Scenario wins: SynapseSeer (66) hayek-bot (49) Panshul42 (38) preseen (23) lewinke-thinking-bot* (15) smingers-bot (8)
Question: how many unique Cornell ILR Labor Action Tracker work‑stoppage starts will appear with start dates from 2026-06-09 through 2026-08-11 on the tracker as visible on 2026-08-11? Forecasters agree the target counts unique work‑stoppages (not map pins) and is sensitive to Cornell’s manual verification/backfill practices, so the Aug. 11 snapshot likely understates some late‑July and Aug. 1–11 starts. Historical Cornell totals (peaking in 2023 then falling to 303 in 2025) and monthly summer activity show June–August is seasonally busy, but published monthly “active” counts double‑count multi‑month stoppages, so the numerical proxy near ~100 is an upper‑biased guide. Teams therefore downweight that proxy and instead blend a uniform annual proration baseline (roughly mid‑50s for a 2025‑like year) with summer seasonality and live 2026 flashpoints (hospitality, aerospace, utilities, auto‑supplier disputes) to get a central range. Individual models land in the mid‑60s to ~80s as medians with broad right tails for a hotter labor summer and left tails for reporting lag or lower activity; one cited a median near ~72 with a ~5–95% range ≈41–110. Recommended monitoring priorities are obtaining Cornell’s event‑level spreadsheet and tracking the tracker weekly in June and daily in early August to separate true activity from listing lag.
Forecast: Probability distribution:
We need the number of work stoppages with start dates in 2026-06-09 to 2026-08-11 (inclusive) that are already present in the Cornell ILR Labor Action Tracker as of 2026-08-11. Key drivers are (1) underlying start frequency in that ~64-day window and (2) reporting/entry lag for late-July/early-August starts.
Across analysts, there is agreement on recent annual totals from the tracker (not BLS “major stoppages”): roughly 279 (2021), 433 (2022), 470 (2023), ~359 (2024), and 303 (2025). A naive annual proration for a 64-day window (64/365 ≈ 0.175) would suggest about 50–85 starts depending on whether 2026 is closer to 2025’s 303 or the higher 2022–2024 period.
The main disagreement is about summer seasonality and whether historical June/July/August monthly figures imply a much higher true-window count (~95–105) versus a more modest summer uplift. I think the “~100 true in-window” estimates (Forecasters 3/4) are somewhat high for 2026 conditional on the clear decline into 2025, and also risk some mismatch between what those monthly report tables count (starts vs other monthly measures). Meanwhile, the lowest forecast (Forecaster 1) likely underweights seasonality by assuming the window is roughly “average days of the year”.
On lag: the question resolves to the snapshot on Aug 11, so some fraction of late July and especially Aug 1–Aug 11 starts will be missing. I weight a moderate undercount (roughly 15–25% vs the eventual fully-updated count for that window), not as low as 5–10% and not as high as ~30% as a central tendency.
Putting these together, my central view is an eventual true-window count around ~85–95 if 2026 resembles a slightly lower-activity continuation of 2025 with summer uplift, and a recorded-by-Aug-11 count around the low-to-mid 70s. I keep a wide upper tail for a contract-cycle spike or unusually rapid entry, and a meaningful lower tail for continued decline plus slower data entry.
I assign negligible probability to exceeding 200 in this 64-day window, so I keep all percentiles within the in-range integers.
An amateur forecast is likely to (a) use BLS “major work stoppages” (tens per year) and badly underpredict, or (b) simply prorate the most recent annual total by days and miss both summer seasonality and the fact the question resolves to a same-day snapshot with reporting lag. My forecast explicitly separates (1) underlying start-rate seasonality and (2) as-of reporting lag, and it also widens the distribution to reflect uncertainty about 2026 being closer to the lower 2025 regime versus a rebound year. Confidence in improvement: moderate (the biggest remaining uncertainty is quantifying the tracker’s real-time completeness near the cutoff date).
