Generate computer offers

The Computer example creates offers of computers for sale: laptops, desktops, mini PCs, all-in-ones, single-board computers and servers, each with its parts, price, stock and reviews.

tabalyst generate --example computer --rows 10000 -d computer-output/

Created: computer-output/computers-en.csv
Created: computer-output/computers-fr.csv
Created: computer-output/computers.meta.json
10,000 rows in 3.5s (2,864 rows/s, 3 processes)

computers-en.csv and computers-fr.csv describe the same 10,000 offers: only the labels change (Laptop and Ordinateur portable, Black and Noir). computers.meta.json records the request, the version of Tabalyst, the anomaly profile if any, and a SHA-256 digest of every resource file used, so the files can be reproduced. The default is 10 rows; --rows and --seed change the count and the data. The same seed gives the same files. The times come from one machine and are not a promise.

One file, with or without errors

-o writes a single CSV, in English unless you ask for French with --locales fr. With an anomaly profile, the file holds the errors and a .changes.csv file beside it lists every changed cell:

tabalyst generate --example computer --rows 30000 -o pcs.csv --anomaly-profile dirty_realistic

Created: pcs.csv
Created: pcs.changes.csv
30,000 rows in 9.1s (3,286 rows/s, 3 processes), 8,506 cells changed: generating rows 7.9s, injecting errors 1.2s

No clean copy, summary or metadata file is written with -o: use -d to get them. --locales en, --locales fr or --locales en,fr chooses the languages; with -d, both are written unless you choose one.

Columns, catalog and market

Option	What it chooses
--output-profile	The columns: compact (40, the default), standard (70) or extended (111).
--generation-profile	The kind of product the catalog favors: balanced (the default), consumer, business, gaming, creator, workstation, enterprise, mixed or portable_only.
--market	The market of the offers: ca (Canadian dollars, the default) or us (US dollars).
--unit-style	numeric (plain numbers, the default) or localized (64 GB in English, 64 Go in French).
--unit-columns	The measured columns written with units, for example memory_gb,display_size_in; it implies localized.

The three profiles nest: standard has every column of compact, and extended has every column of standard. The text values that name something (categories, seller types, warehouses, availability, promotions) are translated in the French file; technical codes such as wifi_7 or tpm_2_0 are the same in both languages.

Compact profile: 40 columns

Group	Columns
Identity (2)	product_id, sku
Catalog (8)	brand, product_family, model, generation, model_year, computer_type, form_factor, product_segment
Processor (5)	cpu_brand, cpu_family, cpu_model, cpu_cores, cpu_threads
Graphics (5)	gpu_brand, gpu_family, gpu_model, gpu_type, gpu_vram_gb
Memory and storage (4)	memory_type, memory_gb, storage_type, storage_gb
Display (4)	display_size_in, display_type, display_resolution, display_refresh_rate_hz
System (3)	operating_system, color, weight_kg
Offer (9)	currency, list_price, current_price, stock_quantity, stock_status, release_date, rating, review_count, market

product_id is unique in a file. sku identifies a configuration (the parts, the color, the operating system and the market), so rows with the same configuration share it.

Standard profile: 70 columns, 30 more

Group	Column	What it holds
Product	mpn	Manufacturer part number: MPN- and a digest. Like sku, it identifies a configuration.
Product	product_category	One of 7 categories: laptops, desktops, mini PCs, all-in-one computers, workstations, single-board computers, servers.
Product	product_subcategory	One of 17, such as consumer laptop, gaming laptop, business laptop, small form factor PC.
Processor	cpu_architecture	x86_64 or arm64.
Processor	cpu_max_clock_ghz	Maximum clock, from 2.4 to 5.6 GHz.
Processor	npu_tops	Power of the AI accelerator in TOPS; 0 without one.
Processor	ai_pc_class	npu_capable from 40 TOPS, otherwise none.
System	os_family	windows, macos, linux or none.
System	keyboard_layout	ca_multilingual for the ca market, us_english for us.
System	language	en_fr for the ca market, en for us.
Display	display_brightness_nits	Brightness, from 250 to 1,000 nits. Empty without a built-in display.
Display	display_touch	true or false; touch screens are rare.
Display	display_finish	matte or glossy. Empty without a built-in display.
Connectivity	wifi_standard	wifi_5, wifi_6e, wifi_7 or wifi_none.
Connectivity	bluetooth_version	5.0, 5.3, 5.4 or none.
Connectivity	ethernet_speed	1_gbps, 2.5_gbps, 10_gbps, 25_gbps or none.
Ports	usb_a_ports, usb_c_ports, hdmi_ports	Number of ports of each kind.
Power	battery_wh	Battery capacity, from 48 to 100 Wh. Only laptops have one; empty otherwise.
Power	power_adapter_w	Rating of the adapter or power supply, from 27 to 1,600 W.
Chassis	chassis_material	aluminum, steel, magnesium_alloy, aluminum_plastic, steel_plastic and a few more.
Camera and security	camera_type	none, 1080p, 1080p_ir or 1440p_ir.
Camera and security	biometric_type	none, fingerprint, face_recognition or a combination.
Offer	discount_percent	0, 5, 10 or 15: the discount that turns list_price into current_price.
Offer	price_tier	entry below a list price of 1,500 (1,125 in US dollars), otherwise premium.
Offer	condition	Always new.
Offer	warranty_period	12_months, 24_months or 36_months.
Lifecycle	product_age_months	Months between release_date and the reference date.
Lifecycle	product_lifecycle_status	current under 24 months, mature after.

Extended profile: 111 columns, 41 more

Group	Column	What it holds
Display	display_aspect_ratio	16:10, 16:9 or 3:2. Empty without a built-in display.
Display	display_hdr	none, hdr10, displayhdr_400, displayhdr_1000 or true_black_500. Empty without a built-in display.
Display	display_color_gamut	srgb_62, srgb_100 or dci_p3_100. Empty without a built-in display.
Processor	npu_brand, npu_family	The maker (Intel, AMD, Apple) and the family of the AI accelerator, such as intel_ai_boost. Empty without one.
Processor	copilot_plus_pc	true from 40 TOPS of AI accelerator, otherwise false.
Connectivity	cellular_type	none or 5g. Only laptops can have 5g.
Connectivity	docking_support	none, usb_c, thunderbolt or thunderbolt_4.
Connectivity	ethernet_port	true when ethernet_speed is not none.
Ports	thunderbolt_ports, displayport_ports	Number of ports of each kind.
Ports	card_reader_type	none, sd, sdxc, sd_uhsii, microsd or nano_sim.
Slots	memory_slots_total, memory_slots_used	Memory slots; the slots used never exceed the total.
Slots	m2_slots_total, m2_slots_used	M.2 storage slots, with the same rule.
Slots	upgradeability_level	none, low, medium, high or very_high.
Dimensions	width_mm, depth_mm, height_mm	Size of the machine, from 85 to 615, 56 to 720 and 9 to 470 mm.
Chassis	chassis_finish	The finish. In the current catalog it takes the same values as chassis_material.
Audio and input	microphone_type	none, dual_array, triple_array or quad_array.
Audio and input	speaker_configuration	none, internal_mono, stereo, quad_speaker or six_speaker.
Audio and input	keyboard_type	One of 8, such as backlit_chiclet, backlit_gaming, external_keyboard, wireless_keyboard or none.
Security	trusted_platform	tpm_2_0, apple_secure_enclave or none.
Promotion	promotion_type	No promotion, percentage discount, seasonal promotion or clearance. There is none when discount_percent is 0.
Promotion	promotion_status	Active when there is a discount, otherwise no promotion.
Price history	lowest_price_30d, highest_price_30d, average_price_30d	Computed from a short synthetic history: the current price and up to two earlier prices of the list price with a discount of 0 to 15%.
Price history	price_change_30d_percent	Change of the current price against the last earlier price of that history, in percent.
Lifecycle	product_generation	current under 12 months since release, previous under 36, legacy after.
Sales	popularity_level	high from 100 units sold in 30 days, medium from 30, otherwise low.
Sales	units_sold_30d	Units sold in 30 days: 0 to 5 when out of stock, otherwise about a third of review_count plus 0 to 60.
Sales	sales_rank	Rank derived from the units sold; the more units, the better (lower) the rank.
Sales	conversion_rate	Share of visits that end in a purchase, from 0.005 to 0.15; 0.005 when out of stock.
Sales	return_rate	Share of units returned, around 0.06 to 0.11; the better the rating, the lower.
Distribution	seller, seller_type	A seller of the market and its kind (retailer, distributor, refurbisher, authorized reseller, marketplace seller, manufacturer). A manufacturer’s own store only sells its brand.
Distribution	warehouse	A warehouse of the market, such as Canada East, Central or West.
Distribution	shipping_availability	Unavailable when out of stock, Limited availability under 5 units, otherwise Available.

