. That is an inference, not a directly observed statistic. (fda.gov)

I modeled the final count with an overdispersed negative-binomial distribution rather than a simple Poisson. That is because counts arrive in weekly batches, and one underlying recall can generate multiple enforcement records. I set the mean to 7.2 and used moderate overdispersion (size parameter 4.0), which keeps the forecast centered in the high-single-digits but leaves room for both a quiet five-week stretch and a clustered burst of multi-record Class I recalls.

Bottom line: my median expectation is in the 6-8 range, with the distribution centered at about 7.2 records. The most likely outcome is therefore a high-single-digit total, with meaningful probability on roughly 3-12 and a thinner but non-zero tail above that.

As of July 5, 2026, openFDA says the drug enforcement endpoint is updated weekly, currently covers data through 2026-06-24, and its downloadable drug-enforcement dataset was last updated on 2026-06-27. The endpoint supports filtering on report_date and classification, so the target is a weekly refreshed enforcement-record total, not a distinct-event count. Because the public data now stop before July 6, 2026, none of the target-window observations are visible yet; and because the feed refreshes weekly, there is some chance that the Aug. 11, 2026 12:00 UTC resolution query will still miss the final few nominal dates in the requested range. That lag point is an inference from the observed coverage and update dates. (open.fda.gov)

For scale, CDER’s Office of Compliance reported 264 classified drug recall events in calendar 2023, including 28 potentially life-threatening Class I events, totaling 1,178 violative drug products. The same office reported 321 classified drug recall events totaling 755 recalled products in calendar 2025. OPQ’s State of Pharmaceutical Quality reported 260 recall events and 421 recalled products in FY2024, and 225 recall events and 674 recalled products in FY2023. Taken together, those official reports imply that a 37-day span typically contains a few dozen all-class drug recall events and something like a few dozen to roughly 75 product or record-level recalls before filtering to Class I. (fda.gov)

The hard part is the Class I filter. The clearest official class split I found is 2023: 28 Class I out of 264 events, about 10.6%. As a recent cross-check, sampled 2025 weekly summaries from IPQ, a trade publication summarizing FDA enforcement reports, show Class I drug recalls were sporadic: many sampled weeks had zero, one May week had one, one June week had one, and one July week had three. That sample looks more like a high-single-digit Class I share than a stable double-digit share, though with obvious week-to-week clustering. My event-rate prior is therefore about 2 to 3.5 Class I recall events over a 37-day window. (fda.gov)

To map events into the resolution value, I used a compound Poisson model: first sample the number of Class I events, then sample how many openFDA enforcement records each event contributes. The annual FDA reports show that product or record counts can materially exceed event counts, so I used a mean just under 3 records per Class I event, with a long but thin tail for larger multi-NDC recalls. I also made a modest downward adjustment, via the event-rate choice, for possible weekly-refresh lag at resolution. That leaves the distribution centered around 6 to 7 records; my single-number point forecast is 6. The code returns the full discrete distribution over 0, 1, 2, …, 120, and greater than 120. (fda.gov)

I treat this as a count of openFDA drug enforcement records, not distinct recall events. openFDA says the drug enforcement endpoint is updated weekly, FDA defines report_date/enforcementreportdt as the weekly enforcement report date, and FDA’s enforcement-report documentation notes that a recall event may include more than one recalled product. CDER also says drug recalls pending classification are published in the Enforcement Report every Wednesday. Combined with openFDA’s current coverage ending at 2026-06-24, that makes the July 6, 2026 through August 11, 2026 window effectively a five-report window, most likely the Wednesday reports dated July 8, July 15, July 22, July 29, and August 5, 2026. (open.fda.gov)

For base rates, FDA’s 2025 Office of Compliance annual report says CDER classified 321 drug recall events totaling 755 recalled products in 2025. A separate FDA pharmaceutical-quality report analyzing FY2017-FY2021 recall data reports 1,220 recall events, of which 113 were Class I. By visually reading Figure 3 in that report, I estimate roughly 50, 50, 25, 80, and 145 Class I recalled products for FY2017-FY2021, or about 67 per year on average; that implies about 6.44 Class I records in a five-week window. Applying a historical roughly 12% Class I share to 2025’s 755 recalled products gives about 8.71 in a five-week window. Those two historical anchors bracket the likely range. (fda.gov)

As a near-term check, the FDA Drug Recalls page showed a cluster of recent 2026 drug recall announcements, including Amneal on March 24, B. Braun on April 28, Pharmacal on May 12, Sun on May 13, Haleon on June 4, BD on June 8, and Beekeeper’s Naturals on June 12. Those examples often span more than one product or lot, such as Gas-X 72 ct and 120 ct, BD’s 1 mL and 1.5 mL applicators, and B. Braun’s two lots, but FDA also states that not all recalls are announced on the media-facing recalls page. Using those visible announcements as a lower-bound proxy and scaling them to a five-report window yields about 5.95 records before any upward adjustment for unpublicized recalls that still appear in Enforcement Reports. (fda.gov)

Blending the three heuristics equally gives 7.03, and I then nudge that up by 10% to 7.74 because the public-announcement proxy systematically misses recalls that do not get separate FDA.gov posts. I model the total with a negative-binomial distribution with mean 7.738076923076924 and variance about 21.04 to capture week-to-week lumpiness and the fact that one underlying recall can create several enforcement records. My point estimate is 8 records, with most probability mass in the mid-single-digits to low teens and a meaningful but smaller tail into the high teens. (fda.gov)

I treated this as a forecast of the exact integer returned by the openFDA drug enforcement endpoint for Class I recalls with report_date from 2026-07-06 through 2026-08-11 inclusive. For this endpoint, openFDA says the data come from FDA’s Recall Enterprise System, the data are updated weekly, and report_date is the date the FDA issued the enforcement report. The same openFDA materials also show that, in early July 2026, the drug enforcement dataset was updated on 2026-07-02 but only covered records through 2026-06-24. That suggests a current lag of roughly about a week between “today” and the latest report_date actually visible in the API. I therefore treated it as quite plausible that the resolution query at 12:00 UTC on 2026-08-11 will still be missing some of the very last days in the nominal window; this is an inference from the present lag pattern, not a guaranteed rule. (open.fda.gov)

