Karthikeyan, Madumita, Bid, Umang Rajnikant, Pan, Rong · Oct 28, 2026
FACTS-MAS: Its authors test which evidence channels help forecasting
Markdown · JSON · Original source
The authors of FACTS-MAS argue that forecasting from mixed evidence should not require every input to contribute. Their multi-agent framework separates evidence into channels for temporal dynamics, exogenous context, seasonality, structural information and discrete events. Each agent makes a forecast that can be tested independently; historical validation estimates its predictive skill against a naive reference, and those estimates determine how forecasts are combined. Evidence that does not help can be removed or given less weight. The authors evaluate the approach on weekly housing inventory and unemployment claims across U.S. metropolitan areas and forecasting horizons. They report that FACTS-MAS outperforms a naive last-value forecast and TimesFM, but that factor-adjusted exponential smoothing performs as well as the combined forecast. They therefore present the contribution not as a universally strongest model, but as a protocol for testing and recomposing evidence before choosing a model for a particular domain. The abstract describes results on two forecasting applications; whether the protocol performs similarly in other domains remains open.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T11:55:28.917Z. First observed: 2026-10-30T06:45:36.347Z. Retrieved: 2026-10-31T12:32:46.403Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The supplied evidence is the abstract, which reports results for weekly housing inventory and unemployment claims; it does not establish performance in other domains. The abstract does not provide numerical performance differences or further detail about the evaluation beyond the stated datasets, locations, and horizons.
Open question
How does FACTS-MAS perform on forecasting applications beyond weekly housing inventory and unemployment claims?
Source-stated
FACTS-MAS separates evidence into specialized channels, with an independently testable forecast from each agent.
decomposes evidence into specialized channels for temporal dynamics, exogenous context, seasonality, structural information, and discrete events. Each agent generates an independently testable forecast
Source-stated
The framework uses historical validation to assess agent forecasts and weight or exclude evidence that does not help.
These skill estimates are used to weight and combine agent forecasts, so that uninformative evidence is either removed or downweighted rather than forced to participate.
Source-stated
The authors evaluate FACTS-MAS on weekly housing inventory and unemployment claims across 15 U.S. metropolitan areas and three forecasting horizons.
We evaluate FACTS-MAS on weekly for-sale housing inventory from Zillow and weekly unemployment insurance claims from FRED, for 15 U.S. metropolitan areas at 4-, 8-, and 13-week forecasting horizons.
Source-stated
The authors report that FACTS-MAS beats the naive last-value forecast and TimesFM, while factor-adjusted exponential smoothing performs as well as the fused forecast.
Our method outperforms both a naive last-value forecast method and the TimesFM foundation model. However, a factor-adjusted exponential smoothing model performs as well as the fused forecast.
Source-stated
The authors frame FACTS-MAS as a protocol for testing and recomposing evidence before selecting a model for an application domain.
The evidence therefore supports FACTS-MAS as a systematic protocol for testing, calibrating, and recomposing evidence channels before assembling them into the strongest single model for a specific application domain.
Source and contributor context
Karthikeyan, Madumita, Bid, Umang Rajnikant, Pan, Rong · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 33d2e872-9846-4166-8e5e-eda7b0e5aa35.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T11:55:28.917Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 263f2def9b0b9a610ad4346f3a19c051a798820ffe1687a0373cf2372f618e21. Source text hash: 2b8237c4daca5dd03dffc21459870adb778c5a47734c93ac4468e6d22b609aa2.
Revision and source-check history
Version 1 · 2026-11-11T11:55:28.917Z
FACTS-MAS: Its authors test which evidence channels help forecasting
The authors of FACTS-MAS argue that forecasting from mixed evidence should not require every input to contribute. Their multi-agent framework separates evidence into channels for temporal dynamics, exogenous context, seasonality, structural information and discrete events. Each agent makes a forecast that can be tested independently; historical validation estimates its predictive skill against a naive reference, and those estimates determine how forecasts are combined. Evidence that does not help can be removed or given less weight. The authors evaluate the approach on weekly housing inventory and unemployment claims across U.S. metropolitan areas and forecasting horizons. They report that FACTS-MAS outperforms a naive last-value forecast and TimesFM, but that factor-adjusted exponential smoothing performs as well as the combined forecast. They therefore present the contribution not as a universally strongest model, but as a protocol for testing and recomposing evidence before choosing a model for a particular domain. The abstract describes results on two forecasting applications; whether the protocol performs similarly in other domains remains open.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T11:55:28.917Z. First observed: 2026-10-30T06:45:36.347Z. Retrieved: 2026-10-31T12:32:46.403Z.
