thesisPrototype

Government data

Initial claims, week ending Aug. 15, 2026

What will the U.S. Department of Labor seasonally adjusted initial unemployment insurance claims (FRED/ALFRED series ICSA), in thousands, be for the week ending 2026-08-15 on the first print?

Forecast

199k
183k–215k
80% prediction interval
gpt-5.5Run record ↗

Cumulative probability

Chance that the outcome is at or below each value.

0%25%50%75%100%159k199k239kForecast value

Shaded band: 80% interval (183k–215k). Dashed line: point estimate (199k).

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

Observed outcomeinside 80% interval
actual
206k
forecast
199k with 80% interval [183k, 215k]
error
+7k · absolute 7k
cdf score
CRPS 4.69 · PIT 0.68
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 ICSA (advance vintage) as the cell's resolver names.

Analysis

§Forecast ICSA for week ending 2026-08-15

The target is the registered U.S. DOL/FRED-ALFRED seasonally adjusted initial claims series ICSA for week ending 2026-08-15, first print, transformed from persons to thousands. The local generated ledger target gives slug initial-claims-week-2026-08-15, unit thousands, dataPointId us.dol.initial_claims.sa.week_2026-08-15, sourceBinding releasePolicy advance_vintage, and resolutionDate 2026-08-24. DOL weekly claims normally publish on Thursdays, so 2026-08-20 is the apparent release day inside the registered expectedReleaseWindow, but I keep the ledger target date and state the discrepancy.

▸ Reported tool use: local.target.lookupmodel report
Read records/targets/2026-08-12-ba7909136ce260dfbd43442621dd88fdef9abd8d2b02c81dfcb114251a76b8bd.json and site/src/data/ledger-targets.generated.ts for this target.
↳ Fetched target fields: catalogSlug initial-claims-week-2026-08-15; target unit thousands; dataPointId us.dol.initial_claims.sa.week_2026-08-15; source field ICSA; transform factor 0.001; expectedReleaseWindow start 2026-08-20 end 2026-08-24; generated resolutionDate 2026-08-24; targetContentHash ba7909136ce260dfbd43442621dd88fdef9abd8d2b02c81dfcb114251a76b8bd.
▸ Reported tool use: dol.schedule.fetchmodel report
curl -L -sS -I https://www.dol.gov/ui/data.pdf and curl -L -sS https://oui.doleta.gov/unemploy/claims.asp
↳ DOL UI data PDF returned HTTP 200 with content-length 606807 and last-modified Thu, 06 Aug 2026 12:30:00 GMT; claims.asp returned the DOL Unemployment Insurance Weekly Claims Data page with final_yr value 2027. This supports the DOL weekly claims data source but did not provide a contrary target-specific 2026-08-24 date in the extracted response.
▸ Reported tool use: fred.lookupmodel report
Fetch https://fred.stlouisfed.org/graph/fredgraph.csv?id=ICSA and read the latest rows echoed this run.
↳ Fetched recent ICSA seasonally adjusted claims in thousands: 2026-07-04 217.0, 2026-07-11 209.0, 2026-07-18 189.0, 2026-07-25 198.0, 2026-08-01 199.0. Earlier fetched rows include 2026-06-06 230.0, 2026-06-13 227.0, 2026-06-20 216.0, 2026-06-27 217.0.
▸ Reported tool use: model_candidates.computemodel report
Compute thesis_model_candidate_v1 persistence candidate from last 24 fetched ICSA observations through 2026-08-01.
↳ Model candidate two_week_persistence_with_two_week_change_residual_interval: point 199.0, p10 183.2, p50 198.0, p90 213.6, 80% interval [183, 215], 90% interval [181, 217], calibration_n 22, train_cutoff 2026-08-01, walk-forward MAE 9.73 thousand, RMSE 12.21 thousand, sigma 12.43 thousand, half_width 15.91 thousand.

Base rate / reference class: the last 24 fetched weekly ICSA prints run from 211.0 thousand on 2026-02-21 to 199.0 thousand on 2026-08-01, with range 189.0-230.0 thousand. The latest three prints were 189.0, 198.0, and 199.0 thousand. For this two-week-ahead target, recent two-week changes over the fetched window ranged from -28.0 to +22.0 thousand, with 8 of the last 10 two-week changes within +/-15.9 thousand.

Persistence benchmark: last fetched print = 199.0 thousand for week ending 2026-08-01. Current evidence fetched this run gives no direct release-specific signal for week ending 2026-08-15, so adjustment = 0.0 thousand and point = 199.0 + 0.0 = 199.0 thousand, rounded to 199.

