Every one of the six historical cases inside the HPI engine — SpaceX, Tesla, Bitcoin, NVIDIA, Shopify, OpenAI — eventually achieved significant recognition. None of them represent an entity that showed a similar early signal profile and then faded into irrelevance. This is survivorship bias, and it's a real, structural limitation of the current library that deserves direct acknowledgment rather than quiet omission.
Why this matters for how matches should be interpreted
If the library only contains successful outcomes, then by construction, every comparison an entity receives is against a pattern that worked out. There is currently no equivalent library of entities that showed similar signal shapes and did not go on to achieve recognition — which means the current matching process cannot, on its own, distinguish a genuinely promising pattern from one that merely resembles successful cases by coincidence.
What addressing this would require going forward
- Building a parallel library of entities that showed comparable early signal profiles and did not subsequently achieve recognition, to establish a genuine base rate
- Disclosing this limitation directly to institutional buyers as part of any serious commercial evaluation, rather than allowing it to be discovered independently
- Continuing to frame every HPI output around historical resemblance rather than predictive confidence, which remains accurate regardless of how the library eventually grows
Why disclosing this strengthens, not weakens, the product
A platform that proactively names its own methodological limitations is more credible to a sophisticated buyer than one that waits to be challenged on them. This disclosure is treated as a standing commitment, not a one-time caveat — every future presentation of the precedent library to a serious institutional audience will include it.