2026 US Cyclosporiasis Outbreak

If a food vehicle is ever identified, it will be a leafy-green/salad blend or a fresh herb (basil/cilantro), consistent with every resolved US Cyclospora outbreak since 2013, not an exotic new commodity.

v2 update (Jul 15): Trending toward confirmation. Michigan officials now cite lettuce or salad greens as the probable vehicle and FDA is investigating produce, consistent with this prediction, though no vehicle is officially confirmed. Confidence raised 70% to 80%.

v3 update (Jul 16): Effectively confirmed pending official word. The reported traceback lands on iceberg lettuce, squarely inside the predicted commodity class, and on a Mexican-grown import, matching the second half of the base rate. Confidence raised 80% to 90%. We withhold full resolution only because no agency has confirmed on the record.

v4 update (Jul 17): Confirmed. FDA and CDC now officially link the five-state Taco Bell cluster to iceberg lettuce. North Carolina's suspected (unconfirmed) foods for its separate cluster, parsley, cilantro, and lettuce, also sit inside this predicted commodity class. Confidence raised 90% to 95%.

v5 update (Jul 18): No new information this cycle bears on the commodity class itself. Taylor Farms' newly published recall covers the same commodity (iceberg lettuce), reinforcing rather than changing this call. Holding at 95%.

v6 update (Jul 19): FDA's own import surveillance returned a direct positive laboratory test for Cyclospora on Taylor Farms iceberg lettuce, the strongest form of confirmation for this commodity class to date, a physical detection rather than an epidemiologic or traceback inference. Raising confidence modestly, 95% to 97%.

v7 correction (Jul 20): reverting the v6 bump. The lab result v6 cited was retracted by FDA the next day as a false positive; see the correction note above the Situation Snapshot. Since the 97% figure was based specifically on that now-withdrawn physical detection, we are reverting to the pre-v6 figure rather than keeping a confidence level whose stated basis no longer holds. This does not disconfirm the commodity-class call itself, which still rests on the epidemiologic and traceback attribution for the named cluster. Confidence back to 95%.

v8 update (Jul 21): raising to 97%, on a different basis than v6 used. FDA's outbreak investigation page, retrieved directly this cycle, instructs consumers plainly: "Do not eat recalled iceberg lettuce from Taylor Farms de Mexico." That is a named agency naming the commodity and the supplier on the record, which is the standard we set for a finding. We are returning to 97%, the same figure v6 asserted and v7 reverted, and we flag that explicitly: the number is the same but its basis is not. v6 rested it on a laboratory detection that did not survive the week; v8 rests it on an agency statement that is still standing. Confidence 95% to 97%.

v9 update (Jul 22): Search-indexed results suggest CDC's own outbreak page has been renamed to explicitly name iceberg lettuce and Taylor Farms de Mexico, closing the interagency divergence v8 flagged, but this remains a lead pending direct confirmation rather than new confirming evidence. The commodity and supplier attribution itself is unchanged this cycle. Holding at 97%.

v10 update (Jul 22): The CDC-retitling lead from v9 remains unconfirmed this cycle. This cycle's material development, the CDC count unfreezing, bears on Predictions 3 and 4 below, not on the commodity/supplier attribution itself. Holding at 97%.

v11 update (Jul 23 to 24): not disconfirmed, but a related claim is now disputed. The CDC-retitling lead is search-confirmed this cycle (see the Bottom Line and the "What changed" note above). Separately, Mexico's Health Secretary disputed on the record whether the outbreak originated at the Taylor Farms de Mexico facility, a dispute about the specific supplier and country-of-origin link, not about the commodity class this prediction tracks (iceberg lettuce remains an undisputed leafy green regardless of where the contamination is ultimately traced). We are holding this prediction's confidence at 97% rather than lowering it, since no source disputes that lettuce is the vehicle for the named cluster; the supplier dispute is reflected instead in the Situation Snapshot's new "Attribution dispute" row.

v12 update (Jul 24 to 25): CDC's expansion of the named cluster to nine states is a widening of the same lettuce-linked cluster, not a new commodity, and Mexico's negative Guanajuato test bears on the supplier/origin dispute (see Prediction 2 and the Situation Snapshot), not on whether lettuce is the vehicle. Holding at 97%.

