thesisPrototype

Government data

US continued claims, week ending Aug. 15, 2026

What will U.S. Department of Labor ETA continued claims, seasonally adjusted, be for the week ending 2026-08-15 on the first print, resolved from ALFRED/FRED series CCSA advance vintage and expressed in millions?

Forecast

1.8M
1.8M–1.9M
80% prediction interval
gpt-5.5Run record ↗

Cumulative probability

Chance that the outcome is at or below each value.

0%25%50%75%100%1.7M1.8M1.9MForecast value

Shaded band: 80% interval (1.8M–1.9M). Dashed line: point estimate (1.8M).

Derived from the point estimate and 80% interval; the agent did not report a full distribution.

Observed outcomeinside 80% interval
actual
1.8M
forecast
1.8M with 80% interval [1.8M, 1.9M]
error
-0M · absolute 0M
cdf score
CRPS 0.015 · PIT 0.31
source
dol_eta Unemployment Insurance Weekly Claims (advance)

DOL ETA UI Weekly Claims news release, advance seasonally adjusted figure for the week ending 2026-08-15, read from FRED CCSA (advance vintage) as the cell's resolver names.

Analysis

§Resolution framing

Target is registered as continued-claims-week-2026-08-15 with unit millions and dataPointId dol.eta.continued_claims.sa.week_2026-08-15.first_print. The registered resolver uses ALFRED graph CSV series CCSA with factor 1e-6. The public specs endpoint requested by the harness returned HTTP 404 in this run, so I could not independently confirm catalog uniqueness there; I kept the canonical ledger slug.

▸ Reported tool use: date.utcmodel report
date -u +%Y-%m-%dT%H:%M:%SZ
↳ 2026-08-12T21:25:18Z
▸ Reported tool use: alfred.fetchmodel report
curl -L -sS 'https://alfred.stlouisfed.org/graph/alfredgraph.csv?id=CCSA' | tail -n 25
↳ Fetched ALFRED CCSA header observation_date,CCSA_20260812. Last 24 values in millions included 2026-02-14=1.827, 2026-02-21=1.871, 2026-03-28=1.787, 2026-04-25=1.758, 2026-05-30=1.786, 2026-06-27=1.821, 2026-07-04=1.798, 2026-07-11=1.789, 2026-07-18=1.777, 2026-07-25=1.801.
▸ Reported tool use: dol.claims_reportmodel report
curl -L -sS -X POST 'https://oui.doleta.gov/unemploy/wkclaims/report.asp' -d 'level=us&strtdate=2026&enddate=2026&filetype=html' | sed -n '446,558p'
↳ DOL national claims table labels Continued Claims S.A. and showed 2026-06-13 S.A.=1,812,000, 2026-06-20 S.A.=1,806,000, 2026-06-27 S.A.=1,821,000, 2026-07-04 S.A.=1,798,000, and 2026-07-11 S.A.=1,789,000; later rows in that table excerpt carried blank cells for 2026-07-18, 2026-07-25, and 2026-08-01, so the registered ALFRED CCSA CSV is the operative history source.
▸ Reported tool use: dol.release_timingmodel report
curl -L -sS 'https://www.dol.gov/newsroom/releases/opa/opa20200701' and curl -L -sS 'https://www.dol.gov/newsroom/releases/eta?date=2026&page=0'
↳ DOL states the ETA Unemployment Insurance Weekly Claims release occurs each Thursday at 8:30 a.m. The DOL ETA release index showed Unemployment Insurance Weekly Claims Report dated August 6, 2026 for week ending August 1 with initial claims 199,000, prior revised from 197,000 to 198,000, and 4-week moving average 198,750; the prior report was July 30, 2026 for week ending July 25 with initial claims 197,000.
▸ Reported tool use: specs.checkmodel report
curl -L -sS -o /tmp/specs_probe -w '%{http_code} %{size_download}\n' 'https://app.thesisinstitute.org/specs.json'
↳ Slug check fetch returned HTTP 404 with 11289 bytes from https://app.thesisinstitute.org/specs.json; fallback probes https://thesisinstitute.org/specs.json and https://app.thesisinstitute.org/api/specs.json also returned 404 with 79 and 13554 bytes respectively.
▸ Reported tool use: model_candidates.localmodel report
Python over fetched ALFRED CCSA CSV, last 24 observations, transformed to millions
↳ thesis_model_candidate_v1 candidates: persistence point=1.801, p10=1.752, p50=1.801, p90=1.850, interval80=[1.752,1.850], interval90=[1.738,1.864], calibration_n=23, train_cutoff=2026-07-25, interval_method=horizon-scaled weekly-change residual; last24_mean point=1.803, p10=1.754, p90=1.852; four_week_drift_to_target point=1.786, p10=1.737, p90=1.835. Last-24 level distribution mean=1.803, median=1.800, min=1.758, max=1.871; weekly changes n=23, mean=-0.001, one-week sigma=0.022, three-week horizon sigma=sqrt(3)*0.022=0.038, 1.28*horizon sigma=0.049.

