pith. machine review for the scientific record. sign in

arxiv: 2604.06376 · v1 · submitted 2026-04-07 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

MTA-Agent: An Open Recipe for Multimodal Deep Search Agents

Xiangyu Peng , Can Qin , An Yan , Xinyi Yang , Zeyuan Chen , Ran Xu , Chien-Sheng Wu
Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:18 UTC · model grok-4.3

classification 💻 cs.CV
keywords multimodal deep searchmulti-hop reasoningtool-augmented agentsvision-language datasetsevidence verificationopen MLLM trainingmulti-stage filteringsynthetic trajectories
0
0 comments X

The pith

An automated pipeline builds 21K verified multi-hop visual search examples that train a 32B open model to exceed GPT-5 and Gemini variants on six benchmarks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that complex multimodal reasoning, which requires chaining visual evidence with external knowledge across multiple steps, can be improved by first synthesizing high-quality training trajectories automatically. MTA-Agent starts from ordinary VQA datasets and uses tool selection plus multi-stage verification to produce structured question-answer paths that enforce factual consistency and unique answers. Training a 32B open model on the resulting 21K examples raises average reasoning steps from 2.27 to 4.28 and delivers 54.63 percent average accuracy, surpassing closed models under matched tool conditions. The work also shows that cached interaction replay allows the same training without repeated live tool calls, lowering cost. The complete dataset, trajectories, and code are released so others can reproduce and build on the approach.

Core claim

MTA-Agent automatically selects tools and parameters to retrieve and validate evidence from visual and textual sources, then generates structured multi-hop question-answer trajectories; after multi-stage filtering for factual consistency and answer uniqueness, the resulting MTA-Vision-DeepSearch dataset of 21K examples trains open multimodal models to achieve higher accuracy and more persistent search behavior than leading closed models on the same benchmarks.

What carries the argument

MTA-Agent, a Multi-hop Tool-Augmented Agent for Evidence-based QA Synthesis that selects tools, retrieves evidence, and applies multi-stage verification to create verified multi-hop trajectories from VQA seed data.

If this is right

  • Training increases the average number of reasoning steps from 2.27 to 4.28 and produces more systematic, persistent search strategies.
  • The same performance gains are obtained when training replays cached tool interactions instead of making live calls, cutting training cost.
  • The open release of the full dataset, trajectories, and implementation details supports direct replication and extension by other researchers.
  • Gains hold across six diverse challenging benchmarks when the trained model uses the same tool settings as the closed-model baselines.
  • The method improves both the depth of reasoning chains and the quality of tool-use behavior in the resulting agents.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same synthesis approach could be applied to other modalities such as video or scientific imagery to test whether verified multi-hop trajectories close performance gaps elsewhere.
  • Releasing the full interaction trajectories allows detailed study of which verification steps most contribute to the observed gains in persistence and step count.
  • The cost reduction from cached replay suggests that large-scale agent training can become feasible for groups without continuous access to paid tool APIs.
  • Future tests could check whether mixing the 21K examples with existing general VQA data yields further gains or whether the method scales to even larger base models.

Load-bearing premise

The multi-stage verification process creates examples that genuinely deepen reasoning and tool-use skills rather than simply matching the target benchmarks or carrying over artifacts from the generator itself.

What would settle it

If a 32B model retrained on the released MTA-Vision-DeepSearch dataset shows no rise in average reasoning steps and fails to exceed the closed-model baselines on the six benchmarks, the claim that the synthesized data improves deep search capability would be falsified.

Figures

Figures reproduced from arXiv: 2604.06376 by An Yan, Can Qin, Chien-Sheng Wu, Ran Xu, Xiangyu Peng, Xinyi Yang, Zeyuan Chen.