Given the long incubation window, produce commingling, perishable evidence, and stressed trace-back capacity, there is a better-than-even chance this outbreak closes without a single conclusively identified food source.

v3 update (Jul 16): trending against us. FDA's traceback reportedly reached a named supplier, commodity, and country of origin inside roughly ten weeks. That is a materially better outcome than we forecast, and we are marking the confidence down 60% to 40%. Two reasons we are not abandoning the call outright: the attribution is not yet on the record, and a five-state restaurant cluster may account for only part of a 34-state outbreak, leaving the remainder unattributed even if Taylor Farms is formally named. The 2013 precedent is instructive here: FDA named Taylor Farms de Mexico and still concluded that not all cases were related to each other.

v4 update (Jul 17): disconfirmed for the named cluster, open for the outbreak as a whole. FDA and CDC named a supplier, commodity, and country of origin on the record within about eleven weeks, for the Taco Bell cluster that call is now clearly wrong. We are splitting the resolution: disconfirmed for that cluster, but North Carolina's separate, still-unattributed cases show a real share of the 34-state case load remains without a named source, exactly the mechanism this prediction anticipated. Lowering overall confidence 40% to 25%, reflecting that the prediction's narrow form failed while its broader spirit is holding for part of the outbreak.

v5 update (Jul 18): No change to the named cluster's resolution. Holding at 25%.

v6 update (Jul 19): The import-surveillance positive test strengthens the named cluster's attribution further but does not bear on the outbreak-wide question this prediction is now tracking. Holding at 25%.

v7 update (Jul 20): The v6 lab result this prediction referenced was retracted (see the correction note above), but this prediction's split resolution never depended on that specific test, only on the on-the-record agency naming for the cluster and North Carolina's unresolved status for the outbreak-wide half. No change to either. Holding at 25%.

v9 update (Jul 22): ambiguous, not scored either way. FDA disclosed this cycle that five distinct cyclosporiasis clusters have been investigated since May, three now considered over. That structure is consistent with this prediction's outbreak-wide half (more than one source in play), but we do not know whether North Carolina's cluster is among the three closed ones, which would cut toward resolution, or the two still open, which would cut against it. Holding at 25% pending that confirmation.

v10 update (Jul 22): No source found this cycle addresses North Carolina's cluster status among FDA's five. The CDC count unfreezing (see above) is a national topline development and does not bear on outbreak-wide attribution specifically. Holding at 25%.

v11 update (Jul 23 to 24): the named-cluster half is less settled than "disconfirmed" now implies. Mexico's Health Secretary disputed on the record whether the outbreak originated at the Taylor Farms de Mexico facility, the same on-the-record naming that led us to disconfirm this prediction for the named cluster in v4. FDA and CDC have not withdrawn their attribution, so we are not reversing that disconfirmation, but a named government's dispute is enough to nudge our confidence in the outbreak-wide half up slightly: a second, separate FDA-disclosed outbreak (72 cases, product unidentified) is itself a concrete instance of a source not being conclusively named, consistent with this prediction's spirit. Raising confidence 25% to 30%.

v12 update (Jul 24 to 25): mixed, net small increase. Mexico's dispute firmed from a spoken assertion to a specific negative lab/environmental test at the implicated Guanajuato facility, a stronger form of disconfirming evidence for the named-cluster attribution than v11's statement alone, though FDA has not withdrawn its conclusion and Mexican officials themselves say a negative plant test does not resolve the question. Working against that: CDC's own expansion of the named cluster to nine states means more of the case load, not less, is now captured under the Taylor Farms attribution, which cuts toward the source being named more conclusively, not less. On balance we weight the negative test as the more direct evidentiary development. Raising confidence modestly, 30% to 32%.

As the under-analysis pool is adjudicated, the CDC-confirmed figure will rise substantially above the v1 figure of 843, narrowing (but not closing) the gap to the state aggregate, validating that the true burden sits well above the confirmed count.

v2 update (Jul 15): Confirmed. CDC-confirmed cases roughly doubled to 1,645 in two days and the under-analysis pool grew to >5,100 (nearly 7,000 combined), exceeding the ~4,700-case 2019 US record.

v3 correction (Jul 16): we called this too early. The prediction had two halves. The count did rise sharply (843 to 1,645), which v2 marked as confirmation. But the second half, narrowing the gap, has not happened and has in fact reversed: CDC has been static at 1,645 since Jul 13 while the state aggregate climbed past 7,500, so the gap widened from roughly 2.4x to 4.6x. Marking this back down from Confirmed to Partially confirmed and reducing confidence 80% to 55%. The underlying claim (true burden far exceeds the confirmed count) is holding; our model of how the numbers would reconcile was wrong. Adjudication appears to be a batch process with multi-day latency, not the continuous convergence we assumed.

v4 update (Jul 17): the gap kept widening, and our own charitable explanation looks weaker. CDC's national count is still static at 1,645, a five-day freeze, while Michigan alone is now above 5,000 and the state aggregate sits well past 7,500. More importantly, agencies published a fresh, current, cluster-specific count (1,644/94) in the same period they left the national topline untouched, which undercuts the pure "batch latency" reading we gave this in v3. We are lowering confidence further, 55% to 40%, and flagging this as the item most deserving of harder scrutiny next cycle.

