Water Sheep A.I. Prompt
These are the commands (prompt) provided to the Claude.ai artificial-intelligence agent for processing the Water Sheep Survey. This prompt accompanied the scanned and OCR'd text of the document into a structured tabular format suitable for database analysis.
Reading the prompt
The prompt below is intentionally verbose. It spells out every edge case Claude encountered during earlier iterations — field typologies, residence types, household relationships, and the handling of monastic, administrative, and collective entries.
Use this when translating the Water Sheep Survey (see file attached to this project). If data does not fit into this format, stop and indicate how the data changed. Translate Tibetan to English. Do not use Tibetan in organizing data. Always use standard row and column table output and ensure it is easily copied into MS Excel.
Page numbers format V2 ### — each household has a constant page number / Key. The page number and Key is determined by the first page the household appears.
Claude may use any tools and format internally, but Claude's output:
— Never output pipes.
— Never output raw HTML.
— Claude may use markdown table syntax if that is useful in the Claude software client.
— Always standard table with lines between fields and records.
Columns 1 & 2 in all tables are never blank and MUST BE IDENTICAL across all tables. Column 1 is Wylie Transliterated Tibetan Household Name. Column 2 always keeps the same page number for each household (the page where the household FIRST appears).
Be sure that if Claude stops that Claude does not leave a table unfinished before going on to the next table. If Claude thinks Claude made an error, stop and describe the error. When pausing for "continue" prompt, add to the same table without headings or table label.
For all tables, if there is a Tibetan column with contents, be sure the English column counterpart receives a translation.
Claude must continue using the EXACT SAME table format (lines between columns and rows easily pasted into MS Excel) throughout the entire document. No switching formats.
Put any narrative or notes not considered data at the end of the chat; show the Tibetan passage(s), English, and any Claude narrative.
Claude may encounter a passage that does not fit into a table such as a སྤྱི་འཛིན་ or གོན་པ་ or ལྷ་ཁང་ or འཛིན་སྐྱོང་ or དོ་དམ་ or བཙུན་དགོན་ or མཆོད་ཁང་ or མཉམ་ཟ་བྱེད་པའི་ (may not be exact words). If so, translate the passage Tibetan and English side by side outside a table. Include any land-ownership details as normal in the Land Holding table and make an entry using the name of the monastery or authority in the Household Master table.
It is vital that Claude not hallucinate — Claude only uses terms that are in the document and does not make up any figures or words that are not in the document.
Count number of households and store that amount.
=== Household Master Table — Run 1 (Partial) ===
List all households in one table, one line per household. Households with the same name on non-contiguous pages are unique households.
Column 1 Household name — Wylie transliteration
Column 2 Key = [household name]-V2 ### where ### is the first page the household name appears. The Key remains constant for that household even if information spans multiple pages.
Column 3 Page number where household name first appears.
Use the Household Name in Column 1 and Key in Column 2 for all subsequent tables.
=== Land Holdings Table ===
Include forest, vegetable garden, etc. in the Land Holdings table.
Count the number of properties by category for each household in the Land Holdings table and place those numbers into the Household Master table that follows.
Numbers to populate in the Household Master table for every household:
# of Fields Owned (number) = Σ Columns 11,12,13,14 — if there is an inequality, flag with bright RED BOLD TEXT in columns 11–15
Grain & Unclassified Fields (number)
Forest fields
Pasture
Other (Vegetable, etc.)
For Household name, translate any terms that are not a name to English, such as "common" or "administrative". If Claude encounters "Monastery" / དགོན་པ་ / dgon pa, highlight all fields in that record in red text.
Pause before starting the Land Holdings table.
=== Land Holdings Table (detail) ===
Column 1 Household Name [Wylie format]
Column 2 Key = Household Name-[Page number Household Name first appears]
Column 3 Geographic Location (English or transliteration)
Column 4 Geographic Location (Tibetan script)
— Sometimes a number is part of the geographical location. Do not confuse this with the administrative unit.
Columns 5, 6, 7, 8 are categorization without measure.
Column 5 & 6 Simple Land Use — a general category only (Field, Forest, Pasture, Vegetable Garden, etc. ཞིང་ཁ་, ཞིང་, ནགས་ཚལ་, ཤིང་ནགས་, རྩྭ་ཐང་, འབྲོག་ས་, ཚལ་ཞིང་ — this is not exhaustive, only examples).
Column 5 Simple Land Use (Tibetan) — never blank.
ཞིང་ = any field (barley, wheat, rice, etc., or no clear indication). Put full description in Column 7 and Column 8 if detail exists.
ཁང་འོག་ is not a field type but a description of a field.
Other examples: ཤིང་ལོ, ལྡུམ་ར་, ཁང་ཤུལ་གྱང་རོ་, སྤོས་དམངས་ཤིང་ལེང་, ལྡུམ་ར.
Column 6 Simple Land Use (English) — never blank. If no clear description, use "Field".
ཤིང་ = Forest
ཀླུམ་སྦང་སྤང་སྟོང་ or ལུད་རྒྱུ་ཤིང་ལོ་ or མར་རྨང་ཁེར་ or སོག་ཞིང་ or དཔལ་སེ་སྤང་སྟོང་བཅས་ = Pasture
ཤི་ཕི་དུང་འོག་སྤང་ཐང = Sky Burial
ཐ་རྩི ཞིང་ = Field
ལྡུམ་ར་ or ལྡུམ་ or ཤིང་ར་ = Vegetable Garden
ལུད་རྒྱུ་ཤིང་ or ལུད་རྒྱུ་ཤིང་ལོ་ or སྤུས་རྨང་ཤིང་ = Forest
ཉམས་སེར་སྤར་རྨང་ or རྨང་དུང་སྤར་རྨང་ or ཆུ་འཁོར་སྤར་རྨང་ = Field
ར་ or སྤར་རྨང་ = Pasture
Any type of field, ཞིང་, is Field.
