DoesItLocal

Score the task, not the model

Does it run local?

Look up a task and get one honest verdict: can a local LLM safely do it, which model clears it, and what to use instead when none can.

Early prototype. This is the task catalog — 6 of 23 tasks have a verdict from a real local-model eval run; the rest are honestly marked "not yet evaluated." Nothing here is fabricated. How a task becomes a verdict →

/

All 23 Coding 4 Data analysis 2 Extraction 4 Instruction following 2 Language 4 Math 2 Reasoning 2 Writing 3

23 tasks in the catalog · 6 evaluated

• Generate a regex from a plain-English description Turn a written matching rule into a working regular expression. Coding 🟡 With a check
• Produce JSON conforming to a given schema Emit JSON that validates against a supplied JSON Schema with no extra or missing keys. Instruction following 🟡 With a check
• Reformat records into a strict output template Rewrite each input record into a fixed line template, same fields, same order, every time. Instruction following 🟡 With a check
• Rewrite a paragraph to a 6th-grade reading level Simplify a passage to a target U.S. grade-6 reading level while preserving all factual content. Language 🟡 With a check
• Translate UI strings to Spanish keeping placeholders intact Translate a JSON/PO file of English UI strings to Spanish without altering interpolation placeholders or markup. Language 🟡 With a check
• Solve a multi-step arithmetic word problem Turn a 3-5 step quantitative word problem into a single correct numeric answer. Math 🔶 More data
• Add type annotations to an untyped Python function Infer and insert PEP 484 type hints on an existing function. Coding Not yet evaluated
• Explain a stack trace and the likely cause Read an exception traceback and name the probable root cause. Coding Not yet evaluated
• Write a unit test for a pure function Given a small pure function, generate a passing test suite covering its branches. Coding Not yet evaluated
• Summarize the key statistics and outliers in a CSV Read a CSV and produce a prose summary of its distributions, key stats, and outlier rows. Data analysis Not yet evaluated
• Write a SQL query from a natural-language question over a known schema Turn a plain-English data question into a SQL query against a provided table schema. Data analysis Not yet evaluated
• Classify a support ticket by topic and urgency Assign an incoming support ticket to one topic from a fixed taxonomy and one of four urgency levels. Extraction Not yet evaluated
• Extract action items and owners from an email thread Pull the open action items, each with its assigned owner and any due date, out of a multi-message email thread. Extraction Not yet evaluated
• Extract structured fields from an invoice Pull vendor, date, line items, and total from a PDF/email invoice into JSON. Extraction Not yet evaluated
• Redact PII from a support transcript Remove or mask every piece of personal data — names, emails, phone numbers, addresses, account and card numbers — from a customer support chat transcript. Extraction Not yet evaluated
• Proofread and fix grammar without changing meaning Correct spelling, grammar, and punctuation in a passage while preserving the author's wording and intent. Language Not yet evaluated
• Summarize a meeting transcript Condense a call transcript into decisions, action items, and owners. Language Not yet evaluated
• Scale a recipe with unit conversions Multiply ingredient quantities by a factor and convert to target units (cups, grams, ml). Math Not yet evaluated
• Order steps into a correct dependency sequence Reorder a shuffled list of steps so every prerequisite comes before the step that needs it. Reasoning Not yet evaluated
• Solve a multi-constraint scheduling problem Given people, rooms, and hard/soft constraints, produce a valid conflict-free schedule. Reasoning Not yet evaluated
• Draft a contract indemnification clause Write an enforceable indemnification clause tailored to a described deal. Writing Not yet evaluated
• Draft a professional email reply from a short brief Turn a one-line intent plus the email being answered into a polished, send-ready reply. Writing Not yet evaluated
• Write release notes from a changelog or diff Convert a raw git log, merged-PR list, or diff into grouped, human-readable release notes. Writing Not yet evaluated

The verdict scale (safe for local / local with a check / needs a bigger model / needs more data) is described in Reading a verdict. Start with the introduction for why this exists.