Prior/update/interval: prior = two-week last-print persistence model candidate using 24 fetched ICSA prints through 2026-08-01; adjustment components = 0.0 thousand because no direct current signal cleared the update test; interval method = normal residual interval from fetched two-week changes. Two-week changes were +2, -9, -2, -2, +7, +5, -3, -18, -16, +22, +11, +0, +15, +18, +2, -14, -10, +1, -8, -28, -11, +10 thousand; sigma = 12.43 thousand; half-width = 1.28*sigma = 15.91 thousand; implied 80% bounds = 199.0 +/- 15.91 = [183.09, 214.91], rounded to [183, 215].

Counter-consideration: upside risk would land above the interval if the delayed post-summer adjustment or a state-level layoff shock pushed two-week claims more than about 16 thousand above the latest print; downside risk would land below the interval if the 2026-07-18 low was repeated and claims fell more than about 16 thousand from 199 thousand.

▸ Reported tool use: slug.checkmodel report
curl -i -L -sS https://app.thesisinstitute.org/specs.json
↳ Slug check attempt returned HTTP 404 with content-length 11289 on 2026-08-12; no competing specs.json entry was fetched. The local generated ledger target contains catalogSlug initial-claims-week-2026-08-15 exactly once for this registered target.

Review disposition: accepted the reviewer note to keep the registered ledger date basis concise and to avoid treating the 404 specs.json attempt as evidentiary source context; no required fixes were raised, and the persistence point estimate and 80% interval are unchanged.

Key drivers

  • Latest fetched print was 199 thousand
  • Two-week persistence is strongest simple benchmark
  • Recent two-week changes have sigma 12.43 thousand
  • No direct release-specific signal to move away from persistence

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
ALFRED graph CSV
Resolved
August 24, 2026
Resolution rule
Resolve from the registered ALFRED graph CSV for FRED series ICSA, U.S. initial unemployment insurance claims, seasonally adjusted, week ending 2026-08-15. Use the advance-vintage first print named by the resolver's releasePolicy=advance_vintage, field ICSA in the ALFRED graph CSV, multiply persons by 0.001, and round to the nearest 0.1 thousand. Later revisions do not change resolution. The generated ledger target records resolutionDate 2026-08-24 within expectedReleaseWindow 2026-08-20 through 2026-08-24; DOL weekly claims normally publish on Thursdays, so 2026-08-20 is the apparent calendar release day, but I retain the registered target date rather than changing the contract.

Forecast history

Select a version to read its estimate and analysis.

Forecast versions, estimates, and intervals
VersionDateEstimate80% intervalCRPS
gpt-5.5 · selectedAug 12, 2026199k183k–215k4.69
persistence.last_printAug 12, 2026199k188k–210k4.17
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 publication-ready against the supplied target contract; no blocking rubric failures found.

  • info optional_suggestion: Consider removing app.thesisinstitute.org/specs.json from sourceContext because the run reports it returned 404.
  • info optional_suggestion: The resolution-date discrepancy is handled appropriately, but it may be worth making the ledger-contract basis even shorter in the final rule text.

disposition not applicable: Review disposition: accepted the reviewer note to keep the registered ledger date basis concise and to avoid treating the 404 specs.json attempt as evidentiary source context; no required fixes were raised, and the persistence point estimate and 80% interval are unchanged.

disposition not applicable: Review disposition: accepted the reviewer note to keep the registered ledger date basis concise and to avoid treating the 404 specs.json attempt as evidentiary source context; no required fixes were raised, and the persistence point estimate and 80% interval are unchanged.

Activity artifacts

Complete original trace

§Forecast ICSA for week ending 2026-08-15

The target is the registered U.S. DOL/FRED-ALFRED seasonally adjusted initial claims series ICSA for week ending 2026-08-15, first print, transformed from persons to thousands. The local generated ledger target gives slug initial-claims-week-2026-08-15, unit thousands, dataPointId us.dol.initial_claims.sa.week_2026-08-15, sourceBinding releasePolicy advance_vintage, and resolutionDate 2026-08-24. DOL weekly claims normally publish on Thursdays, so 2026-08-20 is the apparent release day inside the registered expectedReleaseWindow, but I keep the ledger target date and state the discrepancy.