Figure 1
Figure 1. Figure 1: Data creation pipeline. Starting from filtered seed VQA data, the QA agent [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Tool usage frequency across sce￾narios on all test datasets. Each bar shows the percentage of scenarios in which a given tool is used at least once [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Tool usage distribution on BrowseComp-VL [PITH_FULL_IMAGE:figures/full_fig_p030_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Tool usage distribution on MMSearch-VL [PITH_FULL_IMAGE:figures/full_fig_p030_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Tool usage distribution on HR￾MMSearch-VL [PITH_FULL_IMAGE:figures/full_fig_p030_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Tool usage distribution on MMSearch-Plus [PITH_FULL_IMAGE:figures/full_fig_p031_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Tool usage distribution on FVQA [PITH_FULL_IMAGE:figures/full_fig_p031_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: Image for Example 1. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Image for Example 2. 33 [PITH_FULL_IMAGE:figures/full_fig_p033_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Image for Example 3. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Image for Example 4. 36 [PITH_FULL_IMAGE:figures/full_fig_p036_17.png] view at source ↗
read the original abstract

Multimodal large language models (MLLMs) have demonstrated strong capabilities in visual understanding, yet they remain limited in complex, multi-step reasoning that requires deep searching and integrating visual evidence with external knowledge. In this work, we address this challenge by constructing high-quality, verified multi-hop vision-language training data for multimodal deep-search agents. We propose a Multi-hop Tool-Augmented Agent for Evidence-based QA Synthesis (MTA-Agent), which automatically selects tools and their parameters to retrieve and validate evidence from both visual and textual sources and generates structured multi-hop question-answer trajectories. Starting from diverse VQA seed datasets, our pipeline produces a large-scale training dataset, MTA-Vision-DeepSearch, containing 21K high-quality multi-hop examples. The data is filtered through a multi-stage verification process to ensure factual consistency and answer uniqueness. Using MTA-Vision-DeepSearch, a 32B open-source multimodal search agent achieves state-of-the-art performance, reaching an average of 54.63\% across six challenging benchmarks, outperforming GPT-5 (51.86\%), Gemini-2.5-Pro (50.98\%), and Gemini-3-Pro (54.46\%) under the same tool settings. We further show that training on our data improves both reasoning depth and tool-use behavior, increasing the average number of steps from 2.27 to 4.28, and leading to more systematic and persistent search strategies. Additionally, we demonstrate that training can be performed without real-time tool calls by replaying cached interactions, significantly reducing training cost. Importantly, we present MTA-Agent as a fully open recipe for multimodal deep search: we release the entire dataset, training trajectories, and implementation details to enable reproducibility and future research on open multimodal search agents.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript introduces MTA-Agent, a Multi-hop Tool-Augmented Agent for Evidence-based QA Synthesis that automatically selects tools to retrieve and validate evidence from visual and textual sources, generating structured multi-hop trajectories. Starting from VQA seed datasets, the pipeline produces the MTA-Vision-DeepSearch dataset of 21K high-quality examples filtered by multi-stage verification for factual consistency and answer uniqueness. A 32B open-source multimodal search agent trained on this data achieves state-of-the-art performance with an average of 54.63% across six benchmarks, outperforming GPT-5 (51.86%), Gemini-2.5-Pro (50.98%), and Gemini-3-Pro (54.46%) under identical tool settings. Training also increases average reasoning steps from 2.27 to 4.28 and enables cost-efficient training via cached interactions; the authors release the full dataset, trajectories, and implementation details as an open recipe.

Significance. If the results hold, the work is significant for providing a fully open, reproducible recipe and large-scale verified dataset for training multimodal deep-search agents, which could accelerate progress in open-source tool-augmented visual reasoning. The release of 21K examples and cached trajectories for reduced training cost is a concrete strength that supports reproducibility and community follow-up.

