Experiment Assignment and Exposure discrepancies

We have recently integrated Amplitude analytics and have more recently been testing Amplitude Experiments on our website. For our first test experiment, the ratio between assignment and exposure is roughly 2:1, so there is double the amount of assignments as there is exposures and we don’t know why that is?

We also checked in Pathfinder and could see the following problem: Sometimes exposures seem to occur before the assignment event, is that an issue?

We included the Amplitude browser and experiment sdks via the npm modules @amplitude/experiment-js-client and @amplitude/analytics-browser We then initialize the experiment sdk like so:

  private _experiment = Experiment.initializeWithAmplitudeAnalytics(
    environment.amplitudeDeploymentKey,
    { fetchOnStart: true }
  );

as well as the browser SDK:

amplitude.init(environment.amplitudeApiKey, { defaultTracking: true });

and here is the code that checks the variant with additional experiment.start() call just in case:

  hasVariantValue(
    experimentFlag: string,
    variantValue: string
  ): Promise<boolean> {
    return new Promise((resolve, reject) => {
      this.initExperiments().then(
        () => {
          const variant = this._experiment.variant(experimentFlag);
          resolve(variant.value === variantValue);
        },
        (err) => {
          reject(err);
        }
      );
    });
  }

  initExperiments(): Promise<boolean> {
    if (this._initialized) {
      return Promise.resolve(true);
    }

    return new Promise((resolve, reject) => {
      this._experiment.start().then(
        () => {
          this._initialized = true;
          resolve(true);
        },
        (err) => {
          reject(err);
        }
      );
    });
  }

and is then used like so:

const amplitude = AmplitudeService.getInstance();  
amplitude
    .hasVariantValue('alt-title-searchbox', 'treatment')
    .then((active) => {
      console.debug('[amplitude] split test active = ', active);
      if (active) { /* split test active */ }
     });

We do not set any additional device or user ids.