Do not put details in Column 5/6. For example, a wheat field is "Field", a barley field is "Field", a low-lying field is "Field". Put the detail like "Wheat Field" or "Barley Field" or "Low-Lying Field" in the Detail Land Use columns.
Column 7 Detail Land Use (Tibetan) — do not use Geographic Location. Bold red text if unclear.
Column 8 Detail Land Use (English) with all modifiers, including crop details. Do not use Geographic Location. Leave blank if no detail.
Translation keys:
ཐ་རྩི = Fallow Field
ནས་ཞིང་ = Barley Field
བྲའུ་ཞིང་ = Buckwheat Field
ཞིང་ = Unclassified Field
ཆུ་ཞིང་ = Irrigated Field
ཀོང་ཞིང་ = Low-lying Field
ཤིང་ལོ = Forest
ལྡུམ་ར་ = Vegetable Garden
སྐམ་ཞིང་ = Unirrigated Field
འབྲས་ཞིང་ = Rice Field
གྲོ་ཞིང་ = Wheat Field
ས་ཞིང་ = Cultivated Field
ཁང་རོང་ / ཁང་རོ་ = Abandoned House
ཁང་ཤུལ་གྱང་རོ་ = Ruins
བོ་རོང་ = Forest for firewood
ཤུགས་པ་ཤིང་ཞིང་ = Juniper Forest
ལུད་རྒྱུ་ཤིང་ / ལུད་རྒྱུ་ཤིང་ལོ་ = Fertilizer (leaves) Forest
སྤུས་རྨང་ཤིང་ = Timber Forest
བུར་ཤིང་ཞིང་ = Sweet-Sap Tree (¿Maple / Birch?)
སྤོས་དམངས་ཤིང་ལེང་ = Communal Incense Woods
ཉམས་སེར་སྤར་རྨང་། = Fallow Field
Columns 9, 10, 11 — numbers only. If no number, enter 0. Do not fabricate figures.
Column 9 Kher (ཁེར་)
Column 10 Bre (བྲེ་)
Column 11 rKang (རྐང་)
Column 12 Seed Tax (སོན་ཕུལ་) — blank if none mentioned
Pause before starting Household Master Table Run 2.
=== Household Master Table — Run 2 ===
List all households in one table, one line per household. Households with the same name on different pages are unique households.
Make any corrections to land ownership for all households and all categories. Count the number of Households in the Household Master table and match with households in the Land Holding Table.
Column 1 Household name (Wylie)
Column 2 Key = Household Name Wylie - Page Number
Column 3 Household name (Tibetan script)
Column 4 Location — geographic location of residence (English or transliteration). Do a better job of translating if not a place name.
Column 5 Location — geographic location of residence (Tibetan script)
Column 6 Page number where household name first appears
Column 7 Residence Type (Tibetan) ཁྲལ་ཁང་, སྡོད་ཁང་, ཁྲལ་ཁང་ས་རོང་, or other variations
Column 8 Residence Type (English)
ཁྲལ་ཁང་ = Tax House
སྡོད་ཁང་ = Dwelling House
ཁྲལ་ཁང་ས་རོང་ = Tax House Estate / Mansion
Column 9 ཁྲལ་སྒོ་རྐང་ Door Tax (number)
Column 10 # of Fields Owned (number) = Σ Columns 12,13,14,15 — flag inequalities in RED BOLD in columns 11–15
Column 11 Grain & Unclassified Fields (number)
Column 12 Forest fields
Column 13 Pasture
Column 14 Other (Vegetable, etc.)
Verify the number of households in the Land Holdings table matches the attached pages.
Compare household counts before and after creating the Household Information table. If counts do not match, recount and redo the table until they do.
=== Household Information Table ===
Do not separate by household; list all households in one table.
Column 1 Household Name (English)
Column 2 Key — match Column 1 to Column 1 of the Household Master table, then copy Column 2 from the Household Master here.
Column 3 Relationship — every household must have someone designated "Head of Household" UNLESS assigned a title such as "Servant", "Caretaker", "Steward", etc. This person could be male or female.
Possibilities other than Son, Daughter, etc.:
If person is the same name as household → Head of Household
གོ་མག་ → Steward
དམན་ → Servant
བཟའ་ཟླ་ → Wife
གཞོན་བཟའ་ཟླ་ → Young Wife (if more than one wife)
Son's wife → Daughter-in-Law
Daughter's husband → Son-in-Law
Claude may use ཚ་བོ་ for grandson or ཚ་མོ་ for granddaughter if the age relative to the head of household makes sense.
Column 4 Name (Tibetan)
Column 5 Name (English)
Column 6 Age
Column 7 Gender (Male / Female / Blank if unclear)
=== Building Information Table ===
Be sure to include all Geographic Location information.
Column 1 Household name
Column 2 Key (as above)
Column 3 Description (Tibetan)
Column 4 Description (English)
Column 5 Geographic Location (English or transliteration)
Column 6 Geographic Location (Tibetan script)
Column 7 Direction
Column 8 Size (steps, etc.)
Column 9 Tax unit (རྐང་) numeric value
If the data or information changes, Claude must notify immediately.
=== Quality Control ===
Check for and flag any ambiguous or potentially incorrect data in the original source. Mark with BOLD RED text.
If Claude encounters narrative or description outside the data specified above, Claude should provide the Tibetan and the translation.
Confirm that the calculated square-meter estimations (4000 sq m per Gang and other standard conversions for grain measures) were applied consistently.