#plain-text-accounting
Research grounded in plain-text accounting formats and workflows
Синтетично предизвикателство с три реда Beancount с независимо изведен отговор — проверката приключва на 2958.50 USD, септемврийската печалба на 1958.50 USD — изпълнено от един записан агентски прогон, като неуспехите му са запазени в транскрипта.
ReDAct runs a small model by default and escalates to an expensive model only when token-level perplexity signals uncertainty, achieving 64% cost savings over GPT-5.2-only while matching or exceeding its accuracy — a directly applicable pattern for Beancount transaction-categorization agents.
OpenHands is an MIT-licensed, Docker-sandboxed agent platform where CodeAct achieves 26% on SWE-Bench Lite — a sobering benchmark that establishes what AI agents can reliably do today, and why the first productive finance deployments should be tightly scoped rather than autonomous.
The LLMFinLiteracy benchmark finds that five open-weight ~7B models generate fully correct Beancount transactions only 2.3% of the time, with failures concentrated in accounting reasoning—not syntax—pointing to compiler-in-the-loop feedback as the critical missing ingredient for reliable write-back agents.
TableMaster is a prompting-only pipeline that reaches 78.13% on WikiTQ with GPT-4o-mini—13 points above Chain-of-Table—by combining table-of-focus extraction, semantic verbalization, and adaptive switching between text and symbolic reasoning. Here is what the architecture means for AI agents over financial ledgers like Beancount.
τ²-bench extends agent benchmarking to dual-control settings where both the AI and the user invoke tools over shared state — finding that active users cut success rates by 18–25 percentage points, with direct implications for Beancount agents sharing write access with human users.
GAIA benchmarks 466 real-world tasks across three difficulty levels; frontier agents reached 74.55% in mid-2026 versus 92% for humans, and the remaining Level 3 gap maps directly to the multi-step coordination challenges in automated Beancount ledger workflows.
WorkArena benchmarks LLM web agents on 33 real ServiceNow tasks — GPT-4o reaches 42.7% overall but 0% on list-filter tasks, exposing a hard wall between form-filling and structured UI interaction that maps directly to challenges in Beancount ledger automation.
τ-bench shows that top LLMs like Claude 3.5 Sonnet drop from pass@1 of 0.692 to pass@4 of 0.462 in retail customer-service tasks — a consistency cliff with direct implications for any write-back agent operating on a Beancount ledger.
Chain-of-Table (ICLR 2024) improves LLM tabular reasoning by evolving the table itself as the intermediate state — achieving 67.31% on WikiTQ vs. 61.48% for prior baselines, with a +10.25 point advantage on tables exceeding 4,000 tokens and direct applicability to Beancount ledger query agents.
TableLlama fine-tunes Llama 2 (7B) on 2.6M table-task examples and beats GPT-4 on structural tasks like column type annotation (F1 94 vs 32), but falls 33 points short on WikiTQ compositional reasoning — a calibrated benchmark for what 7B open models can and cannot do in finance AI today.
TAPAS (Google Research, ACL 2020) answers table questions by selecting cells and applying scalar aggregations — no SQL generated. This post analyzes the architecture, its 12-point SQA accuracy gain, and why the cell-selection paradigm fits small Beancount ledger queries but breaks down at scale.