Limits
The supplied evidence is the abstract, which reports results for weekly housing inventory and unemployment claims; it does not establish performance in other domains. The abstract does not provide numerical performance differences or further detail about the evaluation beyond the stated datasets, locations, and horizons.
Open question
How does FACTS-MAS perform on forecasting applications beyond weekly housing inventory and unemployment claims?
Source-stated
FACTS-MAS separates evidence into specialized channels, with an independently testable forecast from each agent.
decomposes evidence into specialized channels for temporal dynamics, exogenous context, seasonality, structural information, and discrete events. Each agent generates an independently testable forecast
Source-stated
The framework uses historical validation to assess agent forecasts and weight or exclude evidence that does not help.
These skill estimates are used to weight and combine agent forecasts, so that uninformative evidence is either removed or downweighted rather than forced to participate.
Source-stated
The authors evaluate FACTS-MAS on weekly housing inventory and unemployment claims across 15 U.S. metropolitan areas and three forecasting horizons.
We evaluate FACTS-MAS on weekly for-sale housing inventory from Zillow and weekly unemployment insurance claims from FRED, for 15 U.S. metropolitan areas at 4-, 8-, and 13-week forecasting horizons.
Source-stated
The authors report that FACTS-MAS beats the naive last-value forecast and TimesFM, while factor-adjusted exponential smoothing performs as well as the fused forecast.
Our method outperforms both a naive last-value forecast method and the TimesFM foundation model. However, a factor-adjusted exponential smoothing model performs as well as the fused forecast.
Source-stated
The authors frame FACTS-MAS as a protocol for testing and recomposing evidence before selecting a model for an application domain.
The evidence therefore supports FACTS-MAS as a systematic protocol for testing, calibrating, and recomposing evidence channels before assembling them into the strongest single model for a specific application domain.
Source and contributor context
Karthikeyan, Madumita, Bid, Umang Rajnikant, Pan, Rong · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 33d2e872-9846-4166-8e5e-eda7b0e5aa35.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T11:55:28.917Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 263f2def9b0b9a610ad4346f3a19c051a798820ffe1687a0373cf2372f618e21. Source text hash: 2b8237c4daca5dd03dffc21459870adb778c5a47734c93ac4468e6d22b609aa2.
uncertain · 2026-11-11T11:55:28.917Z · anthropic/claude-opus-5.5
Publication: 33d2e872-9846-4166-8e5e-eda7b0e5aa35
attribution — pass: The reading consistently credits the FACTS-MAS design, results and framing to the paper's authors (Karthikeyan, Bid, Pan) with verbs such as 'argue', 'report' and 'present'. It does not present the performance comparisons as established fact or as Escape Meridian's view.
Authors:Madumita Karthikeyan, Umang Rajnikant Bid, Rong Pan
summary — uncertain: The summary is faithful and keeps the authors' notable concession. It drops the source's descriptor of TimesFM as a 'foundation model', so a non-specialist cannot tell why beating it matters. 'Factor-adjusted exponential smoothing', 'naive reference' and 'exogenous context' are central but unexplained, so tooltip coverage should be confirmed. Smallest repair: write 'the TimesFM foundation model' and add tooltips for those terms. Minor: 'before choosing a model' loosely renders 'before assembling them into the strongest single model'.
Our method outperforms both a naive last-value forecast method and the TimesFM foundation model.
date — pass: originalPublished 2026-10-28 matches the arXiv v1 submission date. The later feed date of Oct 31 is correctly not used.
[Submitted on 28 Oct 2026]
support — pass: Each claim is quoted verbatim from the abstract. The limitations accurately note that only two applications are covered and that no numerical margins are given. The question about other domains follows directly from the evidence without being leading. No unsupported benefits or costs are implied.
However, a factor-adjusted exponential smoothing model performs as well as the fused forecast.
media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most direct way to reach the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv