Where should deterministic logic stop and generative AI begin? · ZeroOne

Anuj Saxena

Aug 24 (edited) • Builds

Where should deterministic logic stop and generative AI begin?

I have been building 𝗧𝗿𝗮𝗱𝗶𝗻𝗴𝗘𝗱𝗴𝗲𝗜𝗤’𝘀 𝗣𝘂𝗯𝗹𝗶𝗰 𝗙𝗶𝗹𝗶𝗻𝗴𝘀 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻 with one important design principle:

𝘿𝙤 𝙣𝙤𝙩 𝙖𝙨𝙠 𝙖𝙣 𝙇𝙇𝙈 𝙩𝙤 𝙢𝙖𝙠𝙚 𝙩𝙝𝙚 𝙞𝙣𝙫𝙚𝙨𝙩𝙢𝙚𝙣𝙩 𝙙𝙚𝙘𝙞𝙨𝙞𝙤𝙣.

Instead, I separated the responsibilities.

❇️ A deterministic scoring engine evaluates the evidence.

❇️ An AI-assisted explanation layer makes the evidence understandable.

❇️ Product guardrails prevent the result from being presented as a price prediction or recommendation.

The CDNL case in this video shows why that distinction matters.

Seven insiders reported approximately $8.5 million in open-market purchases within 14 days. Richard Lee’s disclosure received an attention score of 87, including 10 points because five colleagues had already reported corroborating purchases.

Without that cluster evidence, the same disclosure would have scored 77.

The lesson is not that AI discovered a stock that will rise. The lesson is that a trustworthy system can distinguish an isolated disclosure from a pattern supported by multiple independent filings, explain exactly why the score changed, and preserve the evidence trail for review.

Claude and Codex have helped me challenge the specification, test edge cases, improve the explanations, and turn the underlying evidence into this case-study video. Generative AI was also used for illustrative scenes and content production, with the AI-generated material disclosed in the video.

For other builders here:

𝘞𝘩𝘦𝘳𝘦 𝘥𝘰 𝘺𝘰𝘶 𝘥𝘳𝘢𝘸 𝘵𝘩𝘦 𝘣𝘰𝘶𝘯𝘥𝘢𝘳𝘺 𝘣𝘦𝘵𝘸𝘦𝘦𝘯 𝘥𝘦𝘵𝘦𝘳𝘮𝘪𝘯𝘪𝘴𝘵𝘪𝘤 