Rhodes, James, Kang, George · Nov 9, 2026
Rhodes and Kang argue work products should carry their context
Markdown · JSON · Original source
James Rhodes and George Kang argue that a document can preserve a professional work product’s visible result while losing the research, calculations, assumptions, evidence and authorization boundaries behind it. They propose delivering that context in a Portable Agent Format (PAF) package: the work product travels with compiled context, evidence, provenance and rules for later interaction. In a four-package evaluation, a hosted governed reader succeeded on 96% of 400 frozen questions, compared with 86% for instructions plus raw-file retrieval. Results varied by package, and the authors report a specific failure: offline retrieval resolved none of 40 historical-record questions in one financial-data package. They also report that the offline reader’s -1 response rate, combining materially wrong and unauthorized answers, was 2.9%, compared with 56% for the source and instructions alone. The authors say the findings support treating compiled knowledge as part of the delivered work product. But the evaluation does not establish independent standards interoperability or end-to-end governance conformance, and its ablations do not separate representation effects from training effects.
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: Nov 9, 2026 (2026-11-09).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T17:49:04.791Z. First observed: 2026-11-11T05:06:21.604Z. Retrieved: 2026-11-11T17:48:14.471Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
Some references could not be checked. Reload this page to try again.
Limits
The abstract reports results from four packages and a held-out evaluation; it does not establish that the findings generalize to other work products or settings. The authors say the ablations do not isolate representation effects from training effects. The abstract says independent standards interoperability and end-to-end governance conformance remain unproven.
Open question
What evidence would establish independent standards interoperability and end-to-end governance conformance for PAF packages?
Source-stated
Rhodes and Kang argue that documents can omit important context behind professional work, and propose delivering the work product with compiled context, evidence, provenance and interaction rules in a Portable Agent Format package.
A document preserves the visible result of professional work but often discards the research, calculations, assumptions, evidence, and authorization boundaries behind it. We study a different unit of delivery: a Portable Agent Format (PAF) package
Source-stated
In the reported evaluation, the hosted governed reader succeeded on 96% of questions, compared with 86% for instructions plus raw-file retrieval.
Across 400 frozen questions and 12,000 evaluated responses, a hosted governed reader succeeds on 96% of questions versus 86% with instructions plus raw-file retrieval
Source-stated
The authors report a failure in which offline retrieval resolved none of the historical-record questions in one financial-data package.
The same run exposes a failure: offline retrieval resolves none of the 40 historical-record questions in one financial-data package.
Source-stated
The authors say the evaluation leaves independent standards interoperability and end-to-end governance conformance unproven.
while leaving independent standards interoperability and end-to-end governance conformance unproven.
Source and contributor context
Rhodes, James, Kang, George · 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: ef4c7989-b14c-4f52-8422-4e54cd017b66.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T17:49:04.791Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 3933df8be31283fe699317c63142d059165d92db4d256a7bac8bb176c389e19d. Source text hash: 2e3822582c323d5723208d9676b0b372a5417136c7f1fd3b39db19c4e413eb50.
Revision and source-check history
Version 1 · 2026-11-11T17:49:04.791Z
Rhodes and Kang argue work products should carry their context
James Rhodes and George Kang argue that a document can preserve a professional work product’s visible result while losing the research, calculations, assumptions, evidence and authorization boundaries behind it. They propose delivering that context in a Portable Agent Format (PAF) package: the work product travels with compiled context, evidence, provenance and rules for later interaction. In a four-package evaluation, a hosted governed reader succeeded on 96% of 400 frozen questions, compared with 86% for instructions plus raw-file retrieval. Results varied by package, and the authors report a specific failure: offline retrieval resolved none of 40 historical-record questions in one financial-data package. They also report that the offline reader’s -1 response rate, combining materially wrong and unauthorized answers, was 2.9%, compared with 56% for the source and instructions alone. The authors say the findings support treating compiled knowledge as part of the delivered work product. But the evaluation does not establish independent standards interoperability or end-to-end governance conformance, and its ablations do not separate representation effects from training effects.
Original source
cs.AI · Article
Source details
Original source date: Nov 9, 2026 (2026-11-09).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T17:49:04.791Z. First observed: 2026-11-11T05:06:21.604Z. Retrieved: 2026-11-11T17:48:14.471Z.
Limits
The abstract reports results from four packages and a held-out evaluation; it does not establish that the findings generalize to other work products or settings. The authors say the ablations do not isolate representation effects from training effects. The abstract says independent standards interoperability and end-to-end governance conformance remain unproven.
Open question
What evidence would establish independent standards interoperability and end-to-end governance conformance for PAF packages?
Source-stated
Rhodes and Kang argue that documents can omit important context behind professional work, and propose delivering the work product with compiled context, evidence, provenance and interaction rules in a Portable Agent Format package.
A document preserves the visible result of professional work but often discards the research, calculations, assumptions, evidence, and authorization boundaries behind it. We study a different unit of delivery: a Portable Agent Format (PAF) package
Source-stated
In the reported evaluation, the hosted governed reader succeeded on 96% of questions, compared with 86% for instructions plus raw-file retrieval.
Across 400 frozen questions and 12,000 evaluated responses, a hosted governed reader succeeds on 96% of questions versus 86% with instructions plus raw-file retrieval
Source-stated
The authors report a failure in which offline retrieval resolved none of the historical-record questions in one financial-data package.
The same run exposes a failure: offline retrieval resolves none of the 40 historical-record questions in one financial-data package.
Source-stated
The authors say the evaluation leaves independent standards interoperability and end-to-end governance conformance unproven.
while leaving independent standards interoperability and end-to-end governance conformance unproven.
Source and contributor context
Rhodes, James, Kang, George · 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: ef4c7989-b14c-4f52-8422-4e54cd017b66.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T17:49:04.791Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 3933df8be31283fe699317c63142d059165d92db4d256a7bac8bb176c389e19d. Source text hash: 2e3822582c323d5723208d9676b0b372a5417136c7f1fd3b39db19c4e413eb50.
uncertain · 2026-11-11T17:49:04.791Z · anthropic/claude-opus-5.5
Publication: ef4c7989-b14c-4f52-8422-4e54cd017b66
attribution — pass: The argument and results are correctly attributed to James Rhodes and George Kang, the listed authors of the arXiv paper. The draft keeps their claims as the authors' assertions.
Authors:James Rhodes, George Kang
summary — uncertain: Central terms go unexplained for a general reader: hosted governed reader, raw-file retrieval, frozen questions, the -1 response rate, ablations, and representation versus training effects. The draft also introduces the offline reader without saying it is a separate 0.8-billion-parameter model, and it gives that reader's 2.9% -1 rate without its 69% overall and 54% answerable success rates. Smallest repair: add the offline reader's size and success rates, and gloss these terms in plain language or with term-library tooltips.
A separate 0.8-billion-parameter offline reader succeeds on 69% overall, but 54% on answerable questions.
date — pass: originalPublished 2026-11-09 matches the v1 submission date. It is not the later feed date of 2026-11-11.
[Submitted on 9 Nov 2026]
support — pass: The figures, the failure case, the authors' conclusion and the unproven items all match the abstract. The phrase 'results varied by package' fairly reflects the abstract's statement that the gap varies by package. The question follows directly from limitations the authors themselves state. No benefits or costs are invented.
These findings support treating compiled knowledge as part of the delivered work product, while leaving independent standards interoperability and end-to-end governance conformance unproven.
media — pass: The selected media is the verified direct link to the original arXiv abstract page, which is the easiest 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 arXiv