Intent
Normalize a business goal, budget, geography, language, category and outcome definition.
Advertising operations control plane
Own the decision record. Partner for the media machinery.
Dharmarth Ads is designed to sit above existing buying and selling infrastructure. The current MVP owns structured intent, deterministic eligibility, planning receipts and approval records—not inventory, auctions, identity, ad serving or media accounts.
Current plus target system
Eight modules around a tamper-evident decision history.
Intent, policy, route planning, business records and human approval are present in the MVP. Execution, normalized measurement, reconciliation and outcome learning remain gated target modules.
Policy
Run deterministic category, destination, budget and known-evidence gates.
Route
Rank eligible channel candidates while preserving reasons, exclusions and rule versions.
Approve
Record a separate human decision before any future material platform action.
Execute
Target state: send typed commands only through contracted, scoped and tested connectors.
Measure
Target state: normalize source events, quality and experiment evidence without hiding provenance.
Reconcile
Target state: separate media, partner cost, service fee, tax, credit and exception.
Learn
Target state: update priors only from permitted, mature and monitored outcome data.
Connector truth
A logo is not an integration.
The catalog describes candidate surfaces and the evidence each would require. No account, contract, credential, booking or inventory access is represented as configured.
RouteMinimum gateCurrent execution
Google AdsManager account, developer access and customer authorizationNot enabled
Display & Video 360DV360 contract, account seat and OAuth permissionsNot enabled
Meta advertisingApp review, business assets and account authorizationNot enabled
Amazon AdsApproved API or partner access and policy fitNot enabled
Retail mediaRetailer-specific commercial and technical agreementNot enabled
Direct publishersVerified property, seller rights and signed insertion orderNot enabled
AAMP / AdCPEmerging adapter interfaces; versioned validation and authority requiredNot enabled
Safe algorithmic boundary
Rules authorize. Models may advise later. People approve material spend.
Cold-start recommendations use declared inputs and deterministic scoring. There is no verified outcome dataset or production connector basis for claiming autonomous optimization today.
LayerAllowed nowRequired before greater automation
EligibilityVersioned deterministic rulesLegal review, policy ownership and regression evidence
RankingExplainable score over eligible routesOutcome labels, calibration, drift monitoring and challenger tests
ForecastingNo fabricated performance forecastComparable history, calibrated intervals and back-testing
ActivationHuman review record onlyContracted connector, least privilege, idempotency, approval and kill switch
OptimizationNot activeSpend caps, exploration ceiling, holdouts and rollback evidence
Test the boundary
A useful architecture starts with one real brief.
The planner shows current rules, candidate routes and missing gates without inventing inventory access or performance.
Create a planning receipt