For base rates, I leaned on official FDA recall statistics. FDA’s FY2015 enforcement statistics show CDER had 45 Class I recall events and 118 Class I recalled products, versus 303 total recall events and 1,822 total recalled products for CDER overall that year. That implies a Class I share of about 14.9% of events and 6.5% of recalled products in that snapshot. Separately, FDA’s FY2021 pharmaceutical-quality report says that FY2017-FY2021 included 1,220 recall events, of which 113 were Class I, or about 9.3% of events. Those numbers tell me Class I is uncommon but not rare, and that the record count can vary meaningfully depending on how many product-level lines each event fans out into. (fda.gov)

I also wanted a recent all-class pace for the openFDA/openFDA-like record stream. A 2024 retrospective study of FDA drug recalls reported 15,710 observations spanning June 2012 through August 2023. openFDA’s statistics page currently lists 17,723 drug enforcement records in the dataset. The difference is about 2,013 records added over roughly 34 months, or about 59 records per month, roughly 700 per year, across all drug enforcement classes. If Class I is something like high-single-digit percent of records, that points to a raw Class I pace on the order of roughly 5 to 7 records over a full 37-day window before accounting for API publication lag. (sciencedirect.com)

That estimate is broadly consistent with the FDA quality reports showing substantial all-class recall volume in recent years. In the FY2024 pharmaceutical-quality report, summing the defect-group counts shown in the figure gives about 691 recalled products in FY2020, 787 in FY2021, and 902 in FY2022, so the denominator for all drug recalls has remained fairly large. Meanwhile, the FY2021 report’s classification chart shows that Class I recalled-product counts can swing a lot from year to year, including visibly larger spikes in some years. That is why I do not want an overly tight Poisson-on-total-records model; the process is lumpy because one serious recall event can create several product/NDC-level enforcement records at once. (fda.gov)

I used two simple models and blended them mentally. Model 1 was a share-of-total model: recent all-class pace (~59/month) times a Class I share bounded by the official FDA snapshots above, then scaled down because the API currently appears to lag current calendar dates by around a week. That model landed around the mid-single digits. Model 2 was an event-clustering model: start from roughly 22-26 Class I events per year, convert that to an effective resolution-window event count after allowing for lag, then multiply by roughly 3-ish records per event as a compromise between FY2015’s 118/45≈2.6 products per Class I event and the fact that some more recent Class I episodes clearly fan out into multiple product lines. That model landed around 6-7 records. (fda.gov)

My final distribution centers just above 6. In the code I express that as a compound-Poisson model with lambda_events = 1.85 and a geometric records-per-event mean of 3.25, which gives an expected value of 6.0125. I intentionally leave meaningful probability on 0 through the low teens, because (a) Class I drug recalls are relatively infrequent, (b) the endpoint is only updated weekly, and (c) one multi-SKU or contamination-driven event can create a noticeable burst of records. Relative to a naive full-37-day count, the biggest downward adjustment is the current evidence that the API may not fully include the last several report_dates by the exact resolution time. (open.fda.gov)

How many Class I drug recall records will openFDA return with a report date between July 6 and August 11, 2026?

• My most likely prediction for the number of Class I drug recall records is 4.50, with a 50% chance of falling between 1.50 and 15.50.
• The prediction reflects a baseline of low-frequency but high-impact reporting, where the majority of weekly FDA batches contain few Class I records.
• A significant right-side tail accounts for ‘batch’ reporting, where a single multi-product recall event can generate dozens of records simultaneously.
• Recent 2026 trends show the pharmaceutical sector at a 12-year high for recall unit volume, although many recent high-profile cases have been Class II rather than Class I.
• Intensified FDA focus on compounded GLP-1 medications and foreign manufacturing cGMP violations increases the risk of a high-volume Class I record release within the window.
• The forecast window (July 6 to August 11, 2026) covers approximately five to six weekly FDA Enforcement Report cycles, which are the primary periods when new records are added.

Key figures

Historical context

• Historically, Class I recalls are reserved for the most serious violations, such as sterile contamination in injectables or critical dosing errors.
• In Q1 2026, the pharmaceutical sector reached a 12-year high in unit volume for recalls, even as the number of distinct events decreased by 10.5%.
• A single recall ‘event’ in the openFDA database often expands into multiple ‘records’ based on NDC numbers; for example, a 2024 duloxetine recall involved multiple lots across two separate records.
• Recent 2026 enforcement has focused heavily on the compounding and telehealth sectors, with 1,700+ adverse event reports linked to compounded GLP-1s by May 2026.
• The FDA shifted leadership in early 2026, moving toward more aggressive post-market surveillance.

Tailwinds

• The FDA’s new policy of ‘continuous post-approval monitoring’ and intensified post-market surveillance under the 2026 commissioner.
• High levels of adverse event reports (1,700+) for compounded medications suggest a backlog of potential Class I actions.
• Foreign manufacturer inspections have increased, often resulting in multi-product ‘batch’ record releases following facility-wide import alerts.
• The programmatic nature of the API (meta.results.total) counts every NDC/packaging variation, which amplifies the impact of any single broad recall event.

Headwinds

• Many large drug recalls reported in June 2026 (e.g., Amgen, duloxetine) were classified as Class II or Class III, not the Class I category required by the query.
• The 2025-2026 multi-agency deregulatory agenda (requiring the elimination of ten regulations for every one new one) may slow down official classification and reporting processes.
• The ‘report_date’ in openFDA reflects the date of official classification, which often lags significantly behind the manufacturer’s voluntary recall initiation.
• Class I recalls are the rarest of the three severity levels; many five-week windows historically contain very few or no Class I drug records.

