Roadmap

9 real operators from marketing, fintech, DTC, and brand-side roles visible on homepage, pricing, RoastIQ, BuyerLens, and About.

SSL Labs, Mozilla Observatory, SecurityHeaders.com — linkable from /security so any buyer can verify our posture on demand.

First full market delivered at 50+ ads per category. Automotive (Ford), Beauty (L'Oréal Paris), Entertainment (Netflix), Fashion (Nike), FMCG (Dove), Financial (GEICO), Food-Beverage (Pepsi + Coca-Cola), Health-Wellness (CVS), Retail (Amazon), Tech (HP).

Reading-path overlay, element breakdown (logo/product/CTA/price/disclaimer), and first-three-seconds placement analysis — added to Deep Dive PDFs for image creatives.

Postgres RPC closes the TOCTOU race on free-tier quota checks. Failed pipeline runs can be resumed from the history view without consuming a second quota slot.

describe-frames and driver-analysis-jobs now trigger automatically after video persist, matching the frontend pipeline. ReportModel exposes frameDescriptions + driverAnalysis.

Pro users receive a Sunday-morning digest summarizing the past week. The Profile page gains a line chart of composite scores over time.

Capture-demo, newsletter-signup, and analyze-demo now validate a Turnstile token before rate-limit and processing.

48 typed events across auth, study lifecycle, pipeline, report interactions, BuyerLens, and upgrade funnel. Server-side capture for subscription_activated / deactivated via the Stripe webhook.

Every RoastIQ now produces two PDFs: a diagnostic Deep Dive and a 12-page Executive Brief. Same data model, different surfaces.

Brand context settings per workspace (via /app/settings) and a campaign brief form at upload (brand name, objective, audience, key message). RoastIQ + BuyerLens both read brand_context when scoring, so synthetic panels are brand-aware rather than generic.

Jest tests (reserveRoastSlotRace) that fire 10 concurrent RPC calls and verify exactly 3 are allowed + 3 reserved rows created (free-tier) or all 10 allowed (pro). Gated behind TEST_LIVE_SUPABASE=1 so production data stays untouched.

L'Oréal, British Airways, Just Eat, Revolut, Nike, Dove, Vodafone, Tesco. One brand per UK category — covers Beauty, Travel, Food-Beverage, Financial, Fashion, FMCG, Telecom, Retail.

Same one-brand-per-category sprint as UK, in parallel. Closes Batch 1 coverage across 5 original markets.

Batch-score every benchmark ad through RoastIQ, then correlate scores against public-transparency signals (TikTok views, YouTube runs, Meta duration). Unlocks "RoastIQ predicts TikTok engagement at ρ=X" claim. 5-day job.

Public methodology doc covering the 44-column taxonomy, scoring engine (Vertex AI Gemini 2.5), benchmark construction, known limitations, and version history of the benchmark pool.

SDK upgrade + instrumentation refactor. No user-visible changes, better error-capture coverage.

Public Discord for early users — peer-to-peer support, product feedback, benchmark-preview previews. Linked from /community and the site footer.

11 standalone tools (ad-length, thumbnail grabber, hook-timer, copy-check, compress, resize, colors, brief, fatigue, ad-specs, benchmarks) indexed at /tools as a zero-auth top-of-funnel.

Redesign of the in-app report and workspace surfaces to match the current landing-page quality bar.

Current demos show strong creative (Coca-Cola RoastIQ, Solstice Skin BuyerLens). Adding two deliberately weak examples — a video that fails Beat-the-Skip and a static image that fails Brand Impact — so first-time visitors see the full score range and understand the product catches actual weakness.

Batch 1 plan includes 12 additional markets on top of the original 5. When we add them depends on demand — MENA markets are likely first given our distribution. 354 additional brands queued.

Site content in French (Morocco + France + Belgium + Senegal + Quebec) and Arabic RTL. Scoring output in user language.

Route creative from browser / Slack share / Shopify product page directly into RoastIQ. Considered only once core paying-customer volume justifies the build.

Score creatives from within Figma. Agency workflow. Evaluated against free-tools hub uptake.

/agents landing page + MCP server + docs for calling RoastIQ from agentic coding tools (Claude, Codex, Cursor). Long tail, not the priority.

Gamified creative testing with real humans via webcam eye-tracking and emotion detection. Separate auth, rewards, AI-calibration feedback loop. Big build, post-Batch-1.

A small XGBoost-class model trained on benchmark outcomes to calibrate Gemini's raw attribute outputs against real-world performance. First genuinely-proprietary ML asset. Unlocks "trained on X outcome-labeled ads" claim.