▸ Reported tool use: local.target.lookupmodel report
Read records/targets/2026-08-12-ba7909136ce260dfbd43442621dd88fdef9abd8d2b02c81dfcb114251a76b8bd.json and site/src/data/ledger-targets.generated.ts for this target.
↳ Fetched target fields: catalogSlug initial-claims-week-2026-08-15; target unit thousands; dataPointId us.dol.initial_claims.sa.week_2026-08-15; source field ICSA; transform factor 0.001; expectedReleaseWindow start 2026-08-20 end 2026-08-24; generated resolutionDate 2026-08-24; targetContentHash ba7909136ce260dfbd43442621dd88fdef9abd8d2b02c81dfcb114251a76b8bd.
▸ Reported tool use: dol.schedule.fetchmodel report
curl -L -sS -I https://www.dol.gov/ui/data.pdf and curl -L -sS https://oui.doleta.gov/unemploy/claims.asp
↳ DOL UI data PDF returned HTTP 200 with content-length 606807 and last-modified Thu, 06 Aug 2026 12:30:00 GMT; claims.asp returned the DOL Unemployment Insurance Weekly Claims Data page with final_yr value 2027. This supports the DOL weekly claims data source but did not provide a contrary target-specific 2026-08-24 date in the extracted response.
▸ Reported tool use: fred.lookupmodel report
Fetch https://fred.stlouisfed.org/graph/fredgraph.csv?id=ICSA and read the latest rows echoed this run.
↳ Fetched recent ICSA seasonally adjusted claims in thousands: 2026-07-04 217.0, 2026-07-11 209.0, 2026-07-18 189.0, 2026-07-25 198.0, 2026-08-01 199.0. Earlier fetched rows include 2026-06-06 230.0, 2026-06-13 227.0, 2026-06-20 216.0, 2026-06-27 217.0.
▸ Reported tool use: model_candidates.computemodel report
Compute thesis_model_candidate_v1 persistence candidate from last 24 fetched ICSA observations through 2026-08-01.
↳ Model candidate two_week_persistence_with_two_week_change_residual_interval: point 199.0, p10 183.2, p50 198.0, p90 213.6, 80% interval [183, 215], 90% interval [181, 217], calibration_n 22, train_cutoff 2026-08-01, walk-forward MAE 9.73 thousand, RMSE 12.21 thousand, sigma 12.43 thousand, half_width 15.91 thousand.

Base rate / reference class: the last 24 fetched weekly ICSA prints run from 211.0 thousand on 2026-02-21 to 199.0 thousand on 2026-08-01, with range 189.0-230.0 thousand. The latest three prints were 189.0, 198.0, and 199.0 thousand. For this two-week-ahead target, recent two-week changes over the fetched window ranged from -28.0 to +22.0 thousand, with 8 of the last 10 two-week changes within +/-15.9 thousand.

Persistence benchmark: last fetched print = 199.0 thousand for week ending 2026-08-01. Current evidence fetched this run gives no direct release-specific signal for week ending 2026-08-15, so adjustment = 0.0 thousand and point = 199.0 + 0.0 = 199.0 thousand, rounded to 199.

Prior/update/interval: prior = two-week last-print persistence model candidate using 24 fetched ICSA prints through 2026-08-01; adjustment components = 0.0 thousand because no direct current signal cleared the update test; interval method = normal residual interval from fetched two-week changes. Two-week changes were +2, -9, -2, -2, +7, +5, -3, -18, -16, +22, +11, +0, +15, +18, +2, -14, -10, +1, -8, -28, -11, +10 thousand; sigma = 12.43 thousand; half-width = 1.28*sigma = 15.91 thousand; implied 80% bounds = 199.0 +/- 15.91 = [183.09, 214.91], rounded to [183, 215].

Counter-consideration: upside risk would land above the interval if the delayed post-summer adjustment or a state-level layoff shock pushed two-week claims more than about 16 thousand above the latest print; downside risk would land below the interval if the 2026-07-18 low was repeated and claims fell more than about 16 thousand from 199 thousand.

▸ Reported tool use: slug.checkmodel report
curl -i -L -sS https://app.thesisinstitute.org/specs.json
↳ Slug check attempt returned HTTP 404 with content-length 11289 on 2026-08-12; no competing specs.json entry was fetched. The local generated ledger target contains catalogSlug initial-claims-week-2026-08-15 exactly once for this registered target.

Review disposition: accepted the reviewer note to keep the registered ledger date basis concise and to avoid treating the 404 specs.json attempt as evidentiary source context; no required fixes were raised, and the persistence point estimate and 80% interval are unchanged.

calibrated forecast · 80% CI
199k[183k · 215k]
Target metadata

Data point: us.dol.initial_claims.sa.week_2026-08-15

More government data forecasts