Forecast rationale (numeric):
— Iteration 1 — Across the forecasts, the core reasoning is consistent:
There is broad agreement that the raw 64-day count should be reduced somewhat to account for reporting delay, producing a central estimate in the mid-50s to low-60s:
The main disagreement is how high the underlying daily rate is and how large the reporting lag discount should be:
The shared view is that the tracker will likely record roughly 55–65 work stoppages as beginning in that period, with substantial uncertainty driven less by the raw strike rate than by verification lag and late-period undercounting.
— Iteration 2 — The core reasoning is a base-rate projection adjusted for tracker timing effects:
Overall consensus: the likely result is around 30–50 recorded stoppages, with a central tendency near the high 30s to mid-40s, and substantial uncertainty driven mainly by reporting lag and volatility in strike activity.
— Iteration 3 — Across the forecasts, the reasoning is broadly consistent:
The forecasts collectively suggest that the count should be estimated by taking a pro-rated share of a fairly high annual stoppage rate, then discounting for summer effects and tracker reporting delays, producing a likely result in the high-60s to low-80s, with substantial uncertainty around that center.
To forecast the number of work stoppages recorded by the Cornell ILR Labor Action Tracker during this specific window, the rationales uniformly focus on three overlapping dynamics: the macroeconomic baseline of labor unrest, structural summer seasonality, and the tracker’s administrative reporting lag.
Declining Macro Baseline Forecasters note a steady downward trend in the total frequency of U.S. strikes since the peak of labor unrest in 2023. This decline is heavily attributed to a structural reduction in “micro-strikes” (such as single-location retail and fast-food walkouts). Unions have increasingly pivoted away from these fragmented actions toward large-scale master contract negotiations. Furthermore, early 2026 indicators of a cooling labor market and slightly rising unemployment are expected to generally depress the baseline frequency of strikes, even as sticky inflation sustains militancy in certain sectors like healthcare and hospitality.
Summer Seasonality The exact timing of the 64-day window significantly influences the expected volume of strikes. While the summer months often see a cluster of strikes tied to fiscal-year collective bargaining expirations (typically around July 1), the overall frequency of labor actions usually dips. This is primarily because K-12 schools and universities are out of session, creating a “structural zero” for the education sector, which historically accounts for a massive portion of the tracker’s total data.
Administrative Reporting Lag and the Cutoff Effect The most critical factor identified across all rationales is the strict resolution criterion: strikes must be visibly listed on the tracker on the exact final day of the window (August 11). The Cornell ILR tracker is not an automated system; it relies on university researchers manually scraping, verifying, and corroborating local news and social media reports. This manual process creates a massive truncation effect.
While work stoppages that begin in June will have ample time to be processed and published, forecasters universally agree that strikes initiating in late July and early August will suffer from severe undercounting. A large portion of these late-window strikes will remain trapped in the verification backlog and miss the August 11 snapshot entirely. Additionally, the possibility of irregular “batch” data uploads and reduced student researcher staffing during the summer introduces further downward pressure on the final observable count.
Modal: 25-50 (52.0%) • partial_shift • 7m9s
Reading: strict
The resolution criteria specify a precise snapshot date (August 11, 2026), an explicit inclusive date window, a named source (Cornell ILR tracker), and explicitly exclude entries added after the snapshot — all of which demand a narrow, literal reading with no room for charitable expansion of scope or timing.
Total evidence registered (shared pool): ?
| Variant | Perspective | Model | Turns | Tools | Status |
|---|---|---|---|---|---|
| 0 | inside_view (inside_view_v1) | openai/gpt-5-mini | 25 | 25 | OK |
| 1 | outside_view (outside_view_v1) | anthropic/claude-sonnet-4-6 | 25 | 33 | OK |
| 2 | contrarian (contrarian_v1) | openai/gpt-5 | 10 | 10 | OK |
Evidence confidence: medium
high evidence]2025 annual pace (~0.83/day) continues into 2026; arrival forecast median of ~44–48 falls in this range; standard reporting lag reduces count modestly below the full underlying rate for the 64-day window; seasonal elevation (June–August is high season per sources 19, 20) partially offsets any cooling trend.
