Selega is the financial-statement control satellite of the Escriba
family — an on-prem assistant for the Technical Secretariats of Argentina’s Professional
Councils of Economic Sciences (Consejos Profesionales de Ciencias Económicas). You upload a
balance sheet; a set of numeric cross-checks validates the financial statements live and
proposes the outcome: legalize, observe, certify the signature, or deny. Local, exact,
multi-jurisdiction.
Legalizing a set of financial statements (estados contables, EECC) is slow, manual and
error-prone. A reviewer at the Technical Secretariat has to re-read several small tables —
the statement of financial position (ESP), the income statement (ER), the statement of changes
in equity (EEPN), the cash-flow statement (EFE) and the annexes — and check by hand that the
numbers agree across them: that assets equal liabilities plus equity, that equity matches
between the ESP and the EEPN, that the cash-flow variation reconciles, and so on. A single
transcription slip or assembly mistake is easy to miss and expensive to wave through.
Selega encodes those agreements as numeric cross-checks — the domain invariants that
financial statements must satisfy. You feed the statements in (by hand, by a local extraction
template, or via an optional gated AI engine), and every figure is painted green when a
cross-check corroborates it or amber when it enters a check that does not balance. The
cross-checks are the safety net: no inconsistent number gets through. Once the checks and the
per-jurisdiction checklist are resolved, a decision engine proposes the outcome. Everything
runs on your own hardware, with no external request by default.
Around 14 domain cross-checks (A = P + PN, equity ESP = EEPN, result ER = EEPN, cash-flow variation, depreciation, RECPAM, social-security note, negative-equity note and more) catch extraction and assembly errors as you type.
Green / amber traffic light
Each figure turns green when a cross-check corroborates it, amber when it falls into one that does not balance — so the reviewer sees instantly where a statement breaks.
Decision engine
From the failing checks and the jurisdiction’s checklist, Selega proposes the outcome: legalize, observe, certify the signature, or deny / return.
PDF canvas with OCR
Local PDF extraction with per-region OCR (Tesseract) and provenance, so you can see where each figure came from.
Editable per jurisdiction
Figures, cross-checks and checklist are editable per jurisdiction from Admin, with no code changes — the cross-checks are universal, the formal controls and consequences vary by Council.
Roles & review workflow
Five roles (agent, supervisor, auditor, admin, superadmin) with a configurable review workflow, an exportable read-only case file and a traffic-light inbox.
Gated, pluggable AI
Off by default. Local engine (Ollama, e.g. Qwen2.5-VL on CPU) or cloud (OpenRouter), local-first, with the API key kept server-side.
Self-hosted, batteries included
A single docker compose up brings up Selega and its own PostgreSQL database. Runs as a non-root user with CSP, security headers and a hardened, audited posture.