v5 update (Jul 18): Every secondary source we found this cycle still cites 1,645, consistent with a continued (now apparent sixth-day) freeze, but this cycle's figure is secondary-sourced, not primary-verified, so we cannot rule out that sources are simply repeating a stale figure rather than confirming no change. Lowering confidence slightly, 40% to 35%, to reflect that verification gap rather than new disconfirming evidence. This remains the item most deserving of a direct primary-source check.

v6 update (Jul 19): Same verification gap persists; every secondary source we found still cites 1,645, and the figure remains secondary-sourced, not primary-verified. No new disconfirming or confirming evidence this cycle. Holding at 35%.

v7 update (Jul 20): Same verification gap, now a third consecutive cycle. Every source we found still cites 1,645, but the figure remains secondary-sourced, not primary-verified. Holding at 35% rather than lowering further on unverified secondary citations alone.

v8 update (Jul 21): freeze confirmed directly. CDC's surveillance page was retrieved and verified: it still reports the Jul 13 figures and was last updated Jul 15. The freeze is real, not a verification artifact. Confidence in the underlying claim (true burden exceeds the confirmed count) remains High; this prediction's specific "narrows toward convergence" clause remains disconfirmed. Holding the calibration figure at 35%.

v9 update (Jul 22): freeze now a ninth day; secondary-sourced this cycle. Every secondary source found this cycle still cites 1,645 as of Jul 21, and Michigan's state total alone has more than tripled the confirmed national figure while the topline has not moved since Jul 13. This cycle's figure is secondary-sourced, not primary-verified, so we cannot independently confirm the freeze this cycle. Lowering confidence slightly, 35% to 30%, to reflect the widening gap and the renewed verification gap together.

v10 update (Jul 22): confirmed, and honestly this is the outcome we described in v1 before losing confidence in it along the way. The freeze broke: CDC's confirmed count rose from 1,645 to 4,173, well above the v1 baseline of 843 (the prediction's first clause), and the gap to total estimated burden narrowed, not closed, exactly as the prediction's second clause specified. Using CDC's own combined figure (confirmed + under analysis, >11,500) as the best available proxy for total burden, the ratio of confirmed-to-total tightened from roughly 4.6x (the widest point, v3 to v9) to roughly 2.8x. We marked this down repeatedly (80% in v2, down to 30% by v9) because the freeze kept extending with no visible movement, and that skepticism was reasonable given what we could observe at each point, but the prediction's underlying model, that adjudication is a batch process which would eventually catch up rather than a permanent gap, is what actually happened. Raising confidence substantially, 30% to 88%, and marking Confirmed. We withhold 100% because the gap has narrowed, not closed, and a single catch-up batch does not establish that future updates will continue converging rather than freezing again.

v11 update (Jul 23 to 24): No source found this cycle shows CDC's national confirmed count moving from 4,173/41 states. One quiet cycle is not enough to call a second freeze, since Jul 22's own catch-up followed nine quiet days before it, so we are holding rather than reading this either way. Holding at 88%; a second consecutive freeze next cycle would be a stronger signal worth revisiting this figure over.

v12 update (Jul 24 to 25): the second consecutive freeze we flagged as the threshold to watch has now happened. No source found this cycle shows the national confirmed count moving from 4,173/41 states, a second straight quiet cycle since the Jul 22 catch-up. This is the specific signal v11 said would be worth revisiting the figure over, though it is only two cycles against a nine-day freeze before the last catch-up, so we are not reading it as a new freeze of comparable length yet. Lowering confidence modestly to reflect the renewed quiet period, 88% to 85%.

v14 update (Jul 26): the national figure is still frozen, but the state-level evidence underneath it moved hard in this prediction's direction. CDC's national count, read directly from its surveillance page rather than inferred from coverage, remains 4,173 across 41 states with 308 hospitalizations, a fourth consecutive quiet cycle. The convergence mechanism this prediction describes is nonetheless visible in the state data: Ohio nearly doubled in a week, 1,316 to 2,574, Indiana moved to 710 and West Virginia to 226, and CDC's own page still carries more than 7,400 cases awaiting laboratory confirmation, explicitly attributing many of them to Michigan and Ohio. That is precisely the backlog this prediction expects to resolve upward into the confirmed count. CDC's deployment of Epi-Aid staff to Ohio for interviews and data management is a further sign the bottleneck is processing capacity rather than case occurrence. We are holding rather than raising, because the confirmed figure itself is what the prediction is scored against and it has not moved, but we note the underlying pressure is building rather than easing. Holding at 85%.

v15 update (Jul 28): A fifth consecutive quiet cycle: no source found this cycle shows CDC's national confirmed count moving from 4,173/41 states. Michigan's own state total kept climbing regardless, to 9,253 (Jul 27, MDHHS), further widening the gap this prediction tracks between the frozen national figure and state-level reality. Holding at 85%.

v16 update (Jul 28): A sixth consecutive quiet cycle for CDC's national figure, still 4,173/41 states. Both Michigan (9,680, Jul 28) and North Carolina (718, Jul 28, up 28% in a week) kept climbing at the state level, widening the gap this prediction tracks further still. Holding at 85%.

