gustavosantos/GTI_GERENCIADOR_FISCAL
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore: capture semantic pipeline baseline

Browse Source
This commit is contained in:
OpenCode
2026-09-21 15:52:17 -03:00
parent 9dcc263217
commit 6731eeca08
47 changed files with 3724 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
central/app/models.py
+125
View File
@@ -339,3 +339,128 @@ class FiscalContextEntry(Base):
reason: Mapped[str | None] = mapped_column(Text, nullable=True)
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
updated_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
# ================= AI semantic layer V1 =================
# The AI layer is NOT a fiscal authority: it only selects among official candidates and
# never writes to Firebird. Knowledge is versioned, provenance-tracked and never deleted.
class KbProduct(Base):
"""GLOBAL product knowledge: only facts inherent to the product (never tenant context)."""
__tablename__ = "kb_products"
id: Mapped[int] = mapped_column(primary_key=True)
fingerprint: Mapped[str] = mapped_column(String(64), unique=True, index=True)
gtin: Mapped[str | None] = mapped_column(String(16), unique=True, nullable=True, index=True)
identity_kind: Mapped[str] = mapped_column(String(16), default="FINGERPRINT") # GTIN | FINGERPRINT | SEMANTIC | UNKNOWN
normalized_description: Mapped[str] = mapped_column(Text, default="")
brand: Mapped[str | None] = mapped_column(Text, nullable=True)
unit: Mapped[str | None] = mapped_column(String(32), nullable=True)
package_quantity: Mapped[str | None] = mapped_column(String(32), nullable=True)
category: Mapped[str | None] = mapped_column(Text, nullable=True)
semantic_attributes: Mapped[dict] = mapped_column(JSON, default=dict)
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
updated_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now(), onupdate=func.now())
class KbKnowledge(Base):
"""Versioned knowledge rows. Unique per (product, key, value): conflicting values for the
same key are kept side by side and surfaced as KNOWLEDGE_CONFLICT, never majority-picked."""
__tablename__ = "kb_knowledge"
__table_args__ = (UniqueConstraint("kb_product_id", "knowledge_key", "value", name="uq_kb_knowledge_value"),)
id: Mapped[int] = mapped_column(primary_key=True)
kb_product_id: Mapped[int] = mapped_column(ForeignKey("kb_products.id"), index=True)
knowledge_key: Mapped[str] = mapped_column(String(48), index=True) # candidate_ncm | validated_ncm | candidate_cest | candidate_ex_tipi | ...
value: Mapped[str] = mapped_column(Text)
source: Mapped[str] = mapped_column(String(32)) # OFFICIAL_DATASET | ERP_OBSERVATION | OPERATOR | AI | AI_PLUS_OFFICIAL_VALIDATION
confidence: Mapped[float] = mapped_column(default=0.0)
evidence: Mapped[dict] = mapped_column(JSON, default=dict)
model: Mapped[str | None] = mapped_column(String(64), nullable=True)
prompt_version: Mapped[str | None] = mapped_column(String(32), nullable=True)
dataset_versions: Mapped[dict] = mapped_column(JSON, default=dict)
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
validated_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
class KbObservation(Base):
"""Raw per-tenant ERP observation of a KB product (evidence aggregation only)."""
__tablename__ = "kb_observations"
id: Mapped[int] = mapped_column(primary_key=True)
kb_product_id: Mapped[int] = mapped_column(ForeignKey("kb_products.id"), index=True)
tenant_id: Mapped[int] = mapped_column(ForeignKey("tenants.id"), index=True)
company_key: Mapped[str] = mapped_column(String(64), index=True)
erp_code: Mapped[str] = mapped_column(String(32))
database_label: Mapped[str] = mapped_column(String(64))
observed_fields: Mapped[dict] = mapped_column(JSON, default=dict)
source: Mapped[str] = mapped_column(String(32), default="ERP_OBSERVATION")
observed_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
class TenantProductContext(Base):
"""TENANT-specific product context: role, ST-withheld, operation profile, operator decisions.
Never promoted to global knowledge."""
