PaceCast Documentation

v1.32.1

PaceCast is an empirical ad forecasting platform and creative intelligence system. It combines calibrated conformal prediction intervals with real-world campaign data to guide budget pacing and creative production.

Last updated: October 2026

Welcome to the PaceCast developer and technical documentation. PaceCast replaces subjective guesswork with mathematically grounded performance forecasts. Rather than projecting a single optimistic estimate, PaceCast calculates empirical 90% prediction bounds, validates accuracy using Weighted Absolute Percentage Error (WAPE) against a 7-day seasonal benchmark, and flags ad creative fatigue before cost-per-acquisition (CPA) spikes.

Platform Architecture

PaceCast is built as a modular system designed for speed, isolation, and privacy. The platform ingests raw performance logs from major ad networks, validates data schema integrity, executes calibrated time-series modeling, and yields actionable creative directives.

Conformal Inference Engine

Computes split-conformal nonconformity quantiles on 5-fold rolling-origin backtests, providing mathematically verified 90% coverage without assuming Gaussian errors.

Attribution Lag Guard

Applies an automated 7-day attribution embargo cutoff before training to prevent immature conversion windows from distorting future trajectories.

Creative Fatigue Telemetry

Monitors hook drop-off and frequency degradation across 6 creative dimensions, generating concrete 15-second production briefs.

Isolated Tenant Enclaves

Every client workspace maintains strictly isolated ad logs, currency parameters, credit ledgers, and OAuth tokens protected by database-level row security.

Core Capabilities

PaceCast delivers three primary analytical deliverables on every analysis run:

  • Pace Line Forecast: A 30 or 60-day daily projection of spend, conversion volume, and blended CPA, flanked by conservative 10th percentile floor and 90th percentile ceiling trajectories.
  • Historical Validation Receipts: Direct WAPE error measurement scored against an unyielding 7-day Seasonal-Naive benchmark to verify that machine learning exceeds simple calendar repetition.
  • Evidence-Informed Creative Briefs: Qualitative breakdown of hook formats, visual pacing, and value proposition messaging correlated with conversion velocity.

Standard Analysis Workflow

Getting started requires four straightforward steps. Data can be supplied via CSV export or direct platform synchronization:

  1. Initialize Client Workspace:Create an isolated workspace configured with your client's reporting currency and operational timezone.
  2. Ingest Daily Ad Metrics: Export daily campaign logs from Meta Ads Manager or connect Google Ads via OAuth 2.0. A minimum of 90 days of historical daily records is required.
  3. Select Budget Scenario: Choose between Hold (run-rate extrapolation), Scale (budget adjustment with modeled diminishing returns), or Plan (custom flight schedules with promotional calendar dates).
  4. Review Output & Deploy Briefs: Inspect calibrated ranges, export executive PDF summaries, or fetch analysis artifacts via MCP in your AI agent environment.

Security & Privacy Guarantees

PaceCast is engineered under strict tenant isolation standards. Customer advertising data is never used to train generalized multi-tenant machine learning models without consent.

Security ControlTypeTechnical Implementation
Token Encryptionrequired
AES-256-GCM
Google Ads refresh tokens are encrypted at rest with authenticated AES-256-GCM using 12-byte initialization vectors and 16-byte authentication tags.
Attribution Embargorequired
7 Calendar Days
Automatically purges the trailing 7 days of ad logs from training data to avoid incomplete attribution bias.
Tenant Isolationrequired
PostgreSQL RLS
All database queries enforce row-level security based on the authenticated user ID and workspace ownership.

Developer Ecosystem

Developers can interact with PaceCast via standard HTTP endpoints or connect their AI assistants (ChatGPT, Claude, custom agents) using our Model Context Protocol server.

curl (Model Context Protocol SSE Handshake)
curl -X POST "https://pacecast.app/api/mcp" \
  -H "Authorization: Bearer <PACECAST_ACCESS_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list", "params": {}}'