v17 update (Jul 29): confirmed a second time. CDC's national confirmed count broke its plateau again, jumping from 4,173 to 6,707 across 45 states (up from 41), with hospitalizations rising from 308 to 423 and the under-analysis pool growing to >11,500. Using CDC's own combined figure (confirmed + under analysis, now >18,200) as the best available proxy for total burden, the confirmed-to-total ratio held roughly steady (about 2.8x before this update, about 2.7x after), continuing to narrow rather than close, exactly this prediction's second clause. That this is now the second such catch-up jump since the outbreak began, and this one followed a shorter plateau (about a week, versus the first freeze's nine days), further supports the batch-adjudication model this prediction has rested on since v10 rather than a one-off event. Raising confidence, 85% to 90%.

v18 update (Jul 30): No source found this cycle shows CDC's national confirmed count moving again from 6,707/45 states; one quiet cycle after two catch-up jumps is not yet a new plateau worth reading either way. Holding at 90%.

v19 update (Jul 30): confirmed against CDC directly, and held. CDC's surveillance page, fetched rather than read through press coverage, confirms 6,707 confirmed cases across 45 states with 423 hospitalizations and 0 deaths, data through Jul 27. The under-analysis pool stands at more than 11,500, so CDC's own combined figure remains above 18,200 and the confirmed-to-total ratio is essentially where v17 left it. Two state figures moved well beyond the national one this cycle (Michigan 10,386, Indiana 1,176), continuing to widen the gap this prediction tracks. No change to the reasoning; holding at 90%.

v20 update (Aug 2): the gap widened again on primary-verified state figures. CDC's national count is unchanged (6,707/45, page still dated Jul 28, next weekly update pending), while Michigan rose to 10,773, Indiana to 1,285 and West Virginia to 268, each confirmed against the state's own page. The mechanism this prediction describes, state totals running ahead of a lagging national confirmed count, held for another cycle. Corroboration without a national move does not raise the prediction; holding at 90%.

Independent of the clinical outcome, the lasting policy narrative will be the FoodNet reduction and CDC workforce cuts as a case study in eroded outbreak-response readiness, cited in future foodborne-illness and public-health-funding debates.

v3 update (Jul 16): Strengthened. Two developments feed this narrative directly: CDC's confirmed count stalling for three days while state totals surged, and the supplier lead reaching the public through a leaked source to reporters rather than an FDA advisory. Whatever the eventual attribution, "the press named the supplier before the agency did" is a durable talking point. Confidence raised 75% to 80%.

v4 update (Jul 17): Reinforced further. The advisory and recall arrived only after five days of a frozen national count and mounting press pressure, and a fresh cluster-specific figure landed before the topline did. Confidence raised 80% to 83%.

v5 update (Jul 18): No material new evidence this cycle bears directly on this prediction. Holding at 83%.

v6 update (Jul 19): mixed evidence, not simply reinforcing. FDA's targeted import surveillance catching a contaminated lot at the border is a capability-positive data point, proactive federal testing worked, which cuts against a pure "eroded surveillance" reading and is worth stating plainly rather than folding only the negative half of this story into the narrative. But the same disclosure did not include the lot number, ship dates, or destination states, a transparency gap Marler Blog's coverage criticized directly, and that gap continues to feed the durable-story thesis on the disclosure side even as the detection side cuts the other way. Net effect: lowering confidence slightly, 83% to 81%, to reflect that the evidence this cycle is genuinely two-directional rather than one-sided.

v7 correction (Jul 20): the "capability positive" half of the v6 read does not survive the retraction. The catch that v6 credited as proactive federal testing working was itself a false positive, meaning the capability point cuts the other way: rather than "surveillance caught contamination," the accurate description is "surveillance produced an erroneous result that required a public correction and apology." That is a capacity-strain data point, not a capacity-strength one. Combined with the pre-existing disclosure-gap criticism (still unresolved) and a still-unverified national topline entering a third cycle, we are not just reverting to 83%, we are raising above it: confidence up 81% to 85%.

v8 update (Jul 21): No material new evidence bearing on this prediction; v8's primary-source verification pass confirmed the freeze rather than adding to the capacity narrative directly. Holding at 85%.

v9 update (Jul 22): mixed again. Finding, this cycle, that our own v8 correction was applied inconsistently across two of the five briefing pages is itself a capacity-and-diligence data point, though about our own product rather than the federal response, and we note it rather than let it pass unremarked. On the federal side, FDA's disclosure that three of five investigated clusters are already closed is a modest capability-positive signal, cases can apparently be resolved, that we had not previously credited. Net effect small and mixed; holding at 85%.

