Echo Oracle interrogates the brief, tests messages, simulates audience friction, audits creator fit and forecasts the media plan based on the budget the user actually enters.
The product does not promise virality. It turns campaign inputs into a transparent, testable decision: what to launch, what to fix, where to spend and what could fail.
The campaign is interrogated, stress-tested, planned and converted into a client-ready launch decision.
Click Generate Sample Report or complete the form and run the trial. The report will include the new evidence layer, assumptions panel, what-would-change-this-verdict section and client-ready action plan.
Messages are read structurally, in English and Romanian: concreteness (numbers, named moments, senses, action verbs), proof markers, emotional tension, and clich\u00e9 / AI-sounding phrasing. No campaign-specific keywords — the same rules apply to every brief.
Total = attention \u00d7.20 + clarity \u00d7.17 + trust \u00d7.20 + share \u00d7.13 + conversion \u00d7.15 + anti-cringe \u00d7.08 + anti-AI \u00d7.07. Every subscore is 0\u2013100 with visible drivers and a rewrite suggestion.
Overall = Campaign DNA \u00d7.18 + Message \u00d7.20 + Audience \u00d7.15 + Creator Fit \u00d7.13 + Risk Safety \u00d7.17 + Media Efficiency \u00d7.17. GO at \u226580, FIX at 60\u201379, DO NOT LAUNCH YET below — with guards: a high Proof Gap or low Risk Safety caps the verdict regardless of the average.
Rate card dated June 2026, EUR base, fixed planning FX (1 EUR \u2248 4.97 RON) and a disclosed \u221218% Romanian-market auction adjustment for RON campaigns. P10/P50/P90 re-run the entire forecast under pessimistic / base / optimistic benchmark scenarios. No conversion value \u2192 ROI is switched off, not invented.
Pasted comments are scanned for six objection patterns (price, proof, trust, timing, complexity, fatigue — RO + EN). Matches feed the Skeptic persona, the comment simulator and the pre-mortem, tagged \u201cUploaded-data-derived\u201d.
Pasted or CSV-loaded comments are segmented into units and analysed in-browser: bilingual sentiment with negation and intensifier handling, behavioral intent (Lavidge–Steiner ladder: purchase / recommend / inquire / complain / switch), aspect × sentiment key drivers, emerging language via Monroe–Colaresi–Quinn weighted log-odds, Language Style Matching (Pennebaker) and COM-B barrier mapping (Michie et al.). All proportions carry Wilson 95% intervals; below n = 80 results are labelled directional. Coverage is disclosed — unclassified units are counted, never guessed. Forecast uncertainty uses 2,000 Monte Carlo simulations. Corpora are cleaned before analysis — exact duplicates and too-short units are removed and the removal is disclosed; ~30 common emoji carry sentiment; custom aspects extend the measurement lexicon; every Key-Driver dot opens its verbatim receipts; the labeled dataset exports as CSV.
Everything runs locally in this browser; nothing is transmitted unless you submit the request form. Confidence is capped at Medium-High in this standalone version: High requires real listening data — that is the paid audit, by design, not a paywall trick.
A working session on your Oracle report, annotated line by line — where the verdict came from, and what would move it. Within 48 hours.
The full research treatment: your real listening data, real method, a document you can put in front of a board. Scope shaped on your case — we talk first.
Privacy by design: your campaign data lives in this browser. Sending this form transmits only your contact details, the chosen offer, the verdict summary above and — if checked — your brief. Nothing else.