major comments (3)
  1. [Experimental Results] The central SOTA claim (54.63% average across six benchmarks) rests on comparisons under 'the same tool settings,' yet the experimental section provides no details on how tool parameters, APIs, or verification rules were equalized between the 32B open model and closed models such as GPT-5; without this, the direct outperformance cannot be evaluated.
  2. [Data Synthesis and Verification Pipeline] The multi-stage verification process is presented as ensuring high-quality data that improves reasoning depth, but no ablation is reported comparing verified trajectories against raw synthetic data or isolating individual verification stages; this is load-bearing for the claim that gains reflect genuine capability rather than generation artifacts or benchmark fitting.
  3. [Ablation and Analysis] The reported increase in average steps (2.27 to 4.28) is used to support improved tool-use behavior, but the results section contains no analysis of whether additional steps correlate with higher accuracy or simply reflect longer trajectories without better outcomes.
minor comments (1)
  1. [Abstract] The abstract refers to 'six challenging benchmarks' without naming them; adding an explicit list or reference to Table 1 in the introduction would improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback, which highlights important areas for improving the clarity and rigor of our manuscript. We address each major comment point by point below, indicating the revisions we will make to strengthen the paper.

read point-by-point responses

  1. Referee: [Experimental Results] The central SOTA claim (54.63% average across six benchmarks) rests on comparisons under 'the same tool settings,' yet the experimental section provides no details on how tool parameters, APIs, or verification rules were equalized between the 32B open model and closed models such as GPT-5; without this, the direct outperformance cannot be evaluated.

    Authors: We agree that explicit details on tool setting equalization are essential to substantiate the comparisons and enable reproducibility. In the revised manuscript, we will expand the Experiments section with a dedicated subsection that specifies the standardized tool parameters, API configurations, verification rules, and any caching mechanisms applied uniformly to the 32B model and the closed models (GPT-5, Gemini variants). This will include concrete examples of parameter values and protocols to allow readers to verify the fairness of the evaluation. revision: yes

  2. Referee: [Data Synthesis and Verification Pipeline] The multi-stage verification process is presented as ensuring high-quality data that improves reasoning depth, but no ablation is reported comparing verified trajectories against raw synthetic data or isolating individual verification stages; this is load-bearing for the claim that gains reflect genuine capability rather than generation artifacts or benchmark fitting.

    Authors: We acknowledge that the absence of ablations leaves the contribution of the verification pipeline insufficiently isolated. While the original work emphasized end-to-end results, we will add a new ablation subsection in the revised manuscript. This will compare model performance when trained on verified versus unverified trajectories (using a representative subset to control cost) and provide an analysis of the impact of each verification stage on data quality metrics such as factual consistency and answer uniqueness. revision: yes

  3. Referee: [Ablation and Analysis] The reported increase in average steps (2.27 to 4.28) is used to support improved tool-use behavior, but the results section contains no analysis of whether additional steps correlate with higher accuracy or simply reflect longer trajectories without better outcomes.

    Authors: We thank the referee for this suggestion. The increase in reasoning steps is intended to indicate more systematic search, but we agree that a correlation analysis is needed to link step count directly to performance gains. In the revision, we will add this analysis to the Results section, including binned accuracy averages by step count, correlation coefficients, and qualitative examples showing that longer trajectories yield higher accuracy rather than redundant steps. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external seeds and benchmarks

full rationale

The paper starts from external VQA seed datasets, uses MTA-Agent to synthesize trajectories, applies multi-stage verification for factual consistency and uniqueness, trains a 32B model on the resulting MTA-Vision-DeepSearch (21K examples), and reports gains on six independent external benchmarks (including comparisons to GPT-5 and Gemini variants). The reported step increase (2.27 to 4.28) and SOTA average (54.63%) are measured against those external benchmarks rather than the synthetic data itself. No equation, definition, or load-bearing claim reduces the final performance result to a fitted parameter or self-generated input by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim depends on the assumption that tool-augmented generation plus verification yields higher-quality training signals than prior methods, with no explicit free parameters fitted to the final benchmarks.

axioms (1)
  • domain assumption Multimodal LLMs can reliably select and invoke external tools to retrieve and validate evidence from images and text
    The MTA-Agent pipeline presupposes that the base model possesses sufficient tool-use capability to produce accurate trajectories.
invented entities (2)
  • MTA-Agent no independent evidence
    purpose: Agent that automatically chooses tools and parameters to retrieve, validate, and synthesize multi-hop vision-language QA trajectories
    Core contribution introduced to generate the training data.
  • MTA-Vision-DeepSearch no independent evidence
    purpose: Dataset of 21K verified multi-hop examples
    Output of the MTA-Agent pipeline used for training.