v10 update (Jul 22): mixed, and this cycle's biggest data point cuts against us for the first time. CDC's confirmed count unfreezing, and Prediction 3's vindication above, is itself evidence that the adjudication backlog was a batch-processing lag rather than the kind of durable capacity failure this prediction's narrative depends on: the system produced a large, coherent catch-up update rather than staying silently frozen. That is a genuine capability-positive data point we should not undercount just because it complicates our thesis. Against that, a nine-day freeze followed by a single 2.5x jump is also consistent with a surveillance system that is stretched thin enough to batch rather than stream its reporting, which is itself the kind of strain this prediction anticipated becoming a policy talking point. We are lowering confidence modestly, 85% to 80%, to reflect that this cycle's central fact cuts against the erosion narrative more than it supports it, while leaving the underlying FoodNet-cuts and workforce-reduction story untouched.

v11 update (Jul 23 to 24): reinforced again. Three developments feed the durable-story narrative this cycle: a named foreign government publicly disputing the US agencies' own traceback conclusion is exactly the kind of contested, uneven public record this prediction anticipated becoming a policy talking point; a Cabinet-level "under control" statement sitting in tension with continued state-level case growth invites exactly the credibility scrutiny this prediction tracks; and a second, wholly separate outbreak with an unidentified product surfacing in the same week as the first is evidence the underlying detection and traceback system is handling more concurrent unknowns than a single, resolved narrative would suggest. None of this is a reversal of last cycle's capability-positive read on the CDC catch-up, which stands on its own; the overall picture is again mixed rather than one-directional. Raising confidence modestly, 80% to 83%.

v12 update (Jul 24 to 25): reinforced further. A named cluster quietly growing from five to nine states between briefing cycles, discovered via CDC's own page retitling rather than a proactive announcement, is itself the kind of uneven, after-the-fact disclosure pattern this prediction has tracked since v3. A foreign government's dispute escalating from a spoken statement to an actual negative environmental test adds a second concrete, verifiable data point to an increasingly public and contested record. Against that, the national topline holding steady for a second cycle is a mild capability-neutral signal (no fresh instability, but no fresh resolution either). Raising confidence modestly, 83% to 85%.

v14 update (Jul 26): the clearest single data point this prediction has had. CDC is deploying Epidemic Intelligence Service officers to Ohio under its Epi-Aid program for roughly three weeks, at Ohio's own request, specifically to assist with case interviews, data management, cluster identification, and coordination with federal traceback. That is a state health department formally declaring it cannot process its own outbreak workload, and a federal agency answering with surge staffing. It maps directly onto this prediction's claim that the durable story is capacity rather than the pathogen: the tasks being backfilled are exactly the interview-and-analysis functions we have argued were thinned, and the request arrives as Ohio's count nearly doubles in a week and more than 7,400 cases nationally sit awaiting confirmation. We stop short of treating it as proof of the causal claim, since Epi-Aid requests are a routine and healthy mechanism that predates any recent reduction, and a state asking for help is evidence of strain but not of what caused the strain. Raising confidence, 85% to 89%.

v17 update (Jul 29): mixed, similar to v10's reasoning the first time this happened. CDC's confirmed count breaking its plateau a second time, cleanly and coherently (6,707 across 45 states, up from 4,173/41), is again a capability-positive data point: the system produced a large, internally consistent catch-up rather than staying silently frozen, and this plateau (about a week) was notably shorter than the first (nine days), which could read as the adjudication pipeline recovering rather than degrading further. Against that, a repeating pattern of plateau-then-batch-jump, now observed twice, is itself evidence of a system that cannot stream its reporting continuously, which is the kind of structural strain this prediction's narrative anticipates becoming a policy talking point, and Indiana's figure this cycle could only be sourced to press rather than verified directly against IDOH's own dashboard, a verification gap consistent with the strained-capacity thesis. Net effect small and mixed, as in v10; holding at 89%.

v18 update (Jul 30): net positive, reversing last cycle's verification-gap concern. Indiana's figure, which could only be sourced to press last cycle, is this cycle directly and consistently attributed to IDOH's own dashboard by three independent outlets, closing the specific verification gap v17 flagged as a capacity-strain signal. Michigan's own health department also volunteered a capacity-relevant explanation for its own numbers this cycle, attributing about 475 recent cases to a reporting backlog rather than new illness and saying the trend is slowing, a level of self-diagnosis and disclosure this prediction's narrative treats as a positive rather than negative data point, though we treat the state's "worst has passed" characterization as a claim rather than a verified fact. Raising confidence modestly, 89% to 90%.

v19 update (Jul 30): the capacity problem this cycle was ours, not the agencies'. Last cycle raised this prediction partly because Indiana's verification gap had closed. It had not: IDOH's dashboard was publishing 1,176 cases while we were preparing to publish approximately 954 on press attribution, and the dashboard answered on the first request. The same applies to the FDA table and the JPML docket, both of which were public and unread. That is a reporting-capacity failure in this briefing rather than evidence about federal or state surveillance capacity, and it cannot support the raise it was used for. Reverting last cycle's increase, 90% back to 89%.

v20 update (Aug 2): unchanged at 89%. Nothing this cycle bears on agency surveillance capacity either way. The one process signal worth logging is our own: the states' dashboards again answered on the first request and the JPML docket sweep ran cleanly, so the constraint remains whether this briefing fetches, not whether the data exists to fetch. CDC's page has not refreshed since Jul 28, which is consistent with the weekly cadence rather than evidence of a stall. Holding at 89%.