If 2026 sees a further decline from the 2025 pace (which itself fell 17% from 2024), or if reporting lag at the August 11 snapshot date is unusually high, the count could fall below 35. Conversely, if the 2022–2023 elevated environment returns, the count could exceed 65.
medium evidence]Reporting lag is substantial for events near the end of the window; stoppages starting in late July and early August may not yet be entered into the tracker by August 11 (source 8, 11); if 2026 activity declines from 2025 (which already fell 17% from 2024), daily rate could be meaningfully below 0.83/day; arrival forecast p25 is approximately 25–31 (sources 9, 10).
Multiple data points suggest sustained or elevated activity in 2026 (source 6); seasonal June–August period is historically high (sources 19, 20); arrival forecast p5 is only 9–15, suggesting sub-35 is a tail outcome not the central case.
medium evidence]If 2026 labor activity rebounds toward the 2022–2023 pace (~433–471/year, ~1.2–1.3/day), a 64-day window would see 75–85 new stoppages; arrival forecast p75 is approximately 69–72 and p90 is 93–104 (sources 9, 10); early 2026 indicators show continued notable strikes (source 6).
The trend from 2023 through 2025 is clearly downward (471→368→303); there is no direct evidence of a reversal in 2026; arrival forecast median is well below 65, so this scenario corresponds to the upper quartile of the distribution.
low evidence]The tracker snapshot on August 11 may be especially sparse if Cornell is behind on data entry (source 11 notes rolling additions with variable lag); if 2026 activity fell sharply — e.g., to 2021 levels or below (~0.73/day) — a 64-day window might yield fewer than 20 entries by that date; arrival forecast p5 is 9–15 (sources 9, 10).
No evidence points to a sharp decline in 2026; the p5 of the arrival model implies this is a ~5% tail scenario; actual reporting at ILR tends to occur within weeks for most stoppages (source 11).
inside_view_v1)Inside View: Causal Chain Analysis
Base mechanism: The ILR tracker records U.S. work stoppages with start dates on a rolling basis. The resolution window is June 9–August 11, 2026 (64 days), measured as a snapshot on August 11.
Key causal chain:
Underlying activity rate: 2025 annual pace = 303 stoppages / 365 days ≈ 0.83/day (source 7). The trend from 2023 peak (471) through 2024 (368) through 2025 (303) is clearly downward. Early 2026 indicators (source 6) suggest comparable to or slightly above 2025, not a rebound to 2022-23 levels. I weight continuity of 2025 pace as the dominant signal.
64-day window gross count: At 0.83/day × 64 days ≈ 53 expected stoppages. With negative-binomial overdispersion (accounts for clustering/burstiness), arrival forecast medians are 44–48 (overdispersion=0.3→0.5 range per sources 9, 10). IQR spans roughly 25–72.
Reporting lag adjustment: The tracker has a nominal 2–4 week lag (source 11). Stoppages starting in late July–early August are systematically underrepresented at the August 11 snapshot. This could reduce the observed count by perhaps 10-20% relative to the true underlying count (roughly 2-3 weeks of activity at 0.83/day = 12-18 stoppages missing). This pushes the modal observed count down toward 35-50.
Seasonal effect: June–November is the high season (source 20, ~12pp higher probability than winter/spring). This supports above-average activity in the window, partially offsetting any secular decline.
Scenario weighting by evidence strength:
- High-evidence scenario (35–65 stoppages, “moderate”): The arrival forecast median (44–48) lands squarely in this range. 2025 pace × 64 days ≈ 53 gross, minus lag → observed ~35-50. This is the best-supported scenario. I place ~55% of mass here (bins 2 and 3: 25-50 and 50-75, with more weight on 25-50).