Detailed reasoning

My analysis is based on the intersection of historical openFDA reporting patterns, recent 2026 enforcement trends, and the specific technical parameters of the API query. The target variable is the ‘meta.results.total’ field, which counts individual enforcement records rather than distinct recall events. Because each unique National Drug Code (NDC), package size, or lot can be logged as a separate record, the data is naturally ‘batchy’ and prone to sudden spikes. Historical base rates for Class I drug recalls—the most severe category involving potential death or serious harm—show they typically comprise about 10% of total drug enforcement actions. Over a five-week window, the neutral expectation is relatively low, often falling between 5 and 15 records.

However, the 2026 context is not neutral. My forecast incorporates the fact that pharmaceutical recall volumes reached a 12-year high in Q1 2026 (218.8 million units). Furthermore, the FDA has intensified its ‘continuous post-approval monitoring’ and launched significant enforcement waves, particularly against compounded GLP-1 medications (semaglutide/tirzepatide) and foreign manufacturers with Current Good Manufacturing Practice (cGMP) violations. As of mid-June 2026, the FDA had issued over 135 warning letters in these sectors. While warning letters are not recall records, they are leading indicators of potential Class I actions if products are found to be sterile-contaminated or otherwise life-threatening.

In constructing the distribution, I weighted several scenarios. The most likely scenario (65% weight) reflects a routine window where the FDA processes a handful of small Class I actions, leading to a median near the lower end of the historical range. A secondary ‘batch-spike’ scenario (20% weight) accounts for the possibility of a single large-scale event—such as a nationwide sterility failure in a compounding pharmacy—which could easily generate 30 to 70 records in a single week. I also included a 15% probability for a ‘quiet window’ (0-3 records), acknowledging that the FDA’s reporting cadence can be irregular, and the 2025-2026 deregulatory agenda may influence the timing of classifications. The final prediction reflects this heavy right skew, with a low most likely value but a broad tail extending toward the upper bound. I have largely discounted the provided equity data for firms like Amgen and Eli Lilly, as their financial performance and stock volatility do not reliably correlate with the specific count of records in a narrow five-week FDA reporting window.

Key uncertainties

• Classification timing: The delay between a manufacturer’s recall and the FDA’s final Class I designation is highly variable.
• Record Multiplicity: The number of NDC records generated by a single recall ‘event’ cannot be predicted without knowing the specific product’s distribution profile.
• Reporting Lags: The openFDA API may experience several days of lag between an enforcement action and the record appearing with a specific ‘report_date’.
• Regulatory Policy: Changes in federal oversight intensity (deregulation vs. increased surveillance) can shift the threshold for Class I classification.

Conclusion

• My analysis suggests a most likely count of 4.50 records, with a 50% chance of falling between 1.50 and 15.50.
• While the broader regulatory environment is highly active, specific Class I drug record surges are unpredictable and often separated by periods of very low activity.
• The distinction between recall “events” and “records” is the primary driver of volatility; a single event involving many product identifiers (NDCs) can shift the outcome by dozens of units in a single reporting day.

Question: How many openFDA Class I drug enforcement-records (meta.results.total) will have report_date in the roughly five-week window July 6–August 11, 2026? Historical counts for the same window (2015–2025) center around a median of 4 and mean ≈4.45 (IQR 2, range 2–7), so all teams use a base-rate anchor near 4–6 records but note that the reconstructed series was manually transcribed and should be verified against openFDA/BigQuery. All forecasters agree on a heavier right tail than the historical series alone because the API counts recall-number records (product-level entries) not distinct events, and one multi-product Class I recall can multiply records in a single weekly report; examples from 2017 and 2023 demonstrate this mechanism. Recent spring/summer 2026 candidates — notably BD ChloraPrep, with a reported June Class I classification — modestly raise the chance of a cluster appearing in early July, but many other spring recalls may already fall outside the target window depending on actual report_date. Modelers placed medians around 5–6 (one model gave mean ≈6.8), with nontrivial mass in the 8–15 range, roughly ~15% chance above 10 and ~4% above 20, and essentially negligible probability above ~120. Recommended workflow and key caveats (verify historical counts, run the exact query after each weekly openFDA update, group results by event_id to detect clustering, and check BD ChloraPrep/report_date directly) were emphasized as critical to final accuracy.

Forecast and Reasoning

Forecast: Probability distribution:

• 1.00% chance of value below 0.0
• 5.00% chance of value below 1.0
• 10.00% chance of value below 2.0
• 20.00% chance of value below 3.0
• 30.00% chance of value below 4.0
• 40.00% chance of value below 4.8
• 50.00% chance of value below 5.5
• 60.00% chance of value below 6.5
• 70.00% chance of value below 8.0
• 80.00% chance of value below 10.5
• 90.00% chance of value below 16.5
• 95.00% chance of value below 26.0
• 99.00% chance of value below 60.0

Forecast Reasoning

We are forecasting the integer value of meta.results.total from the openFDA drug enforcement endpoint for records with classification:"Class I" and report_date in [20260706, 20260811]. This window is 37 days and (per multiple analysts) likely spans about five weekly FDA enforcement report dates (ending around 2026-08-05, since 2026-08-11 is a Tuesday).

Across Forecasters 1–3, the central tendency clusters around ~5–8 records, supported by direct endpoint checks such as: (a) the same window in 2025 yielding 5 records, and (b) 2026 Jan–Jun being unusually low (15 records in ~6 months), implying something like ~3 records if the low 2026 pace persists. Forecaster 2’s broader year-by-year sampling (e.g., 2025 lower than 2024 and 2026 YTD even lower) argues the recent baseline is more like ~3–6 in a 37-day window rather than ~8–10.

Forecaster 4 is an outlier with a very low median and a huge tail; however, it contains structural errors (negative counts are impossible) and likely over-weights “classification lag.” Because the filter is on report_date (publication date) and the query is run several days after the last included weekly report, any “Not Yet Classified” effect should exist but probably not dominate the window total.

Synthesis: I center the distribution at a median around ~5 records, with meaningful overdispersion (multi-NDC recalls can create clusters), but I keep the probability of extreme totals (>120) well below 1%, so no percentiles are placed into the open-above bin. Tails extend to ~60 at p99 to capture rare but plausible multi-record clusters without implying a large chance of exceeding 120.

An amateur forecast is likely to (i) confuse recall events with enforcement records, (ii) ignore the strong recency signal (2026 YTD appears lower than prior years in the analysts’ direct API sampling), and/or (iii) overreact to the open upper bound and place too much mass near or beyond 120. My forecast instead triangulates on the most decision-relevant evidence (same-window prior year and recent endpoint-derived rates), corrects the impossible negative-count issue, and includes a realistic but not exaggerated right tail for occasional multi-record recalls. Confidence in improvement is moderate: the main residual uncertainty is genuine process variance plus potential reporting/update timing quirks in openFDA.

Forecast rationale (numeric):

— Iteration 1 — The forecasts share a common logic: the count should be driven mainly by the short reporting window, the rarity of Class I drug recalls, and the fact that one recall can generate multiple openFDA records.

Main reasoning patterns

• Window length matters: The period from July 6 to August 11, 2026 spans about five weekly FDA Enforcement Report cycles, which sets the basic opportunity for new records to appear.
• Class I drug recalls are uncommon: The baseline expectation is that there are usually fewer than one Class I drug recall event per week, so the total should not be extremely large under normal conditions.
• One event can yield multiple records: A single recall may involve multiple NDCs, product presentations, or affected lots, so the number of openFDA records can be much higher than the number of underlying recall events.
• The distribution is right-skewed: Most of the probability mass is concentrated in low-to-moderate counts, but there is a meaningful chance of a large spike if a broad recall occurs.
• Tail risk is important: Rare but plausible events such as sterile injectable contamination, compounding pharmacy issues, or widespread multi-product recalls can push the count well above typical levels.

Areas of consensus

• The most likely outcome is a modest number of records, not a huge one.
• A quiet period with very few records is plausible.
• A large jump in the count is possible, but should be treated as a tail event rather than the base case.
• Most of the expected mass lies below roughly 120 records.

Main differences in emphasis

• Some reasoning centers the outcome in the low-to-mid teens, while others put it closer to 20–35.
• There is slight variation in how aggressively the tail is weighted:
  • some allow for rare outcomes well above 120,
  • others treat that as a smaller but still meaningful possibility.

Overall synthesis

The shared view is that the count should usually land in the tens rather than the hundreds, with a mid-teen to low-30s central tendency and a long right tail created by occasional large, multi-record recalls.

— Iteration 2 — Overall, the forecasts rely on the same core logic:

• Historical base rate: Class I drug recalls are not extremely frequent, so a 37-day window is expected to produce only a modest number of records in a normal period.
• API/reporting lag: openFDA may not fully reflect the most recent days of activity by August 11, so the count is likely to be slightly understated near the end of the window.
• Records vs. events: The biggest source of uncertainty is that openFDA returns records, not just distinct recall events. A single major recall can generate many records, especially if it affects multiple NDCs or products.
• Heavy right tail: Because of that structure, the distribution is strongly right-skewed: quiet periods could yield very low counts, but a rare mass recall could push the total sharply higher.

Areas of broad agreement

• The most likely outcome is low-to-moderate double digits, not hundreds under ordinary conditions.
• There is meaningful probability of few or even zero records if the period is quiet or ingestion lag is substantial.
• The main reason for a high count is one large multi-product recall, not a steady accumulation of many small ones.

Main differences in emphasis

• Some reasoning places the center of mass around roughly 10–12 records, treating the window as a typical short period with a few Class I recalls.
• Another view is somewhat higher, around 20–25 records, using a broader historical volume assumption.
• Tail assumptions vary widely, but all acknowledge that exceptional recall clusters can produce much larger totals than the median suggests.

Bottom line

The shared forecast logic is that openFDA will likely return a small-to-moderate count of Class I drug recall records, with the exact total heavily dependent on whether one large, multi-record recall occurs during the period and on how much of late-July/early-August reporting has already been ingested.

— Iteration 3 — Across the forecasts, the core reasoning is that this is a short, roughly five-week window in which Class I drug recalls are relatively rare, so the most likely outcome is a low-to-moderate record count, not a huge total.

Main shared reasoning patterns

• Rare-event process: Class I recalls are treated as infrequent, often modeled like a Poisson or compound Poisson process.
• Records vs. events: A key point is that openFDA returns records, not just recall events. A single large recall can generate many records if it affects multiple NDCs/products.
• Right-skewed uncertainty: The distribution is seen as heavily right-skewed: many windows should be modest, but a small number of windows could be very large.
• Possibility of zero or very low counts: Several rationales allow for quiet periods, including the possibility of 0 records.
• Indexing/reporting lag: One model notes that recalls reported near the end of the window may not be fully indexed by the query date, which could slightly depress the observed count.

Areas of consensus

• The central estimate is in the single digits to teens rather than very high.
• The forecasts agree that the upper tail must be wide, because a single multi-NDC Class I recall can produce a burst of records.
• They also agree that uncertainty is substantial, so the forecast should not be tightly concentrated.

Main differences

• The models differ on the most likely central value:
  • one leans lower, around 6
  • others place the median around 11–17
• They also differ on how aggressively to weight extreme spike events:
  • some keep the upper tail wide but more restrained
  • others explicitly allow for 100+ or even hundreds of records in rare black-swan scenarios

Overall synthesis

The collective view is that the expected result is a small-to-moderate number of Class I drug recall records, with a meaningful chance of very low counts and a small but important risk of large spikes if a major multi-product recall occurs. The dominant uncertainty is not whether recalls happen at all, but how many openFDA records each recall generates.