Even if FDA formally names Taylor Farms iceberg lettuce, a substantial share of the 34-state case load will remain unattributed to that vehicle. The reported cluster spans five states and one restaurant chain; the outbreak does not. We expect the final picture to include either much broader distribution of the same supplier's product beyond Taco Bell, or a second concurrent vehicle. The 2013 analogue closed exactly this way, with FDA noting that not all cases were directly related.

v4 update (Jul 17): trending confirmed. North Carolina's health department states its 307 cases are not part of the Taco Bell cluster, direct evidence of exactly the second-vehicle mechanism this prediction anticipated, one week after we wrote it. FDA's own attribution remains scoped to five states. Confidence raised 70% to 85%.

v5 update (Jul 18): complicated, not simply confirmed further. Taylor Farms' newly published recall distributes the same implicated product to 27 states, including North Carolina, the very state whose cases we cited as evidence of a clean second vehicle. That does not disconfirm NC DHHS's statement that its cases are unlinked, but it means the "broader distribution of the same supplier's product" half of this prediction and the "second concurrent vehicle" half are no longer cleanly separable in North Carolina's case: the same supplier's product reached NC through non-Taco-Bell channels, so a future finding that NC's cases trace to Taylor Farms after all would resolve this prediction's first clause rather than its second. We are lowering confidence slightly, 85% to 78%, to reflect this added ambiguity rather than raising it further on an unexamined assumption.

v6 update (Jul 19): The import-surveillance positive test bears on attribution confidence for the named cluster, not on North Carolina's unresolved status. No source we found this cycle addresses the NC overlap. Holding at 78%.

v7 update (Jul 20): The retracted lab result (see correction note above) does not bear on the North Carolina overlap either; no source this cycle addresses it. Holding at 78%.

v8 update (Jul 21): No new evidence on the North Carolina overlap this cycle. Holding at 78%.

v9 update (Jul 22): modest reinforcement. FDA's disclosure that officials have tracked five distinct clusters since May, not just the Taco Bell cluster and North Carolina, is on-the-record confirmation that the outbreak's structure is more than a single vehicle even before North Carolina's own status is resolved. That is consistent with this prediction's core claim, that attribution would not close the whole case. North Carolina's cases themselves remain unaddressed by any source we found this cycle. Raising confidence modestly, 78% to 80%.

v10 update (Jul 22): reinforced. CDC's confirmed count now spans 41 states, up from 34, while the named Taco Bell cluster remains scoped to five. A 41-state confirmed footprint is not plausibly explained by one restaurant chain's five-state cluster alone, which is consistent with this prediction's core claim that attribution will not close the whole case. This is an inference from the widened state count, not a new agency statement addressing the outbreak-wide question directly, so we raise confidence modestly rather than resolve the prediction. 80% to 83%.

v11 update (Jul 23 to 24): confirmed further, on stronger evidence than last cycle's inference. FDA disclosed a second, wholly separate cyclosporiasis outbreak of at least 72 cases, explicitly not linked to the Taylor Farms lettuce cluster, with its own product still unidentified. Unlike v10's inference from the widened state count, this is a direct agency statement that more than one distinct source is active concurrently, precisely this prediction's core claim. Raising confidence 83% to 86%.

v12 update (Jul 24 to 25): this cycle's biggest development cuts against us, and we say so. CDC's expansion of the named cluster from five to nine states means the Taylor Farms attribution now explains more of the national case load, not less: the cluster-specific count rose from 1,644 to 1,947 while the national topline held flat at 4,173, narrowing rather than widening the share of cases outside the named attribution. That is the opposite direction from what this prediction anticipates. We are not reversing the overall call, a 1,947-case cluster is still a minority of the 4,173-to-11,500 national picture, and the separate 72-case outbreak and North Carolina's unresolved status both still stand as genuine second vehicles, but honest calibration means marking this cycle's net movement down. Lowering confidence, 86% to 83%.

