thesisPrototype

Government data

US initial claims, week ending Sep 5

What will the U.S. Department of Labor report as the advance seasonally adjusted initial claims figure for the week ending September 5, 2026, in the first official print?

Forecast

206k
188k–224k
80% prediction interval
gpt-5.5Run record ↗

Cumulative probability

Chance that the outcome is at or below each value.

0%25%50%75%100%161k206k251kForecast value

Shaded band: 80% interval (188k–224k). Dashed line: point estimate (206k).

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

Observed outcomeinside 80% interval
actual
206k
forecast
206k with 80% interval [188k, 224k]
error
0k · absolute 0k
cdf score
CRPS 3.90 · PIT 0.50
source
dol_eta Unemployment Insurance Weekly Claims (advance)

DOL ETA UI Weekly Claims news release, advance seasonally adjusted figure for the week ending 2026-09-05, read from FRED ICSA (advance vintage) as the cell's resolver names.

Analysis

§Forecast for seasonally adjusted initial claims

Resolver framing: the target is DOL seasonally adjusted initial claims, series ICSA-equivalent, for the week ending 2026-09-05, reported as the advance first print in thousands. The registered resolver is ALFRED graph CSV for ICSA with advance_vintage policy and factor 0.001, so the JSON resolver fields follow that contract. DOL's archive page says the UI Weekly Claims News Release is published each Thursday at 8:30 AM Eastern except listed federal-holiday exceptions; the FRED release calendar lists weekly claims releases on 2026-09-10 and 2026-09-17. I believe the registered expectedReleaseWindow ending 2026-09-20 may be one week late for this stated week-ending target, but I keep the catalog slug, dataPointId, resolver source, and resolutionDate tied to the registered target.

▸ Reported tool use: official.lookupmodel report
Read the DOL ETA newsroom listing and latest weekly-claims release text for the most recent print.
↳ Fetched DOL values: week ending 2026-08-22 seasonally adjusted initial claims = 203 thousand; prior week revised level = 207 thousand; 4-week moving average = 205.5 thousand; continuing claims for week ending 2026-08-15 = 1778 thousand.
▸ Reported tool use: official.lookupmodel report
Read the DOL ETA newsroom page for the three preceding weekly claims reports.
↳ Fetched DOL advance seasonally adjusted initial claims: week ending 2026-08-15 = 206 thousand; week ending 2026-08-08 = 209 thousand; week ending 2026-08-01 = 199 thousand; corresponding 4-week averages were 204.0, 199.0, and 198.75 thousand.
▸ Reported tool use: official.lookupmodel report
Read older DOL ETA newsroom entries for the late-June and July reference class.
↳ Fetched DOL advance seasonally adjusted initial claims: week ending 2026-07-25 = 197 thousand; 2026-07-18 = 187 thousand; 2026-07-11 = 208 thousand; 2026-07-04 = 215 thousand; 2026-06-27 = 215 thousand.
▸ Reported tool use: calendar.lookupmodel report
Checked public release-calendar context for the target's publication timing and registered resolve-by bound.
↳ Fetched calendar context: weekly claims releases listed on 2026-09-10 and 2026-09-17; registered sourceBinding expectedReleaseWindow start = 2026-09-16 and end = 2026-09-20; final JSON resolutionDate uses the registered bound 2026-09-20.

Base rate / reference class: the recent official-source reference class has nine consecutive advance prints from 187 to 215 thousand, with the latest at 203 thousand and the latest four-week mean at 204.25 thousand from the fetched Aug 1-Aug 22 values. This argues for a low-200s persistence prior rather than a return to the 220s unless fresh layoff evidence appears.

Prior/update/interval: persistence prior = latest four-week average 204.25 thousand, using the official DOL advance prints from week ending 2026-06-27 through 2026-08-22. Adjustment components: level +0.5 thousand because the latest print, 203, is slightly below the recent mean; momentum +0.5 thousand because the 4-week average rose from 198.75 to 205.5 over the latest three releases; one-off seasonal-noise adjustment +0.5 thousand for late-summer volatility; policy-mechanism adjustment 0 because the latest federal-worker and aggregate claims context does not show a discrete policy shock. Point = 204.25 + 0.5 + 0.5 + 0.5 + 0 = 205.75, rounded to 206 thousand. Successive changes in the fetched level series were 0, -7, -21, +10, +2, +10, -3, -3 thousand; sample one-week sigma = 9.96. For a two-week horizon from the latest released week to week ending 2026-09-05, horizon sigma = sqrt(2)*9.96 = 14.08, and 1.28*sigma = 18.0 thousand, giving an 80% interval of 206 +/- 18 = 188 to 224 thousand.