- Medium-evidence scenario (under 35): Reporting lag + potential further 2026 decline. Arrival forecast p25≈25-31. I place ~20% mass here.
- Medium-evidence scenario (65–100+): Rebound to 2022-23 activity or unusually low lag. Arrival p75≈69-72. I place ~18% mass here.
- Low-evidence scenario (under 20 or 100+): Extreme tails. I place ~7% mass here.
Distribution construction:
- below 0 (bin_0): 0% — impossible
- 0-25 (bin_1): ~7% — very low: severe lag + declining activity needed
- 25-50 (bin_2): ~38% — central case, arrival median 44-48, with lag pulling toward lower end
- 50-75 (bin_3): ~30% — upper portion of central range; requires near-2025 pace with modest lag
- 75-100 (bin_4): ~14% — would require activity above 2025 pace or lower-than-expected lag
- 100-125 (bin_5): ~6% — would require near-2022-23 rebound
- 125-150 (bin_6): ~3% — tail
- 150-175 (bin_7): ~1%
- 175-200 (bin_8): ~0.5%
- above 200 (bin_9): ~0.5%
The confidence_in_evidence is “medium” which warrants moderate spread. The strong NB arrival projection (high-evidence tag) anchors the 25-75 range as the core, with the lag consideration pulling weight toward the 25-50 bin over the 50-75 bin.
outside_view_v1)Reference class: counts of new-start work stoppages in a 64-day summer window, using Cornell ILR annual totals as the baseline. Base rate: 2025 ILR total = 303 stoppages → ~0.83 stoppages/day → 64-day expectation ≈ 53 stoppages (evidence: 2025 annual total ~303, ~0.83/day; Research Brief facts/sources 1 and 7). Adjustment: -~5–10 stoppages to reflect (a) ILR’s documented 2–4 week nominal reporting lag and documented retroactive additions (which make an Aug 11 snapshot likely to miss some late-July/early-August starts) (sources 8 and 11), and (b) medium-quality evidence and uncertainty about whether 2026 activity will equal 2025 (confidence_in_evidence = medium). Final (outside-view anchoring): center the distribution a bit below the raw base-rate expectation (center ~44–48) and widen it to reflect medium evidence/confidence and overdispersion reported in arrival models (arrival forecasts medians/p5/p75 ranges per sources 9–10). I also follow the brief’s scenario weighting: the “Moderate count (35–65)” scenario was tagged high evidence in the brief, so I place the majority of mass in the 25–75 range (bins 2 and 3) rather than centroid-averaging across all scenarios equally (Research Brief scenarios: moderate-count labeled high). Distribution shape and bin-level justification: - 25–50 (bin_2): 0.53 — largest single mass because arrival-model medians (44–48) and the 2025-based base-rate imply the most probability here (sources 9–10), and reported reporting lag argues the observed tracker snapshot will be somewhat lower than the raw underlying-starts expectation (sources 8, 11). - 50–75 (bin_3): 0.32 — captures the plausible chance of counts near or above the raw base-rate (arrival-model upper IQR/p75 values and early-2026 elevated signals; sources 9,10,6); labeled a medium-evidence tail in the brief. - 0–25 (bin_1): 0.07 — reflects the medium-evidence risk of undercounting on the Aug 11 snapshot and the arrival-model lower tail (p5–p25 around 9–31) (sources 9–10, 8, 11). - 75–100 (bin_4): 0.08 — accounts for the moderate tail risk of a rebound toward 2022–2023 elevated yearly pace (sources 9,10,6) and model p75–p90 mass. - bins >=100 (bin_5–bin_9) and bin_0: 0.0 — essentially negligible on the outside view given historical annual totals and the 64-day window (annual totals make >100 in 64 days extremely unlikely). Calibration notes: I am the outside-view trajectory — I anchored on the historical base rate (2025) and only adjusted for clearly reported, evidence-backed effects (reporting lag and medium evidence uncertainty). Because the brief rated the moderate-count scenario as high-evidence, I concentrated >50% probability mass in the 25–50 bin and additional mass in 50–75, rather than spreading mass evenly across all scenarios. Confidence in evidence: medium (per brief); that is why the distribution is relatively wide across neighboring bins rather than a narrow spike.
