A calibrated replica of your RTB market, deployed directly inside your own infrastructure.
Standard columns any DSP already logs, plus two optional columns for improved latency and fraud-detection accuracy. Run the open-source validator on your machine before anything is shared.
CTR, CVR, LTV, floor price, conversion delay, win rate, and bid latency are fit to your data inside an air-gapped Docker container running in your own VPC.
Implement the Enricher and Strategy interfaces. Compete against calibrated archetypes and a ghost market. Measure the impact before committing.
Every calibration run trains a full suite from your log data. These set the hidden ground truths — your bidder sees a realistic market, not a synthetic one.
Auction mechanics, competitor behavior, KPI enforcement, and creative testing — all parameterized from your data.
First-price with floor visibility noise. Publishers adapt floors to fill-rate feedback over time.
Conversions scheduled via AFT delay model. Late attribution carries over across episode resets.
Define a creative with "+20% CTR" and measure the downstream impact without needing live impressions.
Episode ends if KPI miss exceeds a hard threshold after minimum spend to match real guard-rail behavior.
Bidders re-roll their personality each episode. Residual pressure is modeled by a ghost market ensemble trained on your win data.
Latency sampled from your Latency Twin. Configurable timeout thresholds and failure distributions.
BidOptic runs as an encrypted, network-disabled container inside your VPC. Nothing is transmitted.
--network none.Because the container is entirely air-gapped within your VPC, runs with --network none, and contains no telemetry, BidOptic does not qualify as a Data Processor under GDPR. Licensing is enforced offline via a hardware-locked license.bin token — no callback URL, no license server.
You load and run an encrypted container image inside your own VPC. No upload portal, no hosted service, no external API calls.
No upfront cost. Run BidOptic against your own logs for an agreed evaluation period, with a pre-agreed objective accuracy threshold governing conversion to a paid license.
Output is written as structured JSON and CSV to your local output directory — market_intelligence.json, latency_profile.json, simulation_history.csv.
Pin any calibrated parameter for scenario testing without requiring code changes.
BidOptic's models set hidden ground truths. Your pCTR, pCVR, and LTV models receive the same noisy signals as production.
Connect your actual agent — rule-based, RL, or hybrid. Stable-Baselines3 compatible out of the box.
Standard columns any DSP already logs. No bespoke exports.
bid_latency_ms trains the Latency Twin from real data. If absent, a lognormal distribution is synthesized and flagged in the audit output.click_timestamp enables click-to-conversion latency fraud checks in the local validator.If you have a win log, a click log, and a conversion log — you have what BidOptic needs.
Model quality is measured on a held-out time window. Coverage gaps are surfaced before training begins.
Temporal leakage detection, zero-delay artifact removal, and staleness warnings at 30 and 90 days are all handled before the first model trains.
The local validator flags bot-like request volume, click-fraud signatures (high CTR paired with low CVR), and metronomic request timing — before anything is shared with us.
The 11-column schema is the minimum contract. DSPs with richer logs (such as additional user signals, custom auction metadata, or multi-touch attribution) can work with us on a custom calibration build.
Run our open-source Python validator to verify your logs meet the ML requirements. Zero egress means your data never leaves your machine — only aggregate stats go into the receipt you send us.
View Schema Validator on GitHubBidOptic is in a closed Design Partner phase. Reach out directly for a technical conversation.
No pitch. Just a technical conversation.