Level, momentum, one-off, and policy mechanisms: the level is historically low but stable, momentum is a mild drift up from the July trough, one-off risk is mostly seasonal-adjustment noise around late summer, and I do not see a policy mechanism large enough to shift claims materially before the target week. Upside risk would be a sudden layoff burst, a strike-related filing wave, or a seasonal factor miss that would land above the interval. Downside risk would be another July-like low claims print near 187 thousand or below. Outside the interval means below 188 thousand or above 224 thousand in the advance SA first print.

Review disposition: accepted the blocking resolver critiques by changing resolutionDate to 2026-09-20 and the resolver fields to the registered ALFRED ICSA advance-vintage source binding; retained the analytical point estimate and interval because the critique did not identify an evidence or calibration error; retained the date discrepancy only as reasoning context.

Key drivers

  • Recent claims near 200 thousand
  • Four-week average around 205.5 thousand
  • Late-summer seasonal adjustment noise
  • No current layoff-break signal
  • Two-week forecast horizon from latest print

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, ICSA advance vintage
Resolved
September 29, 2026
Resolution rule
Resolve using the registered ALFRED graph CSV source binding for sourceSeriesId ICSA, field ICSA, releasePolicy advance_vintage, table ALFRED graph CSV, transformed by multiplying the published ICSA value by 0.001 to express claims in thousands. Use the first advance vintage for the observation dated 2026-09-05 available by the registered resolve-by bound of 2026-09-20. Ignore later ALFRED/FRED vintage changes and subsequent DOL revisions.

Forecast history

Select a version to read its estimate and analysis.

Forecast versions, estimates, and intervals
VersionDateEstimate80% intervalCRPS
gpt-5.5 · selectedAug 29, 2026206k188k–224k3.90
persistence.last_printAug 29, 2026206k195k–217k2.30
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 · fast · v2.5.11

pre-submit review · completed

Draft is analytically usable but not publication-ready because its resolver source and resolutionDate conflict with the canonical ledger target contract.

  • blocking resolver: The cell uses resolutionDate 2026-09-10 and a DOL data.pdf resolver, but the ledger contract requires resolutionDate 2026-09-20 and the ALFRED/FRED advance-vintage source binding at alfred.stlouisfed.org for ICSA.
  • blocking coherence: The reasoning says it is keeping the registered target while the JSON fields override the registered resolutionDate and source.
  • info optional_suggestion: The prior, update, interval math, tails, point estimate, and unit are otherwise coherent.
  • info optional_suggestion: Consider naming ALFRED advance vintage explicitly in the final forecast step so the first-print rule is unambiguous.

disposition accepted: Review disposition: accepted the blocking resolver critiques by changing resolutionDate to 2026-09-20 and the resolver fields to the registered ALFRED ICSA advance-vintage source binding; retained the analytical point estimate and interval because the critique did not identify an evidence or calibration error; retained the date discrepancy only as reasoning context.

disposition accepted: Review disposition: accepted the blocking resolver critiques by changing resolutionDate to 2026-09-20 and the resolver fields to the registered ALFRED ICSA advance-vintage source binding; retained the analytical point estimate and interval because the critique did not identify an evidence or calibration error; retained the date discrepancy only as reasoning context.

disposition not applicable: Review disposition: accepted the blocking resolver critiques by changing resolutionDate to 2026-09-20 and the resolver fields to the registered ALFRED ICSA advance-vintage source binding; retained the analytical point estimate and interval because the critique did not identify an evidence or calibration error; retained the date discrepancy only as reasoning context.

disposition not applicable: Review disposition: accepted the blocking resolver critiques by changing resolutionDate to 2026-09-20 and the resolver fields to the registered ALFRED ICSA advance-vintage source binding; retained the analytical point estimate and interval because the critique did not identify an evidence or calibration error; retained the date discrepancy only as reasoning context.

