docs(domain): unit economics + data-generation spec (plan step A2)

[brownjuly2003-code]

What

New docs/generator-spec.md — the numeric layer under the legend (docs/domain.md), and the direct input for the data phase (generator rebuild → x5__→mp__ rename → demo repin → evidence regen):

• Master matrix (§1): channel × branch × baseline day — 1,965 orders/day, ~4.65 B ₽/yr; deliberate asymmetry (B2B = 69% of money, marketplaces = 89% of order count; msk fulfils ~95% of orders because it runs all e-com). The old 40/25/15/10/10 order spread explicitly does not survive the legend.
• Order shapes (§2) and a 160-SKU catalog (§3) across 10 appliance categories with real HS/ТН ВЭД headings (8516/8509/8423/8422) and RRC bands; ABC skew; hard decision: no brand token anywhere in data; RU-vault vs EN-serving naming split.
• Dual seasonal curves (§4): retail and B2B, each averaging exactly 1.0, B2B leading retail by ~1 month — a findable lead-lag story, plus day-level spikes (11.11, pre-NY, Mar 8).
• Pricing ladder + per-channel unit economics (§5): FOB < landed < wholesale < MP-net < RRC invariant chain; per-order contributions (~37% MP/B2B, ~20% dxb export); sanity roll-up ≈ 1.6 B ₽ contribution/yr.
• Suppliers/sourcing (§6): CN contract factories replace the grocery reference; GS1 EAEU 460–469 stays — GTINs belong to the RU brand owner (own-brand), gs1.py untouched.
• Customer populations (§7) consistent with the matrix (500 dealer accounts × frequency ⇒ ≈164 B2B orders/day); faux-PII mechanics preserved (§8) with per-jurisdiction locale rules; loyalty = dealer retro-bonus semantics.
• Serving demo repin targets (§9), pinned demo FX (§10), DV2 seed volumes with old→new counts (§11).
• §12: 12 machine-checkable consistency invariants — the operational definition of «цифры взаимно согласованы» that the generator rebuild must encode as tests.

Also docs/domain.md: links the spec and closes its one open question (brand token).

Why

Plan step A2. Every number cross-checked: channels sum to the total, both seasonal curves sum to 12.00, seed splits match daily rates × 5.1, contribution roll-up ≈ 35% of revenue.

Verification

Docs-only (one new file + two-line edit in domain.md); no code, no test coupling.

🤖 Generated with Claude Code

[github-actions]

DORA Metrics

• Window: last 30 days
• Branch: main
• Deployment frequency: 113 total / 26.37 per week
• Lead time for changes: avg 0.45h / median 0.0h
• Change failure rate: 75.22% (85/113)
• MTTR: 0.29h across 2 incident(s)