Summary of Forecasting Rationales

The consensus across the rationales relies on balancing historical FDA enforcement base rates with the structural mechanics of the openFDA API and the highly aggressive regulatory environment of mid-2026. Forecasters highlight the following key dynamics:

• API Mechanics and Data Structure: The openFDA API returns individual product records (National Drug Codes, or NDCs) rather than underlying recall events. Because a single manufacturing failure can affect multiple dosages or package sizes, the data distribution is heavily right-skewed. While the baseline expectation for a typical five-week period is low, a single systemic issue can generate dozens of records. Furthermore, structural reporting delays—such as API synchronization lags and the FDA temporarily listing severe recalls as “Pending Classification”—may depress the final captured count when the query executes on August 11.
• Elevated Baseline and Fiscal Seasonality: Forecasters broadly agree that the 2026 baseline is running significantly higher than historical averages due to stricter post-market surveillance and automated compliance tools. Additionally, the July/August window aligns with the FDA’s Q4 fiscal push to clear administrative classification backlogs before the federal year-end in September.
• Pending Pipeline and “Mega-Recall” Tail Risks: The forecast is heavily influenced by a known pipeline of severe actions initiated in May and June 2026 (including microbial contaminations, major injectables, and OTC products) that are currently maturing through the FDA’s classification lag and are primed to hit the API during the target window. The most significant upside tail risk stems from the FDA’s recent crackdown on compounding pharmacies, particularly those producing GLP-1 drugs. Because compounders produce highly varied custom formulations, a finalized Class I sterility failure from just one of these facilities could flood the database with an massive cluster of records.

Ultimately, forecasters anticipate a moderately elevated baseline driven by seasonal and regulatory trends, counterbalanced by API processing lags, with significant extreme-value risk tied to potential compounding pharmacy “mega-recalls.”

SUMMARY

Question: How many Class I drug recall records will openFDA return with a report date between July 6 and August 11, 2026? Final Prediction: Probability distribution:

• 10.00% chance of value below 0.99
• 20.00% chance of value below 1.895
• 40.00% chance of value below 3.47
• 60.00% chance of value below 5.845
• 80.00% chance of value below 11.99
• 90.00% chance of value below 19.99

Total Cost: extra_metadata_in_explanation is disabled Time Spent: extra_metadata_in_explanation is disabled LLMs: extra_metadata_in_explanation is disabled Bot Name: extra_metadata_in_explanation is disabled

Report 1 Summary

Forecasts

Forecaster 1: Probability distribution:

• 10.00% chance of value below 0.49
• 20.00% chance of value below 1.3
• 40.00% chance of value below 2.45
• 60.00% chance of value below 4.2
• 80.00% chance of value below 8.49
• 90.00% chance of value below 15.49

Forecaster 2: Probability distribution:

• 10.00% chance of value below 1.49
• 20.00% chance of value below 2.49
• 40.00% chance of value below 4.49
• 60.00% chance of value below 7.49
• 80.00% chance of value below 15.49
• 90.00% chance of value below 24.49

Research Summary

The research establishes a baseline using historical openFDA/enforcement volumes and a reference-class proportion for Class I drug recalls: year-to-date 2026 data showed 292 drug enforcement records over 144 days (≈61 records/month), historical annual totals were 765 (2025) and 657 (2024), and a scraped dataset of ~17,000 recall records contained 1,718 Class I recalls (≈9.8–10% of drug enforcement records). Applying the monthly average to the 37-day target window (≈75 total records) and the ~10% Class I share yields a statistical baseline of about 7–8 Class I records; an independent forecasting market anchor (Mantic) estimated 158 total drug enforcement records over a related 78-day window, which scales to the same ~7.5 Class I median for the 37-day window.

The research also identifies concrete recall events and operational factors that could shift that baseline upward: the Target baby-wipes elevation to Class I (July 2) could produce multiple distinct product records (estimated ≥8) if routed through the drug endpoint, Haleon’s Gas-X Class I assignment (early June) could result in delayed or supplemental records in July, and other high-profile Class I recalls in devices/food would not appear in the drug endpoint. It notes API and resolution quirks—weekly updates with historical lags, frequent multi-NDC clustering of single recall events, and possible post-resolution reclassification—that create right-skew risk where a single large recall could raise counts into the mid-teens or higher.

Sources used:

• openFDA Enforcement API: https://open.fda.gov/apis/enforcement/
• FDA Enforcement Reports / Recalls page: https://www.fda.gov/safety/recalls-market-withdrawals-safety-alerts/enforcement-reports
• Mantic forecasting platform: https://mantic.ai/

RESEARCH

Report 1 Research

1. Base Rates and Reference Classes

To establish an outside-view anchor, we first need to look at historical FDA drug enforcement volume and the specific proportion that falls under the “Class I” (highest severity) designation.

Overall Drug Enforcement Volume:

• 2026 Year-to-Date: As of late May 2026, the FDA had published 292 drug enforcement records over 144 days, which averages out to roughly 61 records per month [21].
• Historical Annuals: For broader context, there were 765 total drug enforcement records published in 2025, and 657 in 2024 [21].
• The Target Window: The question’s timeframe of July 6 to August 11, 2026 (inclusive) spans exactly 37 days (or roughly 1.2 months).

Class I Proportion:

• According to a historical scrape of the openFDA API representing over 17,000 pharmaceutical drug recall records, the classification breakdown is roughly: 14,020 Class II recalls (moderate risk), 1,674 Class III recalls (least serious), and 1,718 Class I recalls [4].
• This indicates that historically, Class I recalls make up approximately 9.8% to 10% of all drug enforcement records [4].

Calculated Base Rate Anchor: If we project the 2026 monthly average (61 total records/month) over the 37-day target window (~75 total records expected) and apply the historical Class I proportion (~10%), the statistical baseline for Class I records in this specific window is approximately 7 to 8 records.

Note on clustering: Keep in mind that a single recall event usually results in multiple enforcement records because the API counts unique product identifiers (like NDCs or UPCs) rather than distinct events [21]. A single recall involving multiple dosages or package sizes will generate a batch of records at once [8][11].

