Pith. sign in

REVIEW 2 major objections 6 minor 23 references

Back to the Future: A workbook time machine for spread sheet creation benchmarks

T0 review · 2 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A pipeline that erases derived artifacts from real Excel workbooks yields a 150-task benchmark, and evaluation on it shows that API choice, query specificity, and agent orchestration dominate model identity in spreadsheet-creation…

desk verdict The WTM pipeline is a real step forward for spreadsheet-creation benchmarks, but the unstated handling of Excel's cached formula values is a load-bearing caveat; the paper deserves review, not blind trust. read the letter →

arxiv 2608.07873 v1 pith:AFVUJZKW submitted 2026-08-08 cs.AI

Mansi Uniyal , Agamdeep Singh , Ananya Singha , Priyanshu Gupta , Mukul Singh , Gust Verbruggen , Vu Le , Sumit Gulwani This is my paper · ORCID
classification cs.AI
keywords workbooktimemachinespreadsheetbenchmarkLLMagentsformulagenerationpivottablesconditionalformattinginstructionspecificityreversecurriculum
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper presents the workbook time machine, a pipeline that turns finished, user-authored Excel workbooks into benchmark triples: an input workbook, an output workbook, and a natural-language query. The defining move is to run backward: strip formulas, charts, pivot tables, and conditional-formatting rules from a real workbook to recover the raw data, then run forward to generate instructions at controlled levels of specificity. Applied to public spreadsheet corpora, it produces a corpus of 8,931 queries and a curated 150-task benchmark balanced across artifact types. Evaluations on that benchmark claim that spreadsheet agents succeed or fail mainly because of the scripting API used to control Excel, how detailed the instruction is, and how the agent loop is orchestrated, rather than which frontier model is chosen. If correct, this gives the field a diagnostic instrument that can separate pure execution failures from instruction-following failures.

What carries the argument

The load-bearing mechanism is the Edit DAG, a directed acyclic graph whose nodes are workbook states obtained by adding subsets of the derived artifacts to the raw workbook and whose edges add one artifact at a time. Dependency-aware pruning removes edges whose prerequisite artifacts are absent, so only semantically valid edit histories remain; sub-paths of the pruned DAG become benchmark triples. A companion necessity score labels each action parameter as essential, useful, or incidental, and an LLM turns filtered parameter sets into natural-language instructions at three specificity levels. The DAG also yields a task-complexity measure: the number of artifact transformations between input and output states.

What would settle it

Take a real workbook in which a formula cell contains its last cached computed value, run the workbook time machine's backward step, and inspect the recovered raw workbook: if any numeric cell remains at the stripped formula's position, a model could copy that value instead of deriving it, and every score on the benchmark built from that triple is suspect.

Watch

Extended reading notes

Core claim

The paper's central claim is that WTM-BENCH is the first spreadsheet benchmark to jointly test language models on realistic, user-authored workbooks; on multi-step creation of formulas, charts, pivot tables, and conditional formatting; and on instructions whose specificity is controlled. The argument runs through the workbook time machine: starting from a finished workbook, it removes derived artifacts to recover a raw state, builds a dependency-pruned edit DAG of plausible artifact-creation histories, and generates natural-language queries at three specificity levels for each transformation. Evaluation across eighteen model-API combinations and several orchestration frameworks shows that query specificity, agent orchestration, and the Excel interface API shape performance more than model identity, with the best configuration using a native macro language. Formulas are the most tractable artifact, pivot tables remain nearly unsolved, and performance degrades monotonically as instructions become more abstract and as tasks require more steps.

Load-bearing premise

The load-bearing premise is that the backward step can strip formulas, charts, pivot tables, and conditional formatting from a finished workbook and recover a clean raw input without leaking the answer—in particular, that cached formula values and adjacent descriptor cells do not survive the stripping.