v14 update (Jul 26): reinforced on the strongest evidence yet, and it corrects an understatement of our own. Read directly, FDA's active investigation table lists the iceberg lettuce cluster alongside three other active Cyclospora investigations, refs #1392 (72 cases), #1384 (18 cases), and #1385 (10 cases, up from 8), each with product "not yet identified" and traceback initiated, plus ref #1381 now listed as over with its investigation still open. From v11 through v13 we reported only one additional distinct outbreak and explicitly declined to assert a running total because outlet coverage varied. The agency's own table resolves that: multiple distinct Cyclospora sources are active concurrently, on the record, which is this prediction's core claim stated directly rather than inferred. Because this correction runs in our own favor, we flag it as the kind we should scrutinize hardest, and we note it changes the count of known vehicles rather than resolving where any of them came from. Raising confidence 83% to 87%.

v16 update (Jul 28): North Carolina, the outbreak's clearest standing second-vehicle candidate, grew 28% in a week (561 to 718) while NCDHHS's own no-link statement is unchanged. This is continued growth of a vehicle this prediction already counts, not a qualitatively new one, so we are holding rather than raising confidence again on the same evidence category. No product ID for FDA's three other concurrent investigations this cycle either. Holding at 87%.

v17 update (Jul 29): a new lead, not yet scored. North Carolina's own figures are unchanged this cycle (718/24). A Reuters report quotes former FDA Commissioner Scott Gottlieb, on the record, saying large retailers have pulled back from central-Mexico produce broadly and that officials are searching for other potentially implicated products; this is directionally consistent with this prediction's core claim, but Gottlieb is a named former official, not a current FDA or CDC statement, and no specific new product is identified, so we log it as a watch item rather than move the confidence figure on it. No product ID for FDA's other concurrent investigations this cycle. Holding at 87%.

v18 update (Jul 30): the lead gets more specific, and now names North Carolina. In a CNBC interview, Gottlieb named clusters in the Great Lakes region, New York, and North Carolina specifically as potentially linked to parsley and cilantro from central Mexico, which converges with NCDHHS's own previously reported suspected foods for its unlinked cluster. This is a materially more specific claim than the general "central Mexico produce" framing carried since v17, but Gottlieb remains a named former official, not a current FDA or CDC statement naming a specific product or tracing a specific shipment to North Carolina, so we are not upgrading North Carolina's status from unlinked to linked. Directionally confirming for this prediction's core claim; raising confidence modestly, 87% to 88%.

v19 update (Jul 30): the strongest supporting evidence we were citing turns out to be half its claimed size. This prediction has leaned on FDA carrying four concurrent unidentified-product Cyclospora investigations as evidence that attribution is not closing. Read correctly, two of those four outbreaks are listed as Ended with only the investigations still open, leaving two ongoing (ref #1385 at 10 cases, ref #1392 now at 93). Fewer open unattributed outbreaks is evidence against this prediction, and we had the number wrong in our own favour for five versions. Gottlieb's North Carolina claim is unchanged this cycle and remains a former official's statement, not an agency finding, so it cannot offset this. Lowering confidence, 88% to 86%.

Civil suits against Taco Bell franchisees, Taco Bell corporate, and Taylor Farms will accumulate faster than the regulatory record supports, with the first filed (Ayyad v. Pacific Bells, LLC) before any agency confirmed a source on the record. Expect plaintiff-side filings to become a de facto public information channel about the outbreak, which is a poor substitute for agency reporting because the incentives are adversarial rather than epidemiological.

v4 update (Jul 17): No new litigation developments identified this cycle. Holding confidence unchanged pending the next check.

v5 update (Jul 18): confirmed further. Multiple law firms filed additional suits this week, including Caruso v. Taco Bell Corporation, et al. (N.D. Ohio), that for the first time name Taylor Fresh Foods, Inc. and Taylor Farms California, Inc. directly as defendants alongside Taco Bell, not just a franchisee. Plaintiff filings continue to outpace any formal agency finding about the outbreak as a whole. Confidence raised 85% to 90%.

v6 update (Jul 19): No new litigation developments identified this cycle. Holding at 90%.

v7 update (Jul 20): No new litigation developments identified this cycle. Holding at 90%.

v8 update (Jul 21): No new litigation developments identified this cycle. Holding at 90%.

v9 update (Jul 22): confirmed further. Additional suits reported this cycle, including a 27-year Army veteran's suit against Taco Bell and Taylor Farms and a Michigan couple's federal suit filed Jul 17 (E.D. Michigan), both naming Taylor Farms entities directly, continue to accumulate faster than any agency finding about the outbreak as a whole. Confidence raised 90% to 92%.

v10 update (Jul 22): No new litigation developments identified this cycle. Holding at 92%.

v11 update (Jul 23 to 24): No new litigation developments identified this cycle. Holding at 92%.

v12 update (Jul 24 to 25): No new litigation developments identified this cycle. Jul 24 legal-press coverage of a Michigan couple's suit against Taylor Farms and Taco Bell describes the same Jul 17 E.D. Michigan filing already carried in this briefing since v9, not a new suit. Holding at 92%.