2. Relevant News and Catalysts

When looking at recent news for active or pending Class I recalls that could hit the API’s report_date during your July/August target window, I found several critical events.

• Target ‘Up & Up’ Baby Wipes (Elevated July 2, 2026): The FDA elevated a recall of Target’s fragrance-free and cucumber-scented baby wipes to a Class I designation on July 2 due to bacterial contamination (Burkholderia cepacia and gladioli) [8]. While wipes are generally categorized as cosmetics, medicated wipes can sometimes fall under OTC drugs. Because this recall involves 5 different count sizes for the fragrance-free line and 3 for the scented line, it could generate at least 8 distinct records if routed through the drug API [8]. Because it was elevated right before the question’s open date, the official enforcement report_date is highly likely to fall into your July 6 - August 11 window.
• Haleon Gas-X Extra Strength SoftGels (Assigned June 4, 2026): The FDA assigned a Class I recall to two varieties of Haleon’s Gas-X (120-count and 72-count) due to propylene glycol coolant leakage [11]. While the initial classification happened in June, openFDA enforcement reports are sometimes delayed by weeks in hitting the API [18]. It is possible that supplementary or delayed records for this event will be published in July.
• Irrelevant Noise to Ignore: You might see news about Class I recalls for GE HealthCare Infant Resuscitation Systems [9] and Utz (Zapp’s/Dirty brand) Potato Chips [10][12][14]. These are regulated under the FDA’s device and food jurisdictions, respectively, and will not appear in the drug/enforcement.json endpoint you are tracking.

3. Prediction Markets and Forecast Anchors

While I did not find a highly liquid prediction market specifically for the isolated Class I slice of this question, I did find a highly relevant forecasting benchmark on Mantic (a forecasting platform).

• Mantic Market Anchor: A recent competition question asked forecasters: “How many drug enforcement records will the FDA publish with a report_date between 5/26/2026 and 8/12/2026?” [21].
• Market Consensus: The latest estimate for that 78-day window was 158 total drug enforcement records [21].
• Extrapolating for your forecast: If the consensus expected 158 total records over 78 days, that translates to about 75 records for your 37-day window. Taking 10% of that for Class I records leaves us with a forecasted median of 7.5 records, which aligns perfectly with our calculated historical base rate.

4. API Quirks and Resolution Vulnerabilities

As you finalize your forecast, I want to highlight a few operational quirks of the openFDA API that could affect resolution:

• Update Cadence and Lag: The openFDA enforcement API is officially updated weekly, but historically, developers have noted lags between public FDA announcements and the data appearing in the JSON endpoints [18]. The dataset relies on the FDA’s weekly Enforcement Report [1].
• Reclassification: The question states that records added or reclassified after the resolution time are ignored. However, recalls are frequently initiated as unclassified and later elevated to Class I (as seen with the Target wipes) [1][8]. If a recall occurs in late July but isn’t classified as Class I until late August, it won’t count toward your total.

Summary for your forecast: The quantitative baseline for this question sits around 7 to 8 records. However, because single recall events (like the July 2nd Target wipes recall, if processed as an OTC drug) immediately generate multiple distinct NDC/UPC records, you should account for a heavily right-skewed distribution. A single larger manufacturer recall hitting the API in mid-July could easily push the count into the 15-30 range.

FORECASTS

R1: Forecaster 1 Reasoning

Conditioned on the question not being annulled, I think the Round 4 critic is right that I had probably pushed the center a bit too low in Round 3. I should not treat the fdadri “sparse recent regime” snapshot as authoritative live ground truth. The UPHP table showing two Class I BD ChloraPrep/FREPP rows with FDA notification date July 1, 2026 is important evidence that Class I drug records were still flowing immediately before the window, even though those particular rows miss the July 6 start. (uphp.com)

That said, I am not fully moving to the critic’s higher upper-middle distribution. The strongest bearish facts remain: this is a weekly openFDA/Enforcement Report administrative query, not a 37-day stream of public recall announcements; FDA defines enforcementreportdt as the weekly Enforcement Report date, and openFDA says the drug enforcement API is updated weekly. (fda.gov) The openFDA endpoint I checked currently reports last_updated: 2026-06-24, which reinforces that update latency is real and makes the August 5 report-date contribution at least somewhat at risk by the August 11 12:00 UTC resolution. (api.fda.gov)

I also continue to exclude the most salient non-drug catalyst: Target wipes are explicitly listed by FDA as Cosmetics, despite severe contamination concerns and multiple SKUs. (fda.gov) Known drug catalysts such as BD and Gas-X are real drug recalls, but their public announcements are in early June; they mainly show the product classes capable of producing Class I rows rather than serving as strong target-window upside. (fda.gov) FDA also notes that firms can issue public warnings before classification and Enforcement Report posting, so pre-window public recalls can still affect later report dates, but this cuts both ways with administrative lag. (fda.gov)

My final synthesis is: median back to about 3 records, not 2; about 40–45% on 0–2; meaningful mass on 3–6; and a fat but not dominant tail from a multi-NDC OTC/topical/antiseptic/injectable Class I event. I keep the upper tail broad, but slightly below the Round 4 critic’s suggested tail because of the effective 4–5 weekly-report window and possible openFDA lag.

