Public research · Foundation only
Original data can be useful only when it is reproducible and private.
Eyebiss has not published statistics about customer bills, savings, provider quality, or KEEP / WAIT / SWITCH outcomes. This hub defines the publication boundary for future aggregated work.
Real customer data is not used by this workstream. No sample count, realized-rate distribution, savings claim, or statewide inference should be taken from the demo fixtures or private acceptance corpus.
Required safeguards
Every future result needs a versioned research packet.
Cohort
Defined dates, geography, product scope, inclusion, exclusion, and one household-level unit of analysis.
Privacy
Minimum sample and suppression thresholds, no household identification, and no narrow slices that enable re-identification.
Deduplication
Stable privacy-preserving rules so one household, account, bill, or resubmission is not counted repeatedly.
Reproducibility
Frozen query logic, source requirements, methodology version, and a result hash that another reviewer can reproduce.
Uncertainty
Denominators, missingness, intervals or sensitivity, and limitations shown beside the result.
Population boundary
Customer or qualifying-upload findings never described as representative of all Texas households without a valid design.
Neutrality
No cherry-picking favorable samples, providers, time windows, or outcomes for marketing.
Human gate
Privacy, evidence, legal, and publication review before release; failed gates leave the claim unpublished.
Possible future research families
- Advertised versus realized recurring cost, with compatible definitions.
- Bill-credit threshold exposure across qualifying observed months.
- NEEDS_INFORMATION reasons, with privacy-safe categories.
- TDU share of recurring bill cost by current tariff and supported territory.
- Time-of-use sensitivity using explicit, non-invented load shapes.
These are architecture targets, not findings. See the full internal safeguards in docs/public-research-methodology-v0.md.