Editorial extensions

If this is right

  • WTM-BENCH scores should be read as properties of the whole stack—model, API, query level, and orchestration loop—not of model capability alone.
  • Comparing models through a single API can change their relative ranking, so benchmark results should report API as a condition rather than a confound.
  • Pivot-table creation is an open problem at the current frontier; a score below ten percent soft accuracy marks a clear target for improvement.
  • The three query levels give a direct way to tell whether a model fails because it cannot follow a terse instruction or because it cannot execute the underlying transformation, since the same workbook transformation is described at all three specificities.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If cached formula values are not fully stripped by the backward step, a model could copy the stored result instead of deriving it; a natural stress test is to run the pipeline on workbooks with stale cached values and measure the score drop.
  • The pipeline's direction can be inverted to create deletion and modification tasks by swapping input and output states—an extension the paper lists as future work but does not itself evaluate.
  • The finding that a large share of failed rollouts end with a confident completion claim suggests that practical spreadsheet agents need an independent verification step, not just better instruction-following.
  • If the specificity gradient holds up, training curricula that start at the most detailed query level and move to abstract queries could improve robustness to ambiguous instructions, but this remains to be tested.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper proposes the workbook time machine, a generation pipeline that starts from real user-authored Excel workbooks, removes derived artifacts (formulas, charts, pivot tables, conditional formatting) to recover a raw input state, reconstructs candidate edit histories as a dependency-pruned DAG, and generates natural-language queries at three specificity levels. From Enron and FUSE it produces WTM-CORPUS (8,931 queries over 2,977 tasks) and a curated WTM-BENCH (150 tasks, 450 queries). The paper evaluates six frontier LLMs under three Excel APIs and three orchestration frameworks, reporting that API choice, orchestration, and query specificity explain more variance than model identity, with pivot tables remaining near-unsolved.

Significance. WTM-BENCH fills a real gap: it is the only benchmark I know of that combines realistic workbooks, multi-artifact creation, and controlled instruction specificity. The design has genuine strengths: the grading is deterministic and programmatic with no LLM judge; the three query levels provide a useful diagnostic; the artifact-type balance is deliberate; and the independent Python re-runs in Appendix G show stable model ordering. The main claims are therefore meaningful if the benchmark triples are leakage-free. That conditional is not yet established, and it is the central point on which the paper's validity rests.

major comments (2)
  1. [§4.1 and Figure 2] The backward step's treatment of cached formula values is not described. In the xlsx format, a formula cell stores both the formula and its last computed value in <v>; if the backward step does not remove those cached values, then Win already contains the numeric answer to formula and downstream artifact tasks. The only trace is the unexplained 'Cached - out' label in Figure 2, and Appendix H grades cell values by computing formula results, so a model that copies the cached values (or does nothing) can receive full formula credit. The Future Work section admits that formula descriptor detection is still open, and Appendix A admits that manually entered lookup values are not handled, which are related leakage channels. Please specify exactly how cached values are stripped, verify on the released corpus that no <v> survives in Win, and report a no-op/copy control baseline and a leakage test. Because all 150 tasks and the headline API/query/orchestration ordering depend on these triples, this is load-bearing.
  2. [§4.2 and Appendix B] The Edit DAG pruning and the fallback rule (retain the connected artifact containing W when the pruned graph is disconnected) are justified by assertion rather than validation. No experiment checks that each retained triple has at least one valid edit sequence from Win to Wout, or that the LLM-based semantic descriptor mapping and necessity-score annotations are correct. The human study in Appendix B rates query naturalness and completeness, not task derivability. I would like to see a derivability audit on a sample (for example, executing the intended action parameters from Win and comparing the result with Wout), or at least a human-validated subset with agreement statistics.
minor comments (6)
  1. [§8 and Appendix H] Hard Score is described as 'exact task reconstruction', but Soft Score intentionally ignores axis labels, colors, rule priority, and aggregation function names, so Hard Score is exact only with respect to the weighted component metrics, not exact workbook reconstruction.
  2. [Figure 16(b)] The labels 'lvl1, lvl3, lvl5' appear even though the paper defines three query levels; this is a typo that should be corrected.
  3. [§5 and Appendix C] The phrase 'sequential generation (Level 1→2→3) sequentially in 1 LLM call' is confusing; please clarify whether the three levels are produced in a single call or in sequential calls, since the two interpretations have different cost and leakage implications.
  4. [Appendix G] Re-run robustness is reported only for Python/OpenPyXL; the paper should state explicitly that the VBA and OfficeJS results are single-run, or provide re-runs for those APIs if the API-comparison claim is to carry equal weight.
  5. [§7 and Appendix F] The comparison across APIs is confounded by backend-specific system prompts and function-calling schemas; this should be acknowledged as a limitation of the API-effect interpretation.
  6. [§5 and §7] The same model (GPT-5.4 Reasoning) is used both to generate the benchmark queries and as one of the evaluated models; the paper should discuss whether this creates any style or format advantage and whether contamination was considered.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: benchmark generation and grading are separated, with no fitted parameter renamed as a prediction; the main validity risk (cached formula values in stripped workbooks) is a leakage concern, not an equation-level circularity.