contrarian_v2)Adversarial reading
The bulk of the evidence (2025 annual rate of 303 stoppages, arrival-process medians of 44–48, summer seasonal elevation, and high-evidence moderate scenario) pulls toward an observed count in the low-to-mid 40s once the 2–4 week reporting lag at the August 11 snapshot is applied.
Counter-evidence found (or absence)
The brief explicitly flags substantial reporting lag (retroactive additions of 9 extra 2024 stoppages; nominal 2–4 week delay) and the clear 2023–2025 downward trend (471→368→303), both of which systematically reduce the August 11 snapshot count below the underlying arrival rate. No evidence contradicts the lag effect or the declining baseline.
Distribution implications
Because the high-evidence scenario is moderate (35–65) but lag pushes the realized snapshot lower, mass is shifted leftward relative to the raw arrival median: 25–50 (bin_2) receives the largest share, 0–25 (bin_1) receives more than a naive median-centered distribution would imply, and bin_3 receives the remainder of the moderate mass while bin_4+ is limited to the upper tail of the high-evidence scenario.
Strategy: logit_mean • Drafts: 3/3
anchoring_v1The draft rationale places the p50 near 44–48 stoppages (the NB arrival-model median), but then applies a lag discount described only qualitatively as “perhaps 10-20%” or “12-18 stoppages missing.” This is a classic anchoring pattern: the adjustment is asymmetric — the arrival median is stated as the reference point and only softly discounted, rather than the lag-adjusted modal expectation being treated as the anchor. Given a documented 2–4 week nominal lag and that the window ends exactly on the snapshot date (August 11), roughly 14–28 days of activity (12–23 stoppages at 0.83/day) are likely unrecorded at snapshot time. A flat-prior interpretation of “the snapshot-date count” would center closer to ~30–40, not 44–48. The draft places 44.7% mass in 25–50 (bin_2) but only ~35% weight below the 44–48 median implied by the arrival model — meaning the effective p50 of the distribution sits near 48–52, higher than the lag-adjusted modal expectation would imply.
ceiling_v1The key structural ceiling here comes from the annual totals: the highest annual count in the ILR tracker’s modern history is 471 (2023). The resolution window is 64 days ≈ 17.5% of a year. Even at the all-time peak pace of 471/365 ≈ 1.29/day × 64 days = ~83 stoppages, hitting bins above 100 would require a daily rate exceeding ~1.56/day — roughly 88% above the 2025 baseline and ~21% above the all-time annual peak. The brief confirms the trend is firmly downward (471→368→303), and early-2026 signals suggest “comparable to or slightly above 2025.” The structural ceiling implied by the historical record therefore puts the plausible upper bound around 80–90, making bins 5–9 (100+) collectively deserving of no more than ~2–3% combined mass. The draft assigns ~2.6% to bins 5–9, which is roughly appropriate, but also assigns ~11.5% to 75–100 (bin_4), which is also likely too generous given the structural ceiling: at 2025 pace with a lag adjustment, the p90 from the arrival model is only 93–104, meaning bin_4 mass should be moderate but bin_5+ should be near-negligible. The reporting lag creates an additional structural floor on the upper tail: stoppages starting in late July and the first 11 days of August (roughly 25–30 days of the 64-day window, or ~39–47% of the period) face the strongest lag risk at the August 11 snapshot, meaning the effective observable window is closer to 40–45 days for the trailing portion, further suppressing realized counts toward the lower bins.
math_v1The draft distribution exhibits a critical internal inconsistency between the stated rationale and the probability mass allocation. The forecaster’s own arrival-process analysis yields medians of 44–48 stoppages with reporting-lag adjustment pushing the observed count toward 35–50, yet 25–50 (bin_2) receives only 44.7% of probability mass while the adjacent 50–75 (bin_3) receives 30.9%. This violates the stated logic: if the arrival median is 44–48 and lag reduces it further, the modal bin should clearly contain the plurality of mass, not be diluted into a two-bin split that nearly equally weights 25–50 and 50–75. The distribution shape implies a median near 45–50, but the mass allocation (44.7% + 30.9% split across the two central bins) systematically over-weights the 50–75 range relative to the stated rationale.