__tablename__ = "tenant_product_context"
__table_args__ = (UniqueConstraint("tenant_id", "company_key", "erp_code", name="uq_tenant_product_context"),)
id: Mapped[int] = mapped_column(primary_key=True)
tenant_id: Mapped[int] = mapped_column(ForeignKey("tenants.id"), index=True)
company_key: Mapped[str] = mapped_column(String(64), index=True)
erp_code: Mapped[str] = mapped_column(String(32))
kb_product_id: Mapped[int | None] = mapped_column(ForeignKey("kb_products.id"), nullable=True)
product_role: Mapped[str | None] = mapped_column(String(16), nullable=True) # RESALE | OWN_PRODUCTION | INPUT | SERVICE | UNKNOWN
st_withheld: Mapped[str | None] = mapped_column(String(4), nullable=True) # Y | N — never AI-decided
operation_profile: Mapped[dict] = mapped_column(JSON, default=dict)
source: Mapped[str] = mapped_column(String(32), default="ERP_OBSERVATION")
confidence: Mapped[float] = mapped_column(default=0.0)
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
updated_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now(), onupdate=func.now())
class AiWorkItem(Base):
"""Deduplicated semantic work item (one knowledge need, many pending cells)."""
__tablename__ = "ai_work_items"
__table_args__ = (UniqueConstraint("dedup_key", name="uq_ai_work_item_dedup"),)
id: Mapped[int] = mapped_column(primary_key=True)
task_type: Mapped[str] = mapped_column(String(24), index=True) # INVALID_NCM | CEST | EX_TIPI | PRODUCT_ROLE
dedup_key: Mapped[str] = mapped_column(String(160))
question_refs: Mapped[list] = mapped_column(JSON, default=list)
payload: Mapped[dict] = mapped_column(JSON, default=dict)
status: Mapped[str] = mapped_column(String(32), default="PENDING", index=True) # PENDING | KB_EXACT_HIT | KB_FINGERPRINT_HIT | HIGH_CONFIDENCE_RETRIEVAL | SINGLE_OFFICIAL_CANDIDATE | LLM_SELECTION_REQUIRED | UNRESOLVED | UNRESOLVED_NO_OFFICIAL_MAPPING | CONFLICT | DONE | SKIPPED
kb_product_id: Mapped[int | None] = mapped_column(ForeignKey("kb_products.id"), nullable=True)
cells_covered: Mapped[int] = mapped_column(Integer, default=0)
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
updated_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now(), onupdate=func.now())
class AiRun(Base):
"""One provider execution over a batch of work items, with budget-guard evidence."""
__tablename__ = "ai_runs"
id: Mapped[int] = mapped_column(primary_key=True)
provider: Mapped[str] = mapped_column(String(64))
model: Mapped[str] = mapped_column(String(64))
prompt_version: Mapped[str] = mapped_column(String(32))
work_item_ids: Mapped[list] = mapped_column(JSON, default=list)
input_tokens: Mapped[int] = mapped_column(Integer, default=0)
output_tokens: Mapped[int] = mapped_column(Integer, default=0)
cost_usd: Mapped[float] = mapped_column(default=0.0)
status: Mapped[str] = mapped_column(String(16), default="RUN")
results: Mapped[list] = mapped_column(JSON, default=list)
budget_guard: Mapped[dict] = mapped_column(JSON, default=dict)
started_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
finished_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
class AiEvalCache(Base):
"""Persistent LLM result cache: identical (input_fingerprint, candidate_set_hash,
prompt_version, model, dataset_versions) re-execution makes ZERO new calls."""
__tablename__ = "ai_eval_cache"
__table_args__ = (UniqueConstraint("input_fingerprint", "candidate_set_hash", "prompt_version", "model", "dataset_version_ncm", name="uq_ai_eval_cache"),)
id: Mapped[int] = mapped_column(primary_key=True)
input_fingerprint: Mapped[str] = mapped_column(String(64), index=True)
candidate_set_hash: Mapped[str] = mapped_column(String(64))
prompt_version: Mapped[str] = mapped_column(String(32))
model: Mapped[str] = mapped_column(String(64))
dataset_version_ncm: Mapped[str] = mapped_column(String(64))
phase: Mapped[str] = mapped_column(String(16))
answer: Mapped[dict] = mapped_column(JSON)
gold_ncm: Mapped[str | None] = mapped_column(String(8), nullable=True)
created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())