full rationale

WTM-BENCH is constructed by a backward pass that strips derived artifacts to form Win and a forward pass that generates queries from the stripped artifacts (Sections 4.1-4.3); the evaluation then grades model outputs against Wout with a deterministic programmatic grader. The paper states: "All grading is deterministic and programmatic: no LLM is used to judge model outputs, and the query generator is never consulted during evaluation." Thus the headline findings (API choice, query specificity, and orchestration matter more than model identity) are empirical outcomes rather than identities forced by construction. Necessity scores and task-complexity counts are design-time annotations, not parameters fitted to the evaluation results. Related-work self-citations (InstructExcel, TableTalk, ConDABench) are used only to position the benchmark and do not carry the derivation. The admitted limitations - Future Work lists "formula descriptor detection" to "minimize structural information leakage", and Appendix A notes that manually entered lookup values are not handled - are answer-leakage risks that could undermine validity if cached formula values or descriptor cells survive into Win, but they do not reduce any claimed result to its own input by definition. The internal comparison validating multi-level generation (Section 5 and Appendix C) uses the same necessity-score annotations that define the levels; this is a design-consistency check, not a fitted prediction, and does not affect the model-ordering results. Score 1 reflects the minor internal validation and self-citations without any load-bearing circular step.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The pipeline introduces no new physical entities, but the benchmark claims rest on several hand-set weights and domain assumptions about corpus representativeness, leakage-free stripping, LLM extraction reliability, and grader validity.

free parameters (4)
  • WTM-BENCH artifact distribution targets = formulas 41.3%, charts 23.3%, conditional formatting 20.0%, pivot tables 15.3%
    Chosen by the authors to rebalance WTM-CORPUS (67.5% formulas, 0.8% pivot tables); this composition shapes aggregate scores and benchmark difficulty.
  • Soft Score component weights = chart type 40%, series count 20%, series refs 40%; pivot sheet placement 15%, row fields 25%, column fields 20%, data…
    Hand-set in Appendix H; these weights determine every reported score and the 'nearly unsolved' pivot-table conclusion.
  • Necessity score thresholds = levels: all parameters (L1), s>0 (L2), s=1 (L3)
    Authors' design choice for controlling instruction specificity; the three levels' composition depends on these thresholds.
  • Turn cap = 10 turns
    Multi-turn agents are cut off at 10 turns; 24.2% of Haiku Python rollouts hit the cap, so results depend on this budget.
assumptions (6)
  • domain assumption Enron and FUSE workbooks are representative of real-world spreadsheet use.
    All benchmark tasks are derived from these two public corpora; Section 5 uses them as the sole source.
  • domain assumption Removing derived artifacts and pruning by dependency overlaps yields the true raw state and semantically valid edit histories.
    Section 4.2 assumes the dependency-pruned DAG captures all valid orderings; no ground-truth edit history is available for verification.
  • ad hoc to paper An LLM can reliably segment workbook content into raw data vs derived artifacts and assign necessity scores.
    Sections 4.1 and 4.3 rely on GPT-5.4 Reasoning for table-range identification, semantic descriptor mapping, and necessity annotation; errors propagate into query correctness.
  • domain assumption Sequential multi-level query generation preserves task semantics better than direct Level 3 generation.
    Validated only on 75 tasks by the authors themselves in Appendix C; no independent replication.
  • domain assumption The programmatic grader's component similarity measures task completion.
    Appendix H defines the grader; it ignores labels, colors, fonts, and rule priority, so 'completion' is partial by design.
  • ad hoc to paper If the pruned DAG is disconnected, retaining the connected artifact containing the final state preserves task validity.
    Section 4.2 heuristic, not derived from data or a model of user behavior.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Back to the Future: A workbook time machine for spread sheet creation benchmarks." pith.science (2026). https://pith.science/paper/AFVUJZKW