Updated prediction:
Percentile 0.1: 0.05
Percentile 1: 0.15
Percentile 5: 0.38
Percentile 10: 0.49
Percentile 20: 1.49
Percentile 30: 2.20
Percentile 40: 2.49
Percentile 50: 3.49
Percentile 60: 4.49
Percentile 70: 5.49
Percentile 80: 8.49
Percentile 90: 16.49
Percentile 95: 30.49
Percentile 99: 68
Percentile 99.9: 118

R1: Forecaster 2 Reasoning

I mostly accept the Round 2 critique. The most important correction is as-of-resolution censoring: historical Class I proportions are mostly final-state data, while this market resolves from whatever openFDA has exposed by 12:00 UTC on August 11, 2026. FDA says recalls can be posted before classification and later updated once the hazard assessment is complete, so a late July/August recall that “eventually” becomes Class I may still miss this query at resolution. (fda.gov)

A new reason to take that seriously: the openFDA drug-enforcement endpoint I checked currently reports last_updated: 2026-06-24, even though today is July 5, 2026. That does not prove a persistent lag, but it raises the chance that the August 5 weekly report bin is absent or only partially mature by the August 11 cutoff. (api.fda.gov) The target is therefore more like four to five weekly bins, not a smooth 37-day process; FDA’s API definition describes the relevant enforcement-report date as the weekly report date. (fda.gov)

I still defend the basic outside-view anchor: openFDA is product-record based, not distinct-event based, and FDA’s product view can split one recall event into multiple recalled products. (fda.gov) The historical/Class-I share and recent total-volume anchor still point to a central value in the mid-single digits, not near zero. Also, the right tail remains real: a single sterile-drug, particulate, contamination, hidden-active-ingredient, or compounder event can add many records.

Known catalysts still do not move me much upward. Target baby wipes are officially listed as Cosmetics, not Drugs, so I continue to discount them for this endpoint. (fda.gov) Gas-X is a relevant drug recall, but it was announced June 4 and recent openFDA mirrors show the main Class I row as June 17, before the target window. (fda.gov)

Net update: I am lowering the center by about one record versus my previous round, putting more mass on 0–2, and keeping a fat upper tail. Conditioning on the question not being annulled.

Updated prediction:
Percentile 0.1: 0.05
Percentile 1: 0.20
Percentile 5: 0.49
Percentile 10: 1.49
Percentile 20: 2.49
Percentile 30: 3.49
Percentile 40: 4.49
Percentile 50: 5.49
Percentile 60: 7.49
Percentile 70: 10.49
Percentile 80: 15.49
Percentile 90: 24.49
Percentile 95: 40.49
Percentile 99: 85.49
Percentile 99.9: 119.49

Frontier Forecast — Post 598

Modal: 0-15 (73.0%) • frontier aggregate • 8m1s

Interpret Summary

• Reading: strict
• Type: strict
• Window: Class I drug enforcement records with report_date between 2026-07-06 and 2026-08-11 inclusive, as available at 12:00 UTC on 2026-08-11

Resolution sources/checks:

• https://api.fda.gov/drug/enforcement.json?search=classification:%22Class+I%22+AND+report_date:[20260706+TO+20260811]&limit=1: https://api.fda.gov/drug/enforcement.json?search=classification:%22Class+I%22+AND+report_date:[20260706+TO+20260811]&limit=1 - Resolver URL extracted mechanically from resolution criteria

Edge cases:

• A single recall action covering multiple NDC numbers produces multiple records, so the count can be significantly higher than the number of distinct recall events — the answer is records, not events.
• Records reclassified TO Class I after August 11, 2026 are excluded even if their report_date falls in the window.
• If FDA adds or backdates records with report_dates in this window after the resolution snapshot, they are ignored.

Frontier Views (5/5)

• frontier_1 - Modal: 0-15 (62.0%)

  • Class I drug enforcement records are rare relative to Class II/III. Over a 37-day window (~5–6 weekly Enforcement Reports), base rates suggest a handful to low-teens records, with allowance for a right tail if a large multi-NDC Class I recall occurs.

• frontier_2 - Modal: 0-15 (82.0%)

  • The question resolves to the openFDA meta.results.total for Class I drug enforcement records with report_date in [20260706, 20260811]. Direct queries of the exact resolver endpoint for comparable prior-year windows (Jul 6–Aug 11) returned: 2021=36, 2022=18, 2023=4, 2024=9, 2025=5.

• frontier_3 (revised) - Modal: 0-15 (82.0%)

  • Revised after adjudication: I completely abandoned the flawed 2.03 records/day proxy, which the adjudicator correctly diagnosed as an out-of-scope metric causing excessive probability mass in mid/high bins.

• frontier_4 - Modal: 0-15 (45.0%)

  • Historical Class I enforcement records average ~10-15 per month; 37-day window implies modal outcome 12-18 (bin_0). Right-skewed tail from occasional high-volume multi-NDC actions supports modest probability mass in bins 2-8.

• frontier_5 - Modal: 0-15 (85.0%)

  • Historical baseline for the same Jul 6–Aug 11 (37-day) window over 2019–2025 is 3, 3, 36, 18, 4, 9, 5 records. Excluding the 2021 (hand-sanitizer anomaly) and 2022 (Artri tainted-supplement anomaly) spikes, the non-anomalous base rate is mean ~5.2, median 5, range 3–9.

Adjudication

• Material notes

  • frontier_3: discount/material - Uses an off-target proxy metric (all-records or different timeframe) rather than the exact openFDA Class I meta.results.total for the specified Jul6–Aug11 window. This materially drives the lane to a much heavier mid/high-tail distribution than the exact-resolver historical data support.

• Guidance

  • frontier_3 materially misapplies an off-target proxy (all-records or unrelated forecasting output) to a question that resolves to the exact openFDA Class I meta.results.total for a specific window; this materially biases its distribution and merits discounting.

• Revision

  • Attempted revision for frontier_3; changed frontier_3.

