Loktak v2.2

Why FP&A leaders should care

Your team doesn't care about the pipeline.
They care whether they can trust the numbers.

Loktak is the reason InSightOS outputs are defensible. Here's what that means for the VP Finance presenting to a board, not the engineer building the pipeline.

"Can I trust the number?"

Every number InSightOS shows is produced by Loktak running a deterministic, rule-based pipeline on your actual source data. If someone asks where it came from, Loktak can trace it back to the specific source row, transformation rule, and timestamp that produced it.

"Finance and Sales have different numbers again."

Loktak resolves cross-system conflicts before InSightOS ever reads the data. When systems disagree, it applies your configured resolution rules, produces a single canonical number, and logs that logic so the alignment debate ends at the data layer.

"If auditors ask how we got to that forecast, what do we show them?"

Loktak writes immutable provenance for every value: source system, transformation applied, timestamp, pipeline version, and the rule used. InSightOS logs the decision on top of that, so the full chain from raw ledger row to approved reforecast is exportable as a structured audit trail.

"Our business rules keep changing. How does the system keep up?"

Your allocation logic, elimination methodology, and finance-specific heuristics live in Loktak's versioned knowledge library. Update a rule once and every downstream InSightOS output reflects the new logic, while the previous version stays preserved for historical comparison.

"What if the source data is wrong?"

Loktak runs signal integrity checks before analysis begins. Duplicate rows, missing periods, format inconsistencies, and out-of-range values are flagged before they can turn into a number you present to the board.

Architecture

Where Loktak sits in the PhrasIQ stack.

Loktak is the strict boundary between your raw data sources and the InSightOS execution layer. Nothing crosses that boundary without being resolved, normalised, versioned, and lineage-stamped first.

For the full component boundaries, API surface, and what each layer is and isn't allowed to touch, see the technical documentation.

Pipeline

Eight stages. Same sequence.
Every run.

Every Loktak pipeline run moves through the same eight stages in order, and each one has to clear a data-quality bar before the next begins. If something falls short, the pipeline halts. InSightOS never reasons on data that hasn't been verified.

Pipeline detail

Source connected · NetSuite

Raw data pulled from connected sources

Loktak connects to each configured source system via its native API or SFTP export. Format parsers normalise CSV, JSON, Parquet, and native ERP export formats into a staging schema. Schema validation runs against expected column types, required fields, and referential integrity rules before any further processing begins. A source that fails validation halts the pipeline. No partial data passes forward.

Formats: CSV · JSON · Parquet · XML · native API · SFTP · webhook

Validation: column type · null constraint · referential integrity · range bounds

Failure mode: pipeline halts · source error logged · alert raised

For stage-by-stage mechanics, validation thresholds, failure behavior, and a full pipeline run log, see the technical documentation.

Knowledge layer

Your finance logic, encoded once,
applied everywhere and versioned forever.

Your organisation's specific finance intelligence lives not in a spreadsheet, not in someone's head, and not in undocumented model logic. It is encoded in Loktak, applied to every pipeline run, and versioned permanently.

Business rules, not a black box

Allocation, elimination, and adjustment logic, made explicit

Cost allocation methodologies, intercompany eliminations, management reporting adjustments, and non-GAAP metric definitions are written down as rules Loktak can run, not buried in a spreadsheet macro or a model nobody can inspect.

Versioned forever

Update a rule without rewriting history

When a rule changes, the new version applies going forward while every past pipeline run keeps the rule version that actually produced it. A November-to-October reforecast comparison stays apples-to-apples, automatically.

The compounding advantage

Every approval, override, and manual rule edit feeds back into the knowledge layer, so Loktak gets more aligned with how your finance team actually thinks the longer you use it. Override rates drop, outputs get more defensible, and the rule library becomes something a competitor can't just copy by training a better model — it's built from your decisions, not data anyone else has.

Rule DSL syntax, YAML examples, LoRA-based encoding, and feedback-loop mechanics are in the technical documentation.

Lineage & trust

Every number is traceable.
Permanently.

Every value InSightOS shows links back through Loktak's append-only ledger to the source row it came from, every transformation applied along the way, and the decision it informed. Nothing in that chain can be modified or deleted, not even by Loktak itself.

Transformation steps recorded in Loktak's lineage ledger (append-only)Decision executed in InSightOS

Immutable audit trail

Every transformation is appended to a tamper-evident ledger the moment it happens. No user, process, or agent, including Loktak itself, can modify or delete a record once it's written.

Source → transformation → decision

The same ledger that records Loktak's transformations also records the InSightOS decision that used the result, so the full chain from raw row to approved action lives in one place.

Answers, not investigations

"Where did this number come from" is a lookup, not a project. Full chains export in one click for auditors, board decks, or a skeptical Sales VP.

lineage_for() query examples, comparative lineage between pipeline runs, and decision-layer integration details are in the technical documentation.

Trust & compliance architecture

Built for the compliance
finance operates under.

SOX compliance, data residency, role-based access, and audit readiness are design constraints that shaped Loktak's architecture from the beginning, not features added on top.

SOX-compatible

AES-256 at rest

TLS 1.3 in transit

GDPR-ready

Private VPC

Append-only ledger

RBAC · Object level

7-yr retention default

Human approval gate

SOX-compatible audit trail

The lineage ledger and decision log together produce a complete SOX-compatible audit trail. Every transformation, business rule, analyst decision, and approval is timestamped, attributed, and tamper-evident. Exportable as machine-readable report or human-readable PDF for external auditors.

Private VPC deployment

Loktak can be deployed entirely within your private VPC or on-premise infrastructure. Your financial data never leaves your environment. Full residency controls for GDPR, SOX, and sector-specific requirements.

RBAC at object level

Granular access control at the account, entity, period, and metric level. A salesperson sees pipeline metrics but not COGS. An analyst sees their region's P&L but not the consolidated group. Policies encoded in Loktak, enforced across all InSightOS outputs.

Encryption at rest and in transit

All data at rest encrypted with AES-256. All data in transit uses TLS 1.3. Keys managed via your configured KMS, including AWS KMS, Azure Key Vault, GCP Cloud KMS, or self-hosted HashiCorp Vault.

Human approval gate

No autonomous changes to source data, financial models, or downstream systems without explicit human approval. Loktak processes and InSightOS recommends, but every downstream action requires a named, authenticated user. The approval is logged with approver identity and data version.

Configurable data retention

Retention policies configurable per data category. Lineage records: 7 years default. Raw source snapshots: 13 months. Canonical object versions: 7 years. Regulatory hold policies can override. All retention operations logged.

Compliance requirement	Typical AI finance tool	Loktak
SOX audit trail	No structured audit log	Full transformation + decision chain, structured export
Data lineage to source	Output only, no source trace	Row-level provenance on every value, queryable <100ms
Private deployment	SaaS only or limited VPC	Full VPC and on-premise options, no data egress required
Rule versioning	Model is a black box	Every business rule versioned, pinnable, diffable
Human approval gate	Autonomous or unclear	No downstream action without named human approval
Tamper-evident storage	Logs can be modified	Append-only ledger with SHA-256 hash chain
RBAC at object level	Page or role-level only	Account, entity, period, and metric level access control
Configurable retention	Vendor-determined retention	Configurable per category, regulatory hold supported

Integrations & connectivity

Connect to any API
in under a minute.

Pre-built connectors handle authentication, pagination, incremental sync, and schema mapping for every major ERP, CRM, and data warehouse. Custom sources connect via the Loktak REST API and are mapped to the canonical finance schema automatically. No engineering sprint required.