Activity artifacts

Complete original trace

§Forecast for seasonally adjusted initial claims

Resolver framing: the target is DOL seasonally adjusted initial claims, series ICSA-equivalent, for the week ending 2026-09-05, reported as the advance first print in thousands. The registered resolver is ALFRED graph CSV for ICSA with advance_vintage policy and factor 0.001, so the JSON resolver fields follow that contract. DOL's archive page says the UI Weekly Claims News Release is published each Thursday at 8:30 AM Eastern except listed federal-holiday exceptions; the FRED release calendar lists weekly claims releases on 2026-09-10 and 2026-09-17. I believe the registered expectedReleaseWindow ending 2026-09-20 may be one week late for this stated week-ending target, but I keep the catalog slug, dataPointId, resolver source, and resolutionDate tied to the registered target.

▸ Reported tool use: official.lookupmodel report
Read the DOL ETA newsroom listing and latest weekly-claims release text for the most recent print.
↳ Fetched DOL values: week ending 2026-08-22 seasonally adjusted initial claims = 203 thousand; prior week revised level = 207 thousand; 4-week moving average = 205.5 thousand; continuing claims for week ending 2026-08-15 = 1778 thousand.
▸ Reported tool use: official.lookupmodel report
Read the DOL ETA newsroom page for the three preceding weekly claims reports.
↳ Fetched DOL advance seasonally adjusted initial claims: week ending 2026-08-15 = 206 thousand; week ending 2026-08-08 = 209 thousand; week ending 2026-08-01 = 199 thousand; corresponding 4-week averages were 204.0, 199.0, and 198.75 thousand.
▸ Reported tool use: official.lookupmodel report
Read older DOL ETA newsroom entries for the late-June and July reference class.
↳ Fetched DOL advance seasonally adjusted initial claims: week ending 2026-07-25 = 197 thousand; 2026-07-18 = 187 thousand; 2026-07-11 = 208 thousand; 2026-07-04 = 215 thousand; 2026-06-27 = 215 thousand.
▸ Reported tool use: calendar.lookupmodel report
Checked public release-calendar context for the target's publication timing and registered resolve-by bound.
↳ Fetched calendar context: weekly claims releases listed on 2026-09-10 and 2026-09-17; registered sourceBinding expectedReleaseWindow start = 2026-09-16 and end = 2026-09-20; final JSON resolutionDate uses the registered bound 2026-09-20.

Base rate / reference class: the recent official-source reference class has nine consecutive advance prints from 187 to 215 thousand, with the latest at 203 thousand and the latest four-week mean at 204.25 thousand from the fetched Aug 1-Aug 22 values. This argues for a low-200s persistence prior rather than a return to the 220s unless fresh layoff evidence appears.

Prior/update/interval: persistence prior = latest four-week average 204.25 thousand, using the official DOL advance prints from week ending 2026-06-27 through 2026-08-22. Adjustment components: level +0.5 thousand because the latest print, 203, is slightly below the recent mean; momentum +0.5 thousand because the 4-week average rose from 198.75 to 205.5 over the latest three releases; one-off seasonal-noise adjustment +0.5 thousand for late-summer volatility; policy-mechanism adjustment 0 because the latest federal-worker and aggregate claims context does not show a discrete policy shock. Point = 204.25 + 0.5 + 0.5 + 0.5 + 0 = 205.75, rounded to 206 thousand. Successive changes in the fetched level series were 0, -7, -21, +10, +2, +10, -3, -3 thousand; sample one-week sigma = 9.96. For a two-week horizon from the latest released week to week ending 2026-09-05, horizon sigma = sqrt(2)*9.96 = 14.08, and 1.28*sigma = 18.0 thousand, giving an 80% interval of 206 +/- 18 = 188 to 224 thousand.

Level, momentum, one-off, and policy mechanisms: the level is historically low but stable, momentum is a mild drift up from the July trough, one-off risk is mostly seasonal-adjustment noise around late summer, and I do not see a policy mechanism large enough to shift claims materially before the target week. Upside risk would be a sudden layoff burst, a strike-related filing wave, or a seasonal factor miss that would land above the interval. Downside risk would be another July-like low claims print near 187 thousand or below. Outside the interval means below 188 thousand or above 224 thousand in the advance SA first print.

Review disposition: accepted the blocking resolver critiques by changing resolutionDate to 2026-09-20 and the resolver fields to the registered ALFRED ICSA advance-vintage source binding; retained the analytical point estimate and interval because the critique did not identify an evidence or calibration error; retained the date discrepancy only as reasoning context.

calibrated forecast · 80% CI
206k[188k · 224k]
Target metadata

Data point: us.dol.initial_claims.sa.week_2026-09-05

More government data forecasts