Final Distribution (discrete bins)

Sources Checked

• Resolver: Resolver URL extracted mechanically from resolution criteria - https://api.fda.gov/drug/enforcement.json?search=classification:%22Class+I%22+AND+report_date:[20260706+TO+20260811]&limit=1
• frontier_1: https://open.fda.gov/apis/drug/enforcement/
• frontier_1: https://open.fda.gov/apis/drug/enforcement/understanding-the-api-results/
• frontier_1: https://learn.microsoft.com/en-us/connectors/openfdadrugip/
• frontier_1: https://www.fda.gov/safety/recalls-market-withdrawals-safety-alerts/enforcement-reports
• frontier_2: openFDA drug enforcement API (resolution_source_query)

wooley_rhino • run 36aca6c6 • $2.0 • 8m1s

Extracted Percentile Values: {0.1: 1, 1.0: 2, 5.0: 4, 10.0: 5, 15.0: 6, 20.0: 7, 25.0: 8, 30.0: 9, 35.0: 10, 40.0: 11, 45.0: 12, 46.0: 12, 47.0: 13, 48.0: 13, 49.0: 13, 50.0: 14, 51.0: 14, 52.0: 15, 53.0: 15, 54.0: 16, 55.0: 16, 60.0: 18, 65.0: 20, 70.0: 22, 75.0: 25, 80.0: 30, 85.0: 36, 90.0: 45, 95.0: 58, 99.0: 95, 99.9: 150}

Model’s Answer:

Systematic Analysis

Step one: Identify the exact metric. The question resolves to meta.results.total from the openFDA drug enforcement API for Class I recalls with report_date between 20260706 and 20260811. This is a 37-day window. Enforcement records—not distinct events—are counted, so a single recall with multiple NDCs generates multiple records.

Step two: Establish base rates from historical API data. The most directly relevant data points come from actual API queries:

• June 2025: 9 Class I drug enforcement records (30 days) → ~0.30/day
• July 2025: 5 Class I records (31 days) → ~0.16/day
• 2024 H1 average: ~13 Class I records/month → ~0.43/day
• Historical literature: Class I ≈ 10–14% of total drug recalls, with ~330 recalls/year

For a 37-day window at 0.16–0.43/day, the baseline range is roughly 6–16 records.

Step three: Assess mid-2026 environmental factors. Several large recalls were announced just before the window opens:

• Amgen ivabradine (~935,000 bottles, contamination risk) — potentially multiple NDCs → multiple records if Class I
• Corlanor/Sensipar recall (~1M bottles, foreign substance) — another multi-product recall
• Blood pressure drug recall (~11,000 bottles, unclassified as of publication)

If even 2–3 of these get classified as Class I with 3–5 NDC records each, this could add 6–15 records above baseline. However, classification timing is highly uncertain—some may fall outside the window or never reach Class I.

Step four: Identify critical assumptions.

• Assumption A: The large mid-2026 recalls (ivabradine, Sensipar) are classified as Class I before August 11. Probability of failure: ~40%. If wrong, the total could be much lower (near historical averages of 6–10).
• Assumption B: Historical monthly rates (5–13 records) remain applicable. Probability of failure: ~20%. If wrong (e.g., unusual spike or lull), outcomes could deviate substantially.
• Assumption C: The API query syntax works correctly at resolution time. The VisionAnalysis showed a parse error, but this appears to be a pre-window data access issue—the query format is standard and should work. Probability of failure: ~5%.

Step five: Low and high outcome descriptions.

• Low outcome (P5 ~4 records): Few recalls reach Class I classification within the window; the large contamination recalls are either not classified Class I or have report_dates outside the window; a quiet recall period similar to July 2025 (5 records/month → ~6 in 37 days) but slightly below average.
• High outcome (P95 ~55 records): Multiple large contamination recalls (ivabradine, Sensipar, blood pressure drug) are all classified Class I in the window, each generating 5–10 enforcement records across NDCs, stacking on top of a normal baseline of ~10 records. Additional recalls not yet in news coverage push the total toward 50+.

Step six: Construct distribution. Central estimate is ~10–14 records. Given the elevated 2026 recall environment and possibility of large multi-NDC Class I recalls, I’ll shift the median slightly upward from the 2025 run rate but maintain substantial uncertainty.

Percentile .1: 1 Percentile 1: 2 Percentile 5: 4 Percentile 10: 5 Percentile 15: 6 Percentile 20: 7 Percentile 25: 8 Percentile 30: 9 Percentile 35: 10 Percentile 40: 11 Percentile 45: 12 Percentile 46: 12 Percentile 47: 13 Percentile 48: 13 Percentile 49: 13 Percentile 50: 14 Percentile 51: 14 Percentile 52: 15 Percentile 53: 15 Percentile 54: 16 Percentile 55: 16 Percentile 60: 18 Percentile 65: 20 Percentile 70: 22 Percentile 75: 25 Percentile 80: 30 Percentile 85: 36 Percentile 90: 45 Percentile 95: 58 Percentile 99: 95 Percentile 99.9: 150

Enforcement reporting is constrained by statutory classification criteria and a weekly publication cadence that limit when new Class I records appear.

Counts measure product-specific enforcement records, so one recall event can produce many NDC-level entries and inflate the tally.

Recent same-window and year-to-date counts sit well below long-run highs, producing a central estimate near three records for the target window.

Public recall notices through mid-June show limited evidence of imminent, large-scale Class I activity that would raise the baseline.

A single multi-product Class I action during the window would pivot outcomes into the 20–30 record range and dominate the distribution.

Classification upgrades and the agency’s ingestion timing can move a few records into or out of the snapshot depending on when the cutoff is observed.

Key unresolved uncertainties are whether any pending severe firm actions will be escalated to Class I and whether those escalations post before the cutoff.

Residual tail risk from rare mass-product recalls keeps a small probability of substantially higher counts despite a low central estimate.

Forecast (Q598): median = 4.6683

• Recent pace is softer than usual: Through mid-2026, Class I drug recall records are running at about half the pace seen in 2023–2025 for similar summer windows.
• Historical window analogs are usually in the single digits: The exact July 6–Aug 11 window previously landed at 18 (2022), 4 (2023), 9 (2024), and 5 (2025)—suggesting a typical outcome around low single digits, with one big outlier.
• Main uncertainty is “one big event” multiplying records: Most of the forecast risk on the high side comes from the possibility that a single multi-NDC / multi-lot recall generates many enforcement records in the window.
• Right-edge cutoff/reporting lag matters: Because openFDA updates on a schedule and classification must happen first, records near the end of the window may not be fully captured by the Aug 11 resolution time.
• Result is right-skewed: The median is pulled below the mean because the distribution has a heavier right tail (rare spikes), not because low counts are impossible.