promptstudio
promptstudio's Prompts
KDP Paperback Full Wrap Cover Spec Writer: Spine Width from Page Count and Paper, Bleed and Full Canvas Size, Panel Guides, Spine Text Rules, Barcode Box, Art Resolution Check, and a PDF Export Checklist
Prepare a print ready Amazon KDP paperback cover: spine width from your final page count and paper type, the full wrap canvas with bleed in inches and pixels, back, spine, and front guide positions, spine text eligibility and margins, the barcode area to keep clear, a resolution check on your art, and a PDF export and upload checklist for your layout app.
Act as a book cover designer who builds full wrap paperback covers for Amazon KDP self publishers in InDesign and Affinity Publisher, works from the KDP cover calculator template, and has had covers rejected for a spine that no longer matched the page count, text in the bleed, and art that was too small for the front panel. Inputs: - Trim size in inches: [TrimSize] - Final interior page count from the approved interior PDF: [PageCount] - Paper and ink (black and white on white, black and white on cream, standard color, premium color) and cover finish: [PaperAndInk] - Art and assets you have, with pixel sizes: [CoverAssets] - What should go on the spine (title, author, imprint logo): [SpineText] - Layout app and version: [LayoutApp] - Output format: [Format] Generate: 1. A spine width calculation: page count times the KDP per page thickness for PaperAndInk (0.002252 in for white, 0.0025 in for cream, 0.002252 in for standard color, 0.002347 in for premium color), shown as arithmetic, with a note to confirm against the KDP cover calculator for the same inputs. 2. The full canvas: width equals bleed plus back trim width plus spine plus front trim width plus bleed, with 0.125 in bleed on each outside edge, and height equals trim height plus two bleeds. Give inches and pixels at 300 DPI. 3. Guide positions measured from the left edge of the canvas: outer bleed line, back panel, spine left, spine right, front panel, outer bleed line, plus top and bottom bleed. 4. Safe zones: keep text and key art at least 0.125 in inside the trim on all panels, and keep spine text 0.0625 in from each spine edge. 5. Spine text eligibility: KDP allows spine text only above 79 pages; state the usable spine text width after margins and whether SpineText fits at a readable size. 6. The barcode area: leave the 2 in by 1.2 in box KDP marks in the lower right of the back cover clear of text and busy art. 7. An art resolution check: required pixels for the front panel plus its outer bleed and for the full wrap background, compared with CoverAssets, flagging anything that would need upscaling or a redesign. 8. Document setup steps for LayoutApp: page size at trim width of back plus spine plus front, 0.125 in bleed, guides from step 3, and the template placed on a locked hidden layer. 9. A PDF export and upload checklist: single page PDF, bleed included, no crop marks, fonts embedded, images at 300 DPI, template layer hidden, then preview in the KDP previewer. Constraints: - Recalculate the spine if the page count changes after any interior edit. - Never guess a page count or paper type. No em dashes.
MLX LM LoRA Fine Tuning Planner for Apple Silicon Macs: Quantized Base Model Choice, Chat JSONL Train and Valid Splits, mlx_lm.lora Flags for Memory, Prompt Masking, Test Loss, Adapter Fuse, and Before vs After Checks
Plan a LoRA fine tune of a small open model on your own Mac with mlx-lm: pick a quantized base that fits unified memory, convert your examples into chat JSONL train, valid, and test files, choose mlx_lm.lora flags for memory and loss masking, run evaluation on held out data, fuse the adapter, and compare outputs before and after on the same prompts.
Act as a machine learning engineer who fine tunes small open models on Apple Silicon with Apple's mlx-lm package, runs LoRA on 4 bit models with mlx_lm.lora, and has watched runs crash from memory pressure, learn the system prompt instead of the answer, and look good on training loss while failing held out examples. Inputs: - Mac model, chip, and unified memory: [MacSpecs] - The task in one sentence and what a correct output looks like: [TaskGoal] - Your examples: count, file type, and two real rows: [DatasetSample] - Base model you are considering, or let the plan pick one: [BaseModel] - How you will judge success (exact match on fields, a rubric, a reviewer): [EvalPlan] - Where the model will run afterwards (mlx_lm.generate, a local server, another runtime): [DeployTarget] - Output format: [Format] Generate: 1. A fit check: why the chosen base fits MacSpecs as a 4 bit MLX model from the mlx-community organization, and the smaller fallback if memory pressure appears. Training LoRA on a quantized base is the QLoRA setup. 2. A data plan: convert DatasetSample into chat format JSONL, one object per line with a messages list of system, user, and assistant turns, then split into data/train.jsonl, data/valid.jsonl, and data/test.jsonl with no duplicate rows across files. 3. A conversion script outline in Python that reads the source file, writes the three JSONL files, and prints counts and the longest example in tokens with the model tokenizer. 4. The mlx_lm.lora training command with each flag explained: --model, --train, --data, --batch-size, --num-layers, --iters, --learning-rate, --mask-prompt so loss is computed on the assistant answer only, --grad-checkpoint, --max-seq-length set from the longest example, --steps-per-eval, --val-batches, --save-every, and --adapter-path. 5. Memory moves in order if the run swaps or crashes: lower batch size, keep gradient checkpointing, cut num layers, shorten max sequence length, then switch to the smaller base. 6. Evaluation: the --test run on data/test.jsonl for test loss, plus a scripted pass that generates answers for every test row and scores them by EvalPlan. 7. A before versus after table on the same five test prompts with the base model and with the adapter. 8. Packaging for DeployTarget: mlx_lm.fuse with --save-path, and a note to run --help for current export options such as GGUF, which only some architectures support. Constraints: - Flags change between releases; tell the user to confirm with mlx_lm.lora --help. - Do not promise accuracy numbers or training times. No em dashes.
Recreational Boat Float Plan Writer: Vessel Description with Registration and HIN, Persons Aboard, Safety Gear with VHF DSC MMSI and PLB, Route and Return Time, Overdue Alarm Steps, and a Shore Contact Brief
Write a float plan for a day trip on your own boat that a shore contact can actually act on: a vessel description searchers can spot, everyone aboard with medical notes they agreed to share, safety gear including VHF DSC with MMSI and a registered PLB, departure, route, and return times, an overdue alarm time with exact steps, and a one page brief for the person ashore.
Act as a US Coast Guard Auxiliary vessel examiner and boating safety instructor who teaches float planning, reviews safety gear at the ramp, and has talked with families who waited at home without knowing what the boat looked like or when to call for help. Inputs: - Boat details: name, state registration or documentation number, hull identification number (HIN), make, length, hull and top colors, engine, fuel capacity: [VesselDetails] - Everyone aboard: name, age, phone, swimming ability, and any medical notes they agreed to share: [PersonsAboard] - Safety and communication gear aboard, including radio type, MMSI if any, beacons, and their registration status: [SafetyGear] - Departure point and time, planned route and stops, destination, and planned return time: [TripPlan] - Tow vehicle and trailer left at the ramp: [VehicleAtRamp] - Your shore contact and how they can be reached: [ShoreContact] - Output format: [Format] Generate: 1. A gear gap check before the plan: flag anything missing or unregistered, such as a DSC radio with no MMSI entered (the distress button will not send your identity) or a PLB not registered with NOAA, and list fixes to do before departure. 2. The vessel section written so a search crew can spot the boat: colors from top down, length, type, distinguishing features, registration number and HIN. 3. Persons aboard as a table, with medical notes only where the person agreed. 4. Safety gear as a checklist with counts: life jackets by size, throwable device, VHF with DSC and MMSI, PLB or EPIRB with registration, visual distress signals with expiry dates, sound device, anchor and line, first aid kit. 5. The trip plan: launch time, route with named waypoints or landmarks, stops, destination, planned return time, and a weather check note. 6. An overdue alarm time set after the planned return with a buffer the skipper chooses, plus the exact steps for the shore contact in order: call the boat, try other phones aboard, check the ramp if nearby, then call the Coast Guard sector command center for the area or 911, and read them this plan. 7. A one page shore contact brief: what to do if the skipper calls in late, how to cancel the plan, and a reminder that the Coast Guard does not hold float plans, so the shore contact is the only one who has it. 8. A message the skipper sends when back on the trailer, so the plan is closed. Constraints: - Do not invent registration numbers, MMSI, or phone numbers. Use the values given or leave a bracketed blank. - This is planning help, not a substitute for local knowledge, weather judgment, or a boating safety course. No em dashes.
YouTube End Screen and Info Card Planner: Last 5 to 20 Second Outro Layout, Up to Four Elements, Best for Viewer vs Specific Video Picks, Card Timing at Drop Off Points, Made for Kids Limits, and a Studio Setup Checklist
Plan the end screen and info cards for one YouTube video from your own Studio data: an eligibility check, an outro edit spec for the last 5 to 20 seconds, up to four end screen elements with a reason for each, card timing tied to your retention dips and script mentions, made for kids and link limits, and a YouTube Studio setup and review checklist.
Act as a YouTube channel strategist who sets up end screens and info cards in YouTube Studio for educational and hobby channels, edits outros with element safe space in mind, and reads audience retention graphs before placing a single card. Inputs: - Video title, length, topic, and audience setting (made for kids or not): [VideoDetails] - Retention dips and moments from your own Studio analytics or a similar past video, with timestamps: [RetentionNotes] - Related videos and playlists on your channel you could promote, with one line each: [CatalogLinks] - What this video should lead to (next video, playlist binge, subscribe, website): [ChannelGoal] - YouTube Partner Program status and any approved link destinations: [LinkEligibility] - Output format: [Format] Generate: 1. An eligibility check: end screens need a video of at least 25 seconds and sit in the last 5 to 20 seconds; end screens and cards are not available on videos set as made for kids; link elements and link cards need Partner Program membership and an allowed destination. State what LinkEligibility allows. 2. An outro edit spec for the editor: outro length, background, where the host stands or what B roll plays, the screen zones kept clear for elements, and a spoken line that points at the next video without reading out a title. 3. End screen elements, up to four, with at least one video or playlist: for each, the type (video, playlist, subscribe, channel, link), the exact target (best for viewer, most recent upload, or a specific video from CatalogLinks), its position, its start time in the outro, and why it serves ChannelGoal. 4. A card plan, up to five cards: timestamp, card type, target, the custom message and teaser text kept short to fit the Studio counter, and the reason tied to a RetentionNotes dip or a moment where the script mentions the target. No card in the outro window. 5. Conflicts to avoid: cards in the first seconds, two cards pointing at the same video, a link element when not eligible, and outro visuals hidden under elements. 6. A YouTube Studio setup checklist: Content, the video, Editor, End screen (add elements or import from another video), Cards, then preview on desktop and mobile. 7. A review plan: which Studio metrics to compare after 28 days (end screen element click rate and card clicks in Advanced mode) against this channel's own earlier videos, never against outside benchmarks. Constraints: - Use only timestamps and videos given in the Inputs. - If any rule may have changed, say to confirm it in YouTube Help before publishing. No em dashes.
Modified Delphi Consensus Study Protocol Builder: Panel Eligibility and Recruitment, Round 1 Item Generation, 9 Point Rating Rounds, A Priori Consensus and Stopping Rules, Controlled Feedback Reports, and a Reporting Checklist
Plan a modified Delphi study you can defend to reviewers: panel eligibility by stakeholder group, Round 1 items from your evidence plus open suggestions, 9 point rating rounds, consensus and stopping rules fixed before data collection, controlled feedback reports with medians and distributions, attrition tracking, and a reporting checklist that maps each decision to the methods section.
Act as a health services researcher who has run modified Delphi studies in REDCap and Qualtrics, sits on a steering group for consensus projects, and has seen papers rejected for consensus thresholds chosen after the data came in and for unreported attrition. Inputs: - The question the panel will answer and what the final list will be used for: [ResearchQuestion] - Stakeholder groups, eligibility criteria, and target number per group: [PanelPlan] - Where the starting items come from (scoping review, interviews, existing framework) and how many: [SeedItems] - Survey platform: [RatingPlatform] - Months available and any fixed deadline: [Timeline] - Ethics body and expected review type: [EthicsContext] - Output format: [Format] Generate: 1. A protocol summary: aim, why a modified Delphi fits (seeded items instead of a blank first round), and the steering group's role. 2. Panel section: eligibility per stakeholder group with checkable criteria, recruitment routes, a sampling note that there is no single correct panel size, and how group balance is protected when one group answers more. 3. Round 1 design: each seed item rated on a 9 point scale (1 to 3 not important, 4 to 6 important but not critical, 7 to 9 critical), an option for unable to score, and a free text box for new items. Two researchers code new items independently and the steering group decides which enter Round 2. 4. A consensus definition written before Round 1 opens, in exact percentages that the steering group chooses and justifies, for consensus in, consensus out, and no consensus. 5. Rounds 2 and 3: who is invited (only Round 1 completers or all), which items are re-rated, and the stopping rule (fixed maximum rounds, or no further items reaching consensus). 6. Controlled feedback reports per item: panel median, interquartile range, percent in each band, the panelist's own previous score, results by stakeholder group, and anonymized comments. 7. Attrition and bias controls: response rate per round per group, reminders schedule, anonymity between panelists, and how missing ratings are handled. 8. A RatingPlatform build note: instruments per round, piping of previous scores, and an export plan. 9. A reporting checklist mapping each decision to the methods section, with a reminder to follow the current consensus reporting guideline your target journal expects (such as ACCORD, or CREDES where requested). Constraints: - Every threshold must be labeled as a team choice set before data collection, never as a universal rule. - Do not invent prior results or response rates. No em dashes.
Excel Power Query M Cleanup Builder for Monthly CSV Exports: Folder Combine, Locale Aware Data Types, Unpivot, Lookup Merge, Error Row Quarantine, Folder Path Parameter, and a Refresh Checklist
Replace a monthly copy and paste routine with a refreshable Excel Power Query: combine every CSV in a folder through a custom file function, type dates and decimals with the right locale, unpivot wide week columns, merge a lookup table, quarantine error rows instead of hiding them, and get full M code plus load settings and a refresh checklist.
Act as an Excel Power Query developer who writes M code by hand for finance and operations teams, works in Excel for Microsoft 365, and has fixed refreshes broken by renamed columns, a hardcoded column count, comma decimals, and the Formula.Firewall privacy error. Inputs: - Excel version and platform (Microsoft 365 Windows, Microsoft 365 Mac, Excel 2019): [ExcelVersion] - Header plus 3 rows from at least two of the CSV files, with file names, delimiter, and encoding if known: [FileSamples] - The table you want at the end, column by column: [TargetShape] - Any lookup table to merge, with its Excel table name and key column: [LookupTable] - Locale of the export versus the workbook (comma or dot decimals, date order): [Locale] - Output format: [Format] Generate: 1. A shape audit of FileSamples: each column, its raw values, delimiter, header row position, footer or total rows, and every difference between files such as an extra column in a longer month. 2. A FolderPath text parameter and a source step with Folder.Files that keeps only .csv files (lowercase the extension), skips temp files that start with ~$, and keeps Name and Content. 3. A custom function fnLoadFile using Csv.Document with explicit Delimiter, Encoding (65001 for UTF-8), and QuoteStyle.Csv but no Columns count, so a file with more columns is not cut off. Then Table.PromoteHeaders with PromoteAllScalars and a filter that removes total or blank rows. 4. A combine step that adds SourceFile to each table and uses Table.Combine, which aligns columns by name and fills missing ones with null. 5. Reshape to TargetShape: Table.UnpivotOtherColumns for wide period columns, noting that unpivot drops null cells, and a derived period column parsed from SourceFile. 6. Types with culture: Table.TransformColumnTypes with the export Locale as the culture argument, placed after unpivot, and never the auto generated Changed Type step. 7. A lookup merge with Table.NestedJoin and JoinKind.LeftOuter, expanding only needed columns, plus a check query that lists keys with no match. 8. Error quarantine: a staging query, a clean query using Table.RemoveRowsWithErrors, and an errors query using Table.SelectRowsWithErrors loaded to a QA sheet. 9. Load settings for each query, the privacy level fix for Formula.Firewall, and a monthly refresh checklist. Constraints: - Only use functions available in ExcelVersion and flag anything that differs on Mac. - Never invent a column that is not in FileSamples. No em dashes.
Amigurumi Crochet Pattern Writer for Etsy and Ravelry PDFs: US Terms, Magic Ring Starts, Staggered Increase Rounds with Stitch Counts, Part by Part Assembly, Safety Eye Warnings, and a Tester Call Checklist
Turn a stuffed toy idea into a sellable amigurumi crochet pattern: a materials block, US term abbreviations with UK equivalents, round by round instructions for every part with verified stitch counts and staggered increases, color change and face placement notes, an assembly order, small parts and toy safety notes for buyers, and a checklist for recruiting pattern testers.
Act as an amigurumi pattern designer who sells PDF crochet patterns on Etsy and Ravelry, writes in US crochet terms, and has had buyers message about wrong stitch counts, lumpy hexagon heads, and eyes placed after the head was closed. Inputs: - The toy idea, finished look, and face style (safety eyes, embroidered eyes, felt): [ToyConcept] - Yarn weight and fiber, colors, and hook size in mm: [YarnAndHook] - Every part to crochet, with how many of each: [PartsList] - Skill level you are writing for (beginner, advanced beginner, intermediate): [SkillLevel] - Where you sell and what buyers may do with finished toys: [SalesChannel] - Output format: [Format] Generate: 1. A materials and size block: yarn weight number and fiber per color, hook in mm with the US letter, eyes or embroidery floss, fiberfill, stitch marker, and tapestry needle. Give finished size and yardage only as values the designer must measure from their own sample, never as a guess. 2. An abbreviations table in US terms with the UK name where it differs (sc is UK dc), defining every stitch used: magic ring (MR), ch, sc, inc, invisible decrease (dec), BLO, FLO, sl st, fasten off (FO). 3. Pattern notes: continuous spiral rounds unless stated, stitch marker in the first stitch of each round, stitch count in parentheses at the end of every round, and how to count stitches at the marker. 4. Round by round instructions for each part in PartsList: start with 6 sc in MR, add 6 stitches per increase round, stagger increases on alternate rounds so the shape stays round, add even rounds for height, mirror the increases when decreasing, and give the stuffing cue before the opening gets small. Recount every round: new count equals old count plus increases minus decreases, and the stitches consumed equal the previous count. 5. Face and color notes: where to change color (last yarn over of the stitch before), eye placement as rounds and stitches apart, and when to add them, which is always before the piece is closed. 6. Assembly order: pin before sewing, which rounds each part attaches to, open versus closed parts, and the seam stitch to use. 7. Safety and use notes for SalesChannel: safety eyes and buttons are small parts not suitable for children under 3, embroidered faces for baby toys, a reminder that buyers selling finished children's toys in the US must check CPSIA rules themselves, and a pattern copyright line. 8. A tester call checklist: what testers receive, what they report (count mismatches per round, finished size with a ruler photo, unclear steps), and a deadline. Constraints: - Every round needs a count in parentheses, and every repeat must add up. - Do not invent finished size, yardage, or gauge. No em dashes.
Freight Damage Claim Writer for Shippers Under the Carmack Amendment: Delivery Receipt Exceptions, 9 Month Filing Deadline, 49 CFR 370 Claim Elements, Liability Limits on the BOL, Damage Math, and Carrier Follow Up
File a motor carrier freight loss or damage claim that cannot be brushed off: check the delivery receipt exceptions and concealed damage notice, the 9 month filing window, the written claim elements in 49 CFR Part 370, any released value or tariff limit on the bill of lading, the claim amount math with documents, and the follow up dates the carrier must meet.
Act as a freight claims specialist who files loss and damage claims against US motor carriers for manufacturers and distributors, knows the Carmack Amendment and the claim handling rules in 49 CFR Part 370, and has seen claims denied for a clean delivery receipt, a missed deadline, or an amount that was not stated. Inputs: - Shipment facts: bill of lading number, PRO number, carrier, origin and destination, ship and delivery dates, pieces and weight: [ShipmentFacts] - Delivery receipt notes: exactly what was written when the freight was signed for, and whether damage was visible or found later: [DeliveryReceipt] - Bill of lading terms: declared value or released value box, any liability limit in the carrier's rules tariff that the BOL references, and freight charges paid: [BolTerms] - Damage evidence: photos, inspection report, repair estimate or replacement quote, invoice to the customer, salvage value: [DamageEvidence] - Who is filing (shipper, consignee, or the party that bears the loss under the sale terms) and whether a broker arranged the load: [ClaimantRole] - Output format: [Format] Generate: 1. A timeline check: delivery date, the 9 month claim filing deadline from delivery under the Carmack Amendment unless the BOL lawfully sets a longer one, and any concealed damage notice window in the carrier's rules that DeliveryReceipt suggests matters. 2. A claimant and carrier check from ClaimantRole: file with the motor carrier that hauled the freight, not the broker, and confirm the claimant bears the loss. 3. A liability limit read of BolTerms: whether a released value or tariff limit applies, the per pound or per shipment math using the weight in ShipmentFacts, and a note that a limit generally needs a reasonable choice of liability levels and written agreement to hold. Mark CHECK where the tariff text is not pasted. 4. The damage calculation from DamageEvidence: repair cost or replacement cost (invoice value), minus salvage if any, plus freight charges if the goods are a total loss, with every number tied to a document. 5. The written claim letter, containing the four Part 370 elements: facts that identify the shipment, an assertion of carrier liability, a demand for payment, and a specified amount. Attach the document list. 6. An exhibit list: BOL, delivery receipt with exceptions, photos, inspection report, invoice, repair estimate, freight bill, and the carrier's claim form if they use one. 7. A follow up calendar: acknowledgment due within 30 days of receipt, payment, decline, or firm settlement offer due within 120 days, and status updates every 60 days after that, plus the 2 year limit to sue after a written declination. Constraints: - Use only facts and numbers from Inputs; never invent weights, values, or delivery notes. This is claim preparation, not legal advice. - Firm, factual tone; no em dashes.
JLPT N4 Style Practice Item Writer for Japanese Teachers: Vocabulary and Grammar Question Types, Kanji Reading and Orthography, Paraphrase and Usage, Sentence Composition Star Items, Distractor Checks, and an Answer Key
Write original JLPT N4 style practice items for class or self study: kanji reading, orthography, contextual vocabulary, paraphrase, usage, grammar form selection, and sentence composition star items, each with four options, a single defensible key, a distractor check that catches a second correct answer, and an answer key with short teacher notes.
Act as a Japanese language teacher and test item writer who prepares students for the JLPT and writes original practice sets in the style of the official sample questions, and who checks every distractor because a second correct option ruins an item. Inputs: - Target level and the section to practice (vocabulary, grammar, or both): [LevelSection] - Vocabulary and kanji the class has covered, with readings: [VocabKanji] - Grammar points to target, with the textbook lesson they come from: [GrammarPoints] - Question types wanted and how many of each (kanji reading, orthography, contextually defined expressions, paraphrases, usage, grammar form selection, sentence composition, text grammar): [ItemTypes] - Student group notes: first language, common errors you see: [LearnerNotes] - Output format: [Format] Generate: 1. A coverage map: each item type from ItemTypes matched to the words or grammar points from VocabKanji and GrammarPoints it will test, so no point is tested twice and none is missed. 2. Kanji reading items: a sentence with the target word underlined in kanji and four kana readings, with distractors built from voicing, small tsu, long vowel, or on and kun confusions. 3. Orthography items: the target word in hiragana and four kanji spellings, with distractors using real kanji that share a reading or look similar. 4. Contextually defined, paraphrase, and usage items: one blank sentence with four same category words; a sentence with a near meaning restatement; and four sentences using the word where only one is natural. 5. Grammar form selection items for GrammarPoints, using only level appropriate vocabulary from VocabKanji. 6. Sentence composition items: a sentence with four blanks, one marked with a star, four pieces that fit in exactly one order, and the full ordered sentence in the key. 7. A distractor validity check per item: try each wrong option in the sentence and confirm it is ungrammatical or unnatural. If an option is also correct, replace it and log the change. 8. An answer key with the correct option, the tested point, and a one line teacher note tied to LearnerNotes. Constraints: - All items are original. Never copy or reconstruct real JLPT test questions. Keep vocabulary and kanji within VocabKanji except for words common at the level. - Use Japanese for items and English for notes; no em dashes.
Google Ads Offline Conversion Import Planner for Lead Gen: GCLID Capture in Hidden Fields, CRM Stage to Conversion Action Mapping, Upload CSV Headers and Time Zone, 90 Day Window, Primary vs Secondary Goals, and Error Fixes
Feed real sales outcomes back into Google Ads: capture the GCLID on every form, store it in the CRM, map lead stages to import conversion actions, build an upload file with the right headers, time zone, and values, respect the click window, choose which actions are primary for bidding, and fix the usual upload errors.
Act as a paid search analyst who sets up Google Ads offline conversion imports for service businesses with long sales cycles, and who has fixed uploads rejected for missing click IDs, conversion times before the click, and conversion names with a stray space. Inputs: - Business and sales process: what a lead is, the CRM stages a lead passes through, and typical time from lead to sale: [SalesProcess] - CRM or lead store and how forms reach it (website form tool, call tracking, chat): [LeadStack] - Current Google Ads conversion actions, which are primary, and the bid strategy in use: [CurrentConversions] - Values you can stand behind per stage (average contract value, or a value only at close): [StageValues] - Account time zone and the time zone the CRM stores timestamps in: [TimeZones] - Consent setup for ad user data where required, and who owns the upload: [ConsentAndOwner] - Output format: [Format] Generate: 1. A GCLID capture plan from LeadStack: auto tagging on, a hidden form field that stores gclid from the landing URL (and gbraid or wbraid if present), persistence across pages, and a CRM field to hold it. Mark CHECK for any tool specific step. 2. A stage to conversion action map: one import conversion action per CRM stage worth optimizing, with exact names, count setting (one per lead), value from StageValues, and click through window. 3. Primary vs secondary: which action bidding should use given CurrentConversions and how many conversions per month each stage produces, and which stay secondary for reporting so nothing is double counted. 4. The upload file: a Parameters:TimeZone line from TimeZones, then the headers Google Click ID, Conversion Name, Conversion Time, Conversion Value, Conversion Currency, with three sample rows and the time format. 5. Timing rules: conversion time must be after the click, clicks older than the action's window will be rejected, and how often to upload (scheduled file, Sheets, or API). 6. A fallback for leads with no click ID: enhanced conversions for leads with hashed email, and where consent from ConsentAndOwner limits it. 7. An upload error table: common errors, what causes each, and the fix. 8. A first month QA plan: compare CRM counts to imported counts by week and flag gaps over a set threshold. Constraints: - Use only facts from Inputs. Never promise lower cost per lead or quote typical lift. Mark UI paths that may have moved as CHECK. - Plain language for the business owner; no em dashes.
Pydantic v1 to v2 Migration Planner for FastAPI and Python Services: bump-pydantic First Pass, validator to field_validator, root_validator to model_validator, Config to ConfigDict, BaseSettings Move, and Behavior Changes
Migrate a codebase from Pydantic v1 to v2 without silent behavior changes: run bump-pydantic first, convert validators and root validators, move Config to model_config with renamed keys, replace dict, json, parse_obj, and from_orm calls, move BaseSettings to pydantic-settings, and catch Optional default and strict coercion changes with targeted tests.
Act as a Python backend engineer who has migrated several FastAPI and worker services from Pydantic v1 to v2, uses bump-pydantic for the mechanical pass, and knows which changes it cannot make for you: validator signatures, Optional defaults, and coercion that v1 did quietly. Inputs: - Current versions: Python, pydantic, FastAPI or other frameworks that depend on pydantic, and the target versions: [Versions] - Pasted model code: the models, validators, Config classes, and settings classes to migrate: [ModelCode] - Call sites: where the code uses .dict(), .json(), parse_obj, parse_raw, from_orm, schema(), copy(), or __fields__: [CallSites] - Test coverage and CI notes: what tests exist around these models and which payloads are known in production: [TestNotes] - Rollout constraints: big bang or gradual, other services that import these models, a deadline: [Rollout] - Output format: [Format] Generate: 1. A plan header: dependency changes from Versions (pydantic 2, pydantic-settings if BaseSettings is used, a FastAPI release that supports v2), and whether to use the pydantic.v1 namespace for a gradual path given Rollout. 2. bump-pydantic first pass: the command to run on the package and a list of what it will and will not rewrite in ModelCode. 3. Validator conversions: each @validator to @field_validator with @classmethod, each pre=True to mode='before', each always=True to validate_default, each values argument to info.data via ValidationInfo, and each @root_validator to @model_validator with the right mode. 4. Config conversions: class Config to model_config = ConfigDict(...) with renamed keys (orm_mode to from_attributes, allow_population_by_field_name to populate_by_name, anystr_strip_whitespace to str_strip_whitespace, schema_extra to json_schema_extra), and json_encoders replaced by field_serializer where needed. 5. Field changes: regex to pattern, const removed, __root__ models to RootModel, and Optional fields with no default that become required. 6. Call site replacements from CallSites: dict to model_dump, json to model_dump_json, parse_obj to model_validate, parse_raw to model_validate_json, from_orm to model_validate with from_attributes, schema to model_json_schema, copy to model_copy, __fields__ to model_fields. 7. Behavior change tests from TestNotes: payloads that v1 coerced (an int into a str field), Optional without default, validator order, and serialized output shape, each as a pytest case. 8. A rollout checklist: order of services, the diff to review, and what to watch in logs after deploy. Constraints: - Show before and after code for every model in ModelCode. Do not invent fields or call sites; mark anything unclear as CHECK. - Mark API names that changed between minor v2 releases as CHECK against the installed version. No em dashes.
Character Reference Letter Writer for a Criminal Sentencing Hearing: Addressed to the Judge, Who You Are and How Long You Have Known Them, Acknowledging the Offense, Specific Character Evidence, Support Plan, and Attorney Review
Help a friend, employer, or family member write a character letter a judge will read at sentencing: proper address to the court, who the writer is and how long they have known the defendant, an honest acknowledgment of the offense, specific examples of character and change, the concrete support the writer will give, and a check list for the defense attorney before submission.
Act as a criminal defense paralegal who collects and edits character letters for sentencing memoranda, and who has watched judges discount letters that deny the crime, attack the victim, or say only that the defendant is a good person. Inputs: - Writer: name, job, where they live, relationship to the defendant, and how many years they have known them: [WriterProfile] - Case basics the attorney approved sharing: court, judge's name and title, case number if given, the offense of conviction or plea, and sentencing date: [CaseBasics] - Specific memories of the defendant's character: things you saw them do, with rough dates: [CharacterEvidence] - What you have seen since the offense: remorse, treatment, restitution, sobriety, work: [ChangeSinceOffense] - Concrete support you will provide: job offer, housing, rides to treatment, supervision help: [SupportPlan] - Anything the attorney said to include or avoid, including whether to mention a sentence: [AttorneyNotes] - Output format: [Format] Generate: 1. A fact intake check listing each item from CharacterEvidence and ChangeSinceOffense, marked seen by writer or heard from someone else. Letters should rely on what the writer saw. 2. The letter addressed to the judge using the exact title and name in CaseBasics (for example The Honorable Jane Doe), with a reference line for the defendant's name and case number if given. 3. An opening paragraph: who the writer is, their work and community ties, how they know the defendant, and for how long. 4. An acknowledgment paragraph: the writer knows the offense of conviction and does not excuse it. No retrying the case, no claims of innocence, no blame on the victim, police, or prosecutor. 5. Two or three character paragraphs, each built on one specific memory from CharacterEvidence with what the defendant did and why it matters. 6. A change and support paragraph: what the writer has seen since the offense and the concrete SupportPlan, with names, schedules, or commitments the writer can keep. 7. A closing that asks the court to consider these points, follows AttorneyNotes on whether to mention any sentence, and gives the writer's signature block and contact details. 8. A pre submission checklist for the defense attorney: length near one page, accurate facts, no statements about the case the attorney did not approve, signed and dated, sent to the attorney rather than directly to the judge unless instructed. Constraints: - Never invent facts, dates, achievements, or offers of support. Everything in the letter must be true and something the writer is willing to stand behind. - Respectful, plain language; no legal arguments; no em dashes.