Base rate / reference class: last 24 weekly CCSA observations from ALFRED, transformed to millions, have mean 1.803, median 1.800, min 1.758, max 1.871. The strongest simple prior is last-print persistence at 1.801m because weekly changes are close to zero on average and the latest level is essentially the same as the 24-week center.

Prior/update/interval: prior = persistence candidate 1.801m from ALFRED CCSA 2026-07-25. Historical sample = 24 weekly values from 2026-02-14 through 2026-07-25, with successive changes used for volatility. Forecast horizon = three weekly observation steps from 2026-07-25 to 2026-08-15, so I scale one-week volatility by sqrt(3). Adjustment components: recent four-week drift is -0.005m per week and latest initial claims are low at 199k, but the implied drift candidate 1.786m is a material move from persistence without enough direct continued-claims evidence for week ending 2026-08-15, so weight persistence 90% and drift signal 10%, rounded back to 1.801m at published precision. Interval method = horizon-scaled weekly-change residual: one-week sigma is 0.022m, horizon sigma = sqrt(3)*0.022 = 0.038m, so sigma = 0.038m for the target horizon. Half-width = 1.28*sigma = 1.28*0.038 = 0.049m, so 80% interval = 1.801 +/- 0.049 = [1.752, 1.850]m.

Counter-consideration: upside risk and outside the interval above 1.850m would come from a sharp lengthening in benefit duration or a jump in layoffs not yet visible in the July 25 continued-claims print. Downside risk and outside the interval below 1.752m would come from the low recent initial-claims level flowing through faster than usual, with continued claims falling by more than the recent weekly-change sigma for multiple weeks.

Review disposition: accepted the blocking interval critique by treating the target as three weekly observation steps beyond the latest fetched CCSA print and using horizon-scaled volatility instead of a one-week half-width. Also clarified the expected Thursday first-print date and removed ambiguous first-print/current wording from historical labels.

Key drivers

  • last-print persistence at 1.801m
  • last-24 level mean 1.803m
  • three-week horizon sigma 0.038m
  • recent drift slightly down but not decisive
  • latest initial claims still low at 199k

Tool evidence

This run has no captured tool responses. “Reported tool use” in the analysis is the model’s account, not an archived tool response.

Sources and resolution

Official source
U.S. Department of Labor ETA, Unemployment Insurance Weekly Claims / ALFRED CCSA advance vintage
Resolved
August 27, 2026
Resolution rule
Resolve from ALFRED graph CSV series CCSA for observation_date 2026-08-15, using the advance vintage first available for the DOL ETA Unemployment Insurance Weekly Claims first print expected on Thursday 2026-08-27. Transform by multiplying CCSA by 1e-6 to millions. Round to the nearest 0.001 million. Later revisions are ignored.

Forecast history

Select a version to read its estimate and analysis.