The values follow rules of consistency, not market estimates:

• Each offer takes its parts from a catalog of brands, families, models and generations, and a model only has the processors, graphics, memory, storage, screens and operating systems that the catalog lists for it.
• A dedicated GPU has memory and an integrated one has none, and a processor never has fewer threads than cores.
• current_price never exceeds list_price. Prices are computed from the configuration (more memory, storage, cores or graphics memory cost more) and a discount of 0, 5, 10 or 15%; they are not real prices.
• stock_status is out_of_stock exactly when stock_quantity is 0.
• release_date falls in the model year and never after the reference date (--as-of-date, 30 September 2026 by default).
• rating is empty when review_count is 0.

Many rows share a model, since the catalog is finite: a large file repeats configurations, each row with its own product_id, stock and reviews, while sku and mpn repeat with the configuration.

Add defects on purpose

Computer has four anomaly profiles. They keep the clean file and write an altered copy, a change log and a summary (with -d), or only the altered file and its log (with -o).

Profile	What it does
dirty_realistic	A few defects on a few percent of the rows (table below).
dirty_stress	The same kinds of defect on a larger share of the rows.
unit_variants	Only the units: 16 becomes 16 GB in some memory, storage and graphics memory cells.
unit_variants_extended	The same for 14 measured columns (screen size, refresh rate, battery, weight, dimensions and more). It needs --output-profile extended.

dirty_realistic and dirty_stress change the same columns:

Column	Defect	dirty_realistic	dirty_stress
memory_gb	Emptied	2.5%	12%
cpu_model	Emptied	1.5%	8%
release_date	Written with slashes	2%	10%
current_price	Replaced by N/A	1%	6%
stock_status	Misspelled in_stok	1.5%	8%
gpu_vram_gb	Set to 0 on a dedicated GPU	4%	18%
stock_quantity	Set to 0 on an offer in stock	3%	15%
memory_gb	Followed by GB (16 GB)	8%	15%
storage_gb	Followed by GB	5%	10%
gpu_vram_gb	Followed by GB	3%	6%

The last two columns are the share of rows each defect is requested on. A defect with a condition (a dedicated GPU, an offer in stock) is applied only to the rows that meet it, and the summary counts the others as skipped. Since stock_quantity set to 0 while the status says in_stock contradicts the rule above, such offers are real defects a profile of the file should find.

An unknown profile name stops the command with the list of available names. To get several levels at once, name several profiles with -d: the rows are generated once and each profile makes its own variant.

tabalyst generate --example computer --rows 30000 -d out/ --locales en --anomaly-profile dirty_realistic,dirty_stress

Large files

From 5,000 rows, Computer generates the rows in several processes, one per spare processor and at most 6. --workers N sets the number, and --workers 1 keeps a single process. The files are the same whatever the number, so a seed still gives the same data. On one machine with four processors, 1,000,000 offers with an anomaly profile took about three minutes with three processes.

In a terminal, the command shows its progress on one line with the time left; --no-progress hides it. It always ends with a line giving the rows, the total time, the rate, the changed cells and where the time went.

Next steps

• Tabalyst Generate explains packages and how to write your own anomaly profile.
• The CLI reference lists every option.
• Generate CRM contacts and Generate insurance customers describe the other examples.