v13 correction (Jul 25 to 26): v12's "no new litigation" call was wrong. The Jul 24 legal-press coverage v12 dismissed as duplicating the Jul 17 Parrish filing was in fact reporting a distinct fifth suit, Hercula v. Taco Bell of America, LLC, et al. (filed Jul 20, E.D. Michigan, Clinton Township plaintiffs, via Marler Clark), confirmed by cross-referencing plaintiff names, home counties, restaurant locations, exposure dates, and damages sought against the already-tracked Parrish suit, all of which differ. A second, independently filed Michigan federal suit landing three days after the first is itself supporting evidence for this prediction's core claim, that filings accumulate faster than the regulatory record, and is also a reminder to verify litigation claims against plaintiff-level detail rather than treating "a Michigan couple sued" as a single fact. Raising confidence modestly, 92% to 93%, and recording the correction in the status file per our own hedge-discipline standard.

v15 update (Jul 27 to 28): confirmed further, and a gap in our own tracking closed. Two more suits surfaced this cycle: Gelbspan v. Taylor Farms (Kait Gelbspan and Victor Granados, filed Jul 16, N.D. California), the outbreak's first class action, seeking damages for consumers who bought Taylor Farms lettuce without either plaintiff claiming to have gotten sick; and Ott v. Taco Bell Corp., et al. (David Ott, filed Jul 17, Mahoning County, Ohio, a hospitalized Army veteran naming Taylor Fresh Foods and franchisee Charter Foods directly), which this briefing had cited only as a background source since v9 but never carried into this prediction's tracking or the Situation Snapshot's litigation row. A docket pull through CourtListener's RECAP archive then showed the count was low: at least eight distinct matters are confirmed, spanning four federal districts and one Ohio state court, including two we had never tracked at all (Null, C.D. California Jul 21 and a related N.D. California filing Jul 23; and Burnham, E.D. Michigan Jul 23, six plaintiffs). A new legal theory, the economic-loss class action, now sits alongside the personal-injury suits. This is direct, continuing evidence for this prediction's core claim, and the evidence is stronger than the prediction's own tracking had captured. Raising confidence to 95%. For our own accountability: Ott's suit sat in our sources list for six versions before we integrated it, and two further suits went unnoticed for a week because every prior cycle accepted press coverage as the ceiling on what was findable. The dockets were free and public the whole time. Litigation tracking in this briefing is docket-based from this version forward.

v16 update (Jul 28): No new litigation developments identified this cycle: no new suit, ruling, consolidation, or settlement found among the eight tracked matters. Holding at 95%.

v17 update (Jul 29): No new litigation developments identified this cycle: no new suit, ruling, consolidation, or settlement found among the eight tracked matters. Holding at 95%.

v18 update (Jul 30): a ninth matter, press-sourced only. A Muskegon, Michigan woman sued Taylor Farms, Taylor Fresh Foods, and, for the first time in this briefing's tracking, Taco Bell franchisee Border Foods (plus five unnamed corporations), filed Jul 28 in the US District Court for the Western District of Michigan; we could not obtain a docket number this cycle, so this addition does not yet meet the docket-verified standard this briefing has held litigation to since v15. Continued accumulation of filings, still with no ruling, consolidation, or settlement found. Holding at 95% (near the top of this briefing's confidence scale already).

v19 update (Jul 30): the ninth suit is docket-verified, and consolidation is formally in motion. O'Hara v. Border Foods, LLC is confirmed at W.D. Michigan, No. 1:26-cv-02183, filed Jul 28, opening a fifth federal district and naming Taylor Farms California, Inc. alongside the franchisee. More significantly, a docket sweep found IN RE: 2026 Multistate Cyclosporiasis Outbreak Litigation, MDL No. 3196, docketed before the JPML on Jul 17 with an initial Motion to Transfer covering Granados, Ayyad and Parrish. A consolidation petition within a month of the first filing is a strong structural signal that the plaintiffs' bar is treating this as durable mass-tort litigation, which is precisely this prediction's claim, and it is the first procedural development of consequence in the case. Against that, this prediction's own tracking missed it for thirteen days across three cycles that each affirmatively reported no consolidation found. The evidence supports the prediction; our monitoring of it does not yet merit confidence. Holding at 95% rather than raising, on the view that a prediction should not be rewarded for evidence it failed to find.

v20 update (Aug 2): the MDL cleared its first procedural gate, and this cycle we saw it in real time. A fresh docket sweep found the Panel formally assigned MDL No. 3196 and issued an Order to Show Cause on Jul 29, the first adjudicative movement in the consolidation. That is squarely this prediction's thesis: the litigation is being organized as durable mass tort faster than attribution is resolving. Unlike v19, the tracking caught this within days rather than missing it for weeks, so the monitoring objection that capped v19 at 95 is easing. We nonetheless hold at 95%: the raise this evidence would justify is small, the Panel has not actually ruled on transfer, and one clean cycle does not yet offset three blind ones. Holding at 95%.