Forecast versions, estimates, and intervals
VersionDateEstimate80% intervalCRPS
gpt-5.5 · selectedAug 12, 20261.8M1.8M–1.9M0.015
persistence.last_printAug 12, 20261.8MPoint projection0.016
Run details

The analysis is the model’s written report. Tool-use descriptions in that report are model claims; the activity artifacts contain the execution record.

thesis.analyst · gpt-5.5 · full · v2.5.9

pre-submit review · completed

Draft is publishable except the interval method appears to use one-week volatility for a multi-week-ahead target, making the 80% interval under-explained and likely too narrow.

  • blocking interval: The interval half-width is 1.28 times one-week change sigma, but the forecast is for 2026-08-15 from a latest observed 2026-07-25 print, roughly three weekly steps ahead.
  • warning prior_update_interval: The compact step names the prior and interval formula, but its interval sample/horizon is ambiguous because it uses successive weekly changes while forecasting several weeks forward.
  • info optional_suggestion: Clarify that the 2026-08-27 resolution date is the expected DOL Thursday first print for continued claims week ending 2026-08-15.
  • info optional_suggestion: Consider removing 'first-print/current' from historicalContext labels unless those exact values were verified as first vintage values rather than the current ALFRED vintage as of 2026-08-12.

disposition accepted: Review disposition: accepted the blocking interval critique by treating the target as three weekly observation steps beyond the latest fetched CCSA print and using horizon-scaled volatility instead of a one-week half-width. Also clarified the expected Thursday first-print date and removed ambiguous first-print/current wording from historical labels.

disposition accepted: Review disposition: accepted the blocking interval critique by treating the target as three weekly observation steps beyond the latest fetched CCSA print and using horizon-scaled volatility instead of a one-week half-width. Also clarified the expected Thursday first-print date and removed ambiguous first-print/current wording from historical labels.

disposition not applicable: Review disposition: accepted the blocking interval critique by treating the target as three weekly observation steps beyond the latest fetched CCSA print and using horizon-scaled volatility instead of a one-week half-width. Also clarified the expected Thursday first-print date and removed ambiguous first-print/current wording from historical labels.

disposition not applicable: Review disposition: accepted the blocking interval critique by treating the target as three weekly observation steps beyond the latest fetched CCSA print and using horizon-scaled volatility instead of a one-week half-width. Also clarified the expected Thursday first-print date and removed ambiguous first-print/current wording from historical labels.

Activity artifacts

Complete original trace

§Resolution framing

Target is registered as continued-claims-week-2026-08-15 with unit millions and dataPointId dol.eta.continued_claims.sa.week_2026-08-15.first_print. The registered resolver uses ALFRED graph CSV series CCSA with factor 1e-6. The public specs endpoint requested by the harness returned HTTP 404 in this run, so I could not independently confirm catalog uniqueness there; I kept the canonical ledger slug.