Reviews consumed: 3
All three reviewers converged on the same direction: shift_lower, with the math reviewer flagging the most concrete inconsistency. The math reviewer correctly identified that the draft’s stated rationale (arrival median 44–48, lag pulling toward 35–50, “more weight on 25-50”) is inconsistent with the actual allocation (bin_2=44.7%, bin_3=30.9%) — the bin_2/bin_3 ratio should be more skewed toward bin_2. The anchoring reviewer independently noted the draft anchored on the gross arrival median rather than the lag-adjusted snapshot expectation (~30–38), and the ceiling reviewer reinforced that the ~28 days near the window end suffer the most reporting lag, compressing observable counts downward. I partial-shifted by moving roughly 8 points of mass from bin_3 into bin_2 to honor the stated rationale, and trimmed bin_4 modestly toward bin_2 as well. I kept the upper tails (bins 5–9) close to the draft since the ceiling reviewer judged ~2.6% combined as roughly appropriate. I rejected the anchoring reviewer’s secondary suggestion to add more weight to under 25 (bin_1): the trend-extrapolation argument is speculative and the early-2026 signal (source 6) suggests pace comparable to 2025, not further decline.
[anchoring] Secondary suggestion to place more mass in 0–25 (bin_1) based on linear extrapolation of the 2023→2025 downward trend implying a 2026 rate of ~0.63–0.70/day.
[ceiling] Structural ceiling from historical annual max (471 in 2023) implies bins 5–9 deserve no more than ~2–3% combined.
| Bin | Probability |
|---|---|
| Below 0 (bin_0) | 0.1% |
| 0-25 (bin_1) | 10.0% |
| 25-50 (bin_2) | 52.0% |
| 50-75 (bin_3) | 25.0% |
| 75-100 (bin_4) | 9.0% |
| 100-125 (bin_5) | 2.5% |
| 125-150 (bin_6) | 0.9% |
| 150-175 (bin_7) | 0.3% |
| 175-200 (bin_8) | 0.1% |
| Above 200 (bin_9) | 0.1% |
wooley_rhino • run dd264e66 • $2.75 • 7m9s
The Labor Action Tracker is manually curated and constrained by persistent reporting lag and seasonal clustering.
Summer starts consistently absorb a stable share of annual stoppages, concentrating structural risk into a narrow calendar window.
Year‑to‑date starts in 2026 are softer than the recent five‑year mean, reducing raw expected additions in the June–August window.
Outstanding contract expirations and authorization deadlines in early June create upward pressure for new stoppages to emerge before or during the snapshot.
Given the same‑calendar historical undercount for recent events and the soft 2026 pace, the snapshot expectation centers near 45 listed starts.
A late cluster of multi‑site actions would push the listed count above that center, while typical lag behavior would leave many late‑July and early‑August starts unlisted.
Primary uncertainties are the magnitude of same‑date undercount for early‑August events and the frequency of late‑reported local actions.
Sensitivity to a handful of backfilled entries and rare lockouts means the snapshot can plausibly swing by several events in either direction.
Forecast (Q415): median = 67.2048 work stoppages
How many work stoppages will the Cornell ILR Labor Action Tracker record as beginning between 9 June and 11 August 2026, as listed on the tracker on 11 August 2026?
Key figures
Historical context
Tailwinds
Headwinds
Detailed reasoning
Key uncertainties
Conclusion