@misc{pith2026260807873,
  author       = {Pith},
  title        = {Pith review of: Back to the Future: A workbook time machine for spread sheet creation benchmarks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AFVUJZKW}},
  note         = {Machine review of arXiv:2608.07873}
}
read the original abstract

We introduce the workbook time machine, a pipeline that automatically creates benchmarks evaluating the ability of language models to create derived objects in spreadsheets (formulas, charts, pivot tables, and conditional formatting). Applied to public workbook corpora, it produces wtmcorpus--a collection of (input workbook, output workbook, query) triples spanning four artifact types and varying complexity. From this corpus we curate wtmbench, a 150-task evaluation benchmark with queries at three levels of specificity. We evaluate existing spreadsheet manipulation agents and baselines on wtmbench across artifact types, step complexity, and instruction granularity. Our evaluations show that query specificity, agent orchestration, and interface API used to control spreadsheets play a big role in LLM performance on Excel tasks.

Figures

Figures reproduced from arXiv: 2608.07873 by the authors.

Figure 1
Figure 1. Example of creating a benchmark with the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Generation pipeline flowchart 4.1 Component extraction. Recall from Section 3 that a workbook W = {D} ∪ C comprises raw data D and derived artifacts C = {c1, . . . , cn}. Starting from the final workbook W, the backward step strips away derived artifacts to recover the raw state W0 = {D}, and then reconstructs the candidate edit histories–all semantically valid orderings in which the artifacts could have been added … view at source ↗
Figure 3
Figure 3. WTM-BENCH achieves better artifact distribution while preserving other attributes. Blue bars for WTM-CORPUS and red bars for WTM-BENCH dataset. Addressing distribution imbalance. Our primary design challenge was the extreme skew in the original distribution where formulas dominate (67.5%) while pivot tables represent only 0.8% of tasks. This imbalance would render aggregate performance metrics meaning￾less, as they … view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Task distribution analysis over WTM-BENCH. Level 1 Level 2 Level 3 Query Level 0 20 40 60 80 100 Percentage (%) 76.0% 24.0% 51.3% 47.3% 29.3% 67.3% Specificity Distribution Specificity Level Ambiguous Reasonably Specified Very Well-Specified Well-Specified (a) Specific…
Figure 5
Figure 5. Figure 5: Query characteristic analysis. 7 Experiment Setup Models. We evaluate 6 state-of-the-art language models spanning two major families: Anthropic (Claude Haiku 4.5, Claude Sonnet 4.5, Claude Opus 4.6) and OpenAI (GPT-4.1 Mini, GPT-5.2 Reasoning, GPT-5.4 Reasoning). Orche…
Figure 6
Figure 6. Figure 6: Soft scores decrease with complexity; Haiku 4.5 outperforming GPT-4.1 Mini. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Query quality assessment: score distributions and annotation reliability analysis. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Information retention comparison between sequential multi-level generation (left) [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Comparing query quality assessment over a multi-level vs. single-generation pass [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Pipeline construction efficiency: time distribution across stages (left) and scaling [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Task-type and capability analysis of all query levels in the dataset (8,931 queries). [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: NAICS sector and financial scenario distributions across the complete 3k dataset. [PITH_FULL_IMAGE:figures/full_fig_p018_12.png]
Figure 13
Figure 13. Figure 13: Level analysis across the complete dataset: specificity (left), query length (centre), [PITH_FULL_IMAGE:figures/full_fig_p018_13.png]
Figure 14
Figure 14. Figure 14: Evaluation metrics comparison. For it, Hard match: [PITH_FULL_IMAGE:figures/full_fig_p022_14.png]
Figure 15
Figure 15. Figure 15: Example 1: Clubbed formula group being added. Where the colored texts are the [PITH_FULL_IMAGE:figures/full_fig_p022_15.png]
Figure 16
Figure 16. Figure 16: Examples of complex spreadsheet tasks 22 [PITH_FULL_IMAGE:figures/full_fig_p022_16.png]
Figure 17
Figure 17. Figure 17: Example 4: Multi-step formula tasks 23 [PITH_FULL_IMAGE:figures/full_fig_p023_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 14 canonical work pages

  1. [3]

    sequentially would provide better information retention and query quality than directly generating Level 3 queries. The rationale was that starting with verbose, well-specified Level 1 queries would preserve essential task details that could be progressively condensed while maintaining semantic fidelity. Empirical comparison methodology.To validate this d...

    work page 2026

  2. [8]

    11 Published as a conference paper at COLM 2026 Jenny T

    URL https://proceedings.neurips.cc/paper_files/paper/2023/ file/0ff30c4bf31db0119a6219e0d250e037-Paper-Conference.pdf. 11 Published as a conference paper at COLM 2026 Jenny T. Liang, Aayush Kumar, Yasharth Bajpai, Sumit Gulwani, Vu Le, Chris Parnin, Arjun Radhakrishna, Ashish Tiwari, Emerson Murphy-Hill, and Gustavo Soares. Tabletalk: Scaffolding spreadsh...

    work page 2023

  3. [9]

    Zeyao Ma, Bohan Zhang, Jing Zhang, Jifan Yu, Xiaokang Zhang, Xiaohan Zhang, Sijia Luo, Xi Wang, and Jie Tang

    doi: 10.1145/3765286. Zeyao Ma, Bohan Zhang, Jing Zhang, Jifan Yu, Xiaokang Zhang, Xiaohan Zhang, Sijia Luo, Xi Wang, and Jie Tang. Spreadsheetbench: Towards challenging real world spreadsheet manipulation.Advances in Neural Information Processing Systems, 37:94871–94908,

    work page doi:10.1145/3765286

  4. [10]

    Code volume is the strongest negative predictor, suggesting that long code generations often mark unsuccessful repair attempts

    show that tool failure is not the main bottleneck: frontier models sit at 3–5% tool-call failure, and score correlation with tool-call failure is weak. Code volume is the strongest negative predictor, suggesting that long code generations often mark unsuccessful repair attempts. Exploration polarity differs by family: read-only exploration correlates nega...

    work page 2026

  5. [12]

    NAICS.North American Industry Classification System, 2022,

    URL https://techcommunity.microsoft.com/blog/excelblog/ agent-mode-in-excel-is-now-generally-available-on-desktop/4457408. NAICS.North American Industry Classification System, 2022,

    work page arXiv 2022

  6. [14]

    Justin Payan, Swaroop Mishra, Mukul Singh, Carina Negreanu, Christian Poelitz, Chitta Baral, Subhro Roy, Rasika Chakravarthy, Benjamin Van Durme, and Elnaz Nouri

    URLhttps://chatgpt.com/apps/spreadsheets. Justin Payan, Swaroop Mishra, Mukul Singh, Carina Negreanu, Christian Poelitz, Chitta Baral, Subhro Roy, Rasika Chakravarthy, Benjamin Van Durme, and Elnaz Nouri. In- structexcel: A benchmark for natural language instruction in excel.arXiv preprint arXiv:2310.14495,

    arXiv

  7. [15]

    The impact of ai on developer productivity: Evidence from github copilot.arXiv preprint arXiv:2302.06590,

    Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. The impact of ai on developer productivity: Evidence from github copilot.arXiv preprint arXiv:2302.06590,

    arXiv

  8. [16]

    12 Published as a conference paper at COLM 2026 Dixin Tang, Fanchao Chen, Christopher De Leon, Tana Wattanawaroon, Jeaseok Yun, Srini- vasan Seshadri, and Aditya G

    doi: 10.1145/3732294. 12 Published as a conference paper at COLM 2026 Dixin Tang, Fanchao Chen, Christopher De Leon, Tana Wattanawaroon, Jeaseok Yun, Srini- vasan Seshadri, and Aditya G. Parameswaran. Efficient and compact spreadsheet formula graphs.Proceedings of the VLDB Endowment,

    doi:10.1145/3732294 2026

Show all 23 references
  1. [17]

    If llm is the wizard, then code is the wand: A survey on how code empowers large language models to serve as intelligent agents

    Ke Yang, Jiarui Liu, Jing Wu, Chaoqi Yang, Yifei Fung, Sha Li, Zhenqiao Huang, Xue Nan Cao, Xingyao Wang, Yizhong Wang, et al. If llm is the wizard, then code is the wand: A survey on how code empowers large language models to serve as intelligent agents. arXiv preprint arXiv:...

    arXiv

  2. [18]

    This scope is taxonomic rather than purely frequency-based

    13 Published as a conference paper at COLM 2026 A Limitations and Scope The current version of WTM-BENCHfocuses on derived analytical artifacts: formulas, charts, conditional formatting, and pivot tables. This scope is taxonomic rather than purely frequency-based. These artifa...

    2026

  3. [20]

    single-generation pass of 75 queries

    0.0 0.2 0.4 0.6 0.8 1.0 Average Correlation Coefficient (b) Inter-annotator agreement Figure 9: Comparing query quality assessment over a multi-level vs. single-generation pass of 75 queries. intermediate workbook states; Component extraction accounts for 25% (mean 102 s) and ...

    2026

  4. [21]

    It offers comprehensive support for reading and writing Excel 2010 xlsx/xlsm files, including formulas, charts, and conditional formatting

    is a Python library providing direct program- matic access to Excel files. It offers comprehensive support for reading and writing Excel 2010 xlsx/xlsm files, including formulas, charts, and conditional formatting. However, limitations include lack of VBA macro support, missin...

    2010

  5. [23]

    For it, Hard match: Fail, Soft match: Pass

    21 Published as a conference paper at COLM 2026 (a) Generated (b) Ground Truth Figure 14: Evaluation metrics comparison. For it, Hard match: Fail, Soft match: Pass. Calculate the percentage change for all years. Figure 15: Example 1: Clubbed formula group being added. Where th...

    2026

  6. [245]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    OpenAI. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    arXiv

  7. [2011]

    Felienne Hermans

    doi: 10.1145/1926385.1926423. Felienne Hermans. Spreadsheets are code. In2016 IEEE 23rd International Conference on Software Analysis, Evolution, and Reengineering (SANER),

  8. [2015]

    Faster, higher, stronger: Redesigning spreadsheets for scale

    Mangesh Bendre, Tana Wattanawaroon, Sajjadur Rahman, Kelly Mack, Yuyang Liu, Shichu Zhu, Yu Lu, Ping-Jing Yang, Xinyan Zhou, Kevin Chen-Chuan Chang, et al. Faster, higher, stronger: Redesigning spreadsheets for scale. In2019 IEEE 35th International Conference on Data Engineeri...

    1972

  9. [2016]

    doi: 10.1109/SANER.2016

    2016 doi

  10. [2017]

    openpyxl: A python library to read/write excel 2010 xlsx/xlsm files,

    Eric Gazoni and Charlie Clark. openpyxl: A python library to read/write excel 2010 xlsx/xlsm files,

    2010

  11. [2019]

    Teaching large language models to self-debug.arXiv preprint arXiv:2304.05128,

    Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. Teaching large language models to self-debug.arXiv preprint arXiv:2304.05128,

    arXiv

  12. [2023]

    Sheetagent: towards a generalist agent for spreadsheet reasoning and manipulation via large language models

    Yibin Chen, Yifu Yuan, Zeyu Zhang, Yan Zheng, Jinyi Liu, Fei Ni, Jianye Hao, Hangyu Mao, and Fuzheng Zhang. Sheetagent: towards a generalist agent for spreadsheet reasoning and manipulation via large language models. InProceedings of the ACM on Web Conference 2025, pp. 158–177,

    2025

  13. [2024]

    Microsoft 365 Consumer sub- scribers grew to 82.5 million

    URL https://www.microsoft.com/en-us/ investor/earnings/fy-2024-q4/press-release-webcast . Microsoft 365 Consumer sub- scribers grew to 82.5 million. July 30,

    2024

  14. [2025]

    Spreadsheetllm: Encoding spreadsheets for large language models.arXiv preprint arXiv:2407.09025,

    Haoyu Dong, Jianbo Zhao, Yuzhang Tian, Junyu Xiong, Shiyu Xia, Mengyu Zhou, Yun Lin, José Cambronero, Yeye He, Shi Han, et al. Spreadsheetllm: Encoding spreadsheets for large language models.arXiv preprint arXiv:2407.09025,

    arXiv

  15. [2026]

    Microsoft

    URLhttps://arxiv.org/abs/2601.11868. Microsoft. Earnings release FY24 Q4,

    arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.