▸ Reported tool use: date.utcmodel report
date -u +%Y-%m-%dT%H:%M:%SZ
↳ 2026-08-12T21:25:18Z
▸ Reported tool use: alfred.fetchmodel report
curl -L -sS 'https://alfred.stlouisfed.org/graph/alfredgraph.csv?id=CCSA' | tail -n 25
↳ Fetched ALFRED CCSA header observation_date,CCSA_20260812. Last 24 values in millions included 2026-02-14=1.827, 2026-02-21=1.871, 2026-03-28=1.787, 2026-04-25=1.758, 2026-05-30=1.786, 2026-06-27=1.821, 2026-07-04=1.798, 2026-07-11=1.789, 2026-07-18=1.777, 2026-07-25=1.801.
▸ Reported tool use: dol.claims_reportmodel report
curl -L -sS -X POST 'https://oui.doleta.gov/unemploy/wkclaims/report.asp' -d 'level=us&strtdate=2026&enddate=2026&filetype=html' | sed -n '446,558p'
↳ DOL national claims table labels Continued Claims S.A. and showed 2026-06-13 S.A.=1,812,000, 2026-06-20 S.A.=1,806,000, 2026-06-27 S.A.=1,821,000, 2026-07-04 S.A.=1,798,000, and 2026-07-11 S.A.=1,789,000; later rows in that table excerpt carried blank cells for 2026-07-18, 2026-07-25, and 2026-08-01, so the registered ALFRED CCSA CSV is the operative history source.
▸ Reported tool use: dol.release_timingmodel report
curl -L -sS 'https://www.dol.gov/newsroom/releases/opa/opa20200701' and curl -L -sS 'https://www.dol.gov/newsroom/releases/eta?date=2026&page=0'
↳ DOL states the ETA Unemployment Insurance Weekly Claims release occurs each Thursday at 8:30 a.m. The DOL ETA release index showed Unemployment Insurance Weekly Claims Report dated August 6, 2026 for week ending August 1 with initial claims 199,000, prior revised from 197,000 to 198,000, and 4-week moving average 198,750; the prior report was July 30, 2026 for week ending July 25 with initial claims 197,000.
▸ Reported tool use: specs.checkmodel report
curl -L -sS -o /tmp/specs_probe -w '%{http_code} %{size_download}\n' 'https://app.thesisinstitute.org/specs.json'
↳ Slug check fetch returned HTTP 404 with 11289 bytes from https://app.thesisinstitute.org/specs.json; fallback probes https://thesisinstitute.org/specs.json and https://app.thesisinstitute.org/api/specs.json also returned 404 with 79 and 13554 bytes respectively.
▸ Reported tool use: model_candidates.localmodel report
Python over fetched ALFRED CCSA CSV, last 24 observations, transformed to millions
↳ thesis_model_candidate_v1 candidates: persistence point=1.801, p10=1.752, p50=1.801, p90=1.850, interval80=[1.752,1.850], interval90=[1.738,1.864], calibration_n=23, train_cutoff=2026-07-25, interval_method=horizon-scaled weekly-change residual; last24_mean point=1.803, p10=1.754, p90=1.852; four_week_drift_to_target point=1.786, p10=1.737, p90=1.835. Last-24 level distribution mean=1.803, median=1.800, min=1.758, max=1.871; weekly changes n=23, mean=-0.001, one-week sigma=0.022, three-week horizon sigma=sqrt(3)*0.022=0.038, 1.28*horizon sigma=0.049.

Base rate / reference class: last 24 weekly CCSA observations from ALFRED, transformed to millions, have mean 1.803, median 1.800, min 1.758, max 1.871. The strongest simple prior is last-print persistence at 1.801m because weekly changes are close to zero on average and the latest level is essentially the same as the 24-week center.

Prior/update/interval: prior = persistence candidate 1.801m from ALFRED CCSA 2026-07-25. Historical sample = 24 weekly values from 2026-02-14 through 2026-07-25, with successive changes used for volatility. Forecast horizon = three weekly observation steps from 2026-07-25 to 2026-08-15, so I scale one-week volatility by sqrt(3). Adjustment components: recent four-week drift is -0.005m per week and latest initial claims are low at 199k, but the implied drift candidate 1.786m is a material move from persistence without enough direct continued-claims evidence for week ending 2026-08-15, so weight persistence 90% and drift signal 10%, rounded back to 1.801m at published precision. Interval method = horizon-scaled weekly-change residual: one-week sigma is 0.022m, horizon sigma = sqrt(3)*0.022 = 0.038m, so sigma = 0.038m for the target horizon. Half-width = 1.28*sigma = 1.28*0.038 = 0.049m, so 80% interval = 1.801 +/- 0.049 = [1.752, 1.850]m.

Counter-consideration: upside risk and outside the interval above 1.850m would come from a sharp lengthening in benefit duration or a jump in layoffs not yet visible in the July 25 continued-claims print. Downside risk and outside the interval below 1.752m would come from the low recent initial-claims level flowing through faster than usual, with continued claims falling by more than the recent weekly-change sigma for multiple weeks.

Review disposition: accepted the blocking interval critique by treating the target as three weekly observation steps beyond the latest fetched CCSA print and using horizon-scaled volatility instead of a one-week half-width. Also clarified the expected Thursday first-print date and removed ambiguous first-print/current wording from historical labels.

calibrated forecast · 80% CI
1.8M[1.8M · 1.9M]
Target metadata

Data point: dol.eta.continued_claims.sa.week_2026-08-15.first_print

More government data forecasts