pith-pipeline@v0.9.0 · 5636 in / 1437 out tokens · 75767 ms · 2026-05-10T19:18:02.271856+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. TRACER: Verifiable Generative Provenance for Multimodal Tool-Using Agents

    cs.CL 2026-05 unverdicted novelty 7.0

    TRACER attaches verifiable sentence-level provenance records to multimodal agent outputs using tool-turn alignment and semantic relations, yielding 78.23% answer accuracy and fewer tool calls than baselines on TRACE-Bench.

Reference graph

Works this paper leans on

57 extracted references · cited by 1 Pith paper

  1. [1]

    People.Famous individuals who are uniquely named, including actors, musicians, scientists, politicians, historical figures, and athletes (e.g.,Marie Curie,Lionel Messi)

  2. [2]

    Organizations.Named institutions such as technology companies, NGOs, universities, sports teams, and government bodies (e.g.,UNICEF,Manchester United)

  3. [3]

    Locations.Specific geographic or architectural referents, including cities, countries, landmarks, airports, and nature reserves (e.g.,Eiffel Tower,Yellowstone)

  4. [4]

    Products.Commercially named goods such as consumer electronics, vehicles, and software (e.g.,iPhone 15,Tesla Model 3)

  5. [5]

    Events.Named occurrences with a defined scope, including global sporting events, historical events, conferences, and cultural festivals (e.g.,FIFA World Cup,Diwali)

  6. [6]

    Creative Works.Titled artistic or intellectual productions, including films, television series, books, artworks, and music albums (e.g.,Inception,Mona Lisa)

  7. [7]

    8.Medical & Health.Named diseases or medical conditions (e.g.,COVID-19,Malaria)

    Science & Technology.Named scientific entities such as biological species, chemical compounds, devices, AI models, and benchmark datasets (e.g.,GPT-4,ImageNet). 8.Medical & Health.Named diseases or medical conditions (e.g.,COVID-19,Malaria)

  8. [8]

    a bridge

    Financial & Business.Named companies, stock indexes, and commercial brands (e.g., Coca-Cola,S&P 500). 10.Space & Astronomy.Named celestial objects, space missions, and astronomical events (e.g.,Hubble Telescope,Mars). Answers that are numbers, adjectives, generic nouns (e.g., “a bridge”, “the car”), actions, or multi-entity phrases are explicitly excluded...

  9. [9]

    in the image

    The question should mention “in the image” and remove any answer choices

  10. [10]

    This person is playing a similar sport to whom?

    Ensure the question makes sense as a free-form question without choices. Questions like “This person is playing a similar sport to whom?” require choices to be answerable and should be marked as invalid

  11. [11]

    valid”: false, “reason

    If the question cannot be reasonably converted to free-form without choices, respond with {“valid”: false, “reason”: “Question requires multiple choice options to be answerable”}

  12. [12]

    The answer should be just the final entity/answer without any reasoning or explanation

  13. [13]

    Apple” (technology company)→complete_answer: “Apple, a technology company

    Provide both a short answer (output_answer) and acomplete_answer: • Add descriptive contextfrom the provided contextto disambiguate it. • Keep the new description less than 3 words—add the minimum context needed for disambiguation. Examples of disambiguation: • “Apple” (technology company)→complete_answer: “Apple, a technology company” • “Birmingham” (cit...

  14. [14]

    Queries are constrained to always include the complete disambiguated entity name to prevent irrelevant re- trievals

    Web Search.Given a text query, this tool issues a search request via the Tavily API and returns up to 10 results, each consisting of a URL, title, short snippet, and—when available—the full extracted page content in Markdown format. Queries are constrained to always include the complete disambiguated entity name to prevent irrelevant re- trievals

  15. [15]

    It is used to deepen coverage of a specific source when the snippet returned by web search is insufficient for synthesizing a well-grounded question

    Web Reader.Given a URL, this tool fetches and returns the full textual content of the target webpage. It is used to deepen coverage of a specific source when the snippet returned by web search is insufficient for synthesizing a well-grounded question

  16. [16]

    It is particularly useful for identifying specific objects, landmarks, or entities that are visually distinctive but difficult to describe in text alone

    Google Lens.Given an image, this tool performs a reverse image search and returns visually similar results with their associated titles and snippets. It is particularly useful for identifying specific objects, landmarks, or entities that are visually distinctive but difficult to describe in text alone

  17. [17]

    {entity_name}

    Image Search.Given a text query, this tool retrieves image URLs along with auto- generated descriptions and captions. It is used when visual context about the current entity—such as product appearance, architectural style, or logo design—can surface facts that text-only pages do not explicitly state. 19 Preprint. Under review. Together, these tools allow ...

  18. [18]

    The question must have a UNIQUE (singular) answer — exactly ONE correct answer

  19. [19]

    {entity_name}

    The question must include “{entity_name}” in the question

  20. [20]

    the passage

    The question must be standalone (no references to “the passage” or “the context”)

  21. [21]

    {entity_name}

    Removing “{entity_name}” from the question should make it have multiple possible answers

  22. [22]

    The question should be USEFUL and NATURAL

  23. [23]

    Ask about meaningful relationships, key facts, or important attributes

  24. [24]

    What movies has X appeared in?

    Do not add any redundant details in the questions The question must have ONLY ONE POSSIBLE ANSWER — AVOID: • Multiple possible answers (e.g., “What movies has X appeared in?”) • Time-dependent answers without time constraint (e.g., “Who is the current CEO?”) • Subjective answers (e.g., “What is the best...”) • Weak relational phrases (“associated with”, “...

    1995

  25. [25]

    Keep the question grammatically correct and natural

  26. [26]

    The simplified question must still have the SAME unique answer

  27. [27]

    Remove redundant temporal info (years, dates) if not needed for uniqueness

  28. [28]

    Remove redundant location info if not needed for uniqueness

  29. [29]

    {entity_name}

    Keep the entity “{entity_name}” in the question Do NOT simplify if all details are necessary for uniqueness, or if the question is already concise. Return: { "needs_simplification": <true|false>, "reason": "<why simplification is/isn't needed>", "simplified_question": "<simplified question, or original if no change needed>" } A.5.4 Step 4: Factual Verific...

  30. [30]

    Is the answer factually correct according to the given context?

  31. [31]

    {entity_name}

    Does the question contain the given entity “{entity_name}”?

  32. [32]

    Is this answer unique (not multiple possible answers)?

  33. [33]

    Is this answer stable over time (not changing)?

  34. [34]

    {entity_name}

    Removing “{entity_name}” from the question, is it impossible to get the unique answer? Return “VERIFIED” if all criteria are met, “REJECTED” if any fail. Explain your reasoning briefly. Candidates that pass verification receive acomplete answer—a short disambiguating phrase appended to the bare answer (e.g.,“Birmingham, city in Alabama”)—generated by GPT-...

  35. [35]

    Add descriptive contextfrom the provided contextto disambiguate the answer

  36. [36]

    Apple”→“Apple, a technology company

    Keep the added description to fewer than three words. Examples: •“Apple”→“Apple, a technology company” •“Birmingham”→“Birmingham, city in Alabama” •“Baltimore Orioles”→“Baltimore Orioles, a Baseball team” •“google.com”→“google.com, a website” •“Comcast”→“Comcast, a company” Return ONLY the complete answer, nothing else. The disambiguating phrase is always...

  37. [37]

    Q:In which city is Domenico Rambelli’s monument to Francesco Baracca located? A:Lugo Merged Questions (step-by-step composition): 1.After Hop 1→2:Who founded the brand of the vehicle in the image?

  38. [38]

    After Hop 2→3:Which flying ace’s prancing horse emblem was adopted for the cars of the brand of the vehicle in the image?

  39. [39]

    After Hop 3→4:Which sculptor created the monument dedicated to the flying ace whose prancing horse emblem was later adopted for the cars of the brand of the vehicle in the image?

  40. [40]

    32 Preprint

    After Hop 4→5 (Final):In which city is the monument located that honors the flying ace whose prancing horse emblem was later adopted for the cars of the brand of the vehicle in the image? Final Answer:Lugo Figure 14: Image for Example 1. 32 Preprint. Under review. Example 2 Image:See Figure 15. Individual Hops: 1.Q (Image):What organization is represented...

  41. [41]

    Q:What is the name of the underground metro line in Istanbul that is considered the world’s second-oldest? A:Istanbul Tünel 4.Q:Which sultan granted the concession to build the Istanbul Tünel? A:Sultan Abdülaziz

  42. [42]

    Q:Which British monarch invested Sultan Abdülaziz with the Order of the Garter? A:Queen Victoria Merged Questions (step-by-step composition):

  43. [43]

    After Hop 1→2:In which city is the home stadium of the organization represented by the emblem in the image located?

  44. [44]

    After Hop 2→3:What is the name of the underground metro line in the city where the home stadium of the organization represented by the emblem in the image is located, which is considered the world’s second-oldest?

  45. [45]

    After Hop 3 →4:Which sultan granted the concession to build the underground metro line in the city where the home stadium of the organization represented by the emblem in the image is located, which is considered the world’s second-oldest?

  46. [46]

    33 Preprint

    After Hop 4→5 (Final):Which British monarch invested with the Order of the Garter the sultan who granted the concession to build the underground metro line in the city where the home stadium of the organization represented by the emblem in the image is located, which is considered the world’s second-oldest? Final Answer:Queen Victoria Figure 15: Image for...

  47. [47]

    Q:Which saint is associated with giving Glasgow its name meaning ‘dear green place’? A:Saint Mungo 5.Q:In which Scottish town was Saint Mungo born? A:Culross Merged Questions (step-by-step composition):

  48. [48]

    After Hop 1→2:Who wrote the book in the image that predicted the fall of Detroit into poverty?

  49. [49]

    After Hop 2→3:In which Scottish city did the author of the book in the image that predicted the fall of Detroit into poverty die?

  50. [50]

    After Hop 3→4:Which saint is associated with giving the name meaning ‘dear green place’ to the Scottish city where the author of the book in the image that predicted the fall of Detroit into poverty died?

  51. [51]

    Under review

    After Hop 4→5 (Final):In which Scottish town was the saint born who is associated with giving the name meaning ‘dear green place’ to the Scottish city where the author of the book in the image that predicted the fall of Detroit into poverty died? Final Answer:Culross 34 Preprint. Under review. Figure 16: Image for Example 3. 35 Preprint. Under review. Exa...

  52. [52]

    president issued the executive order that first placed Spirit Island in Mille Lacs Lake under federal protection? A:Woodrow Wilson

    Q:Which U.S. president issued the executive order that first placed Spirit Island in Mille Lacs Lake under federal protection? A:Woodrow Wilson

  53. [53]

    Q:On which transport ship did Woodrow Wilson travel to the peace negotiations after World War I? A:USS George Washington Merged Questions (step-by-step composition):

  54. [54]

    After Hop 1→2:Which city is known as the capital of the holiday the boy in the image is likely celebrating?

  55. [55]

    After Hop 2→3:From which lake does the river flowing out of the city known as the capital of the holiday the boy in the image is likely celebrating originate?

  56. [56]

    After Hop 3→4:Which U.S. president issued the executive order that first placed Spirit Island in the lake from which the river flowing out of the city known as the capital of the holiday the boy in the image is likely celebrating originates, under federal protection?

  57. [57]

    After Hop 4→5 (Final):On which transport ship did the U.S. president who first placed Spirit Island in the lake from which the river flowing out of the city known as the capital of the holiday the boy in the image is likely celebrating originates, travel to the peace negotiations after World War I? Final Answer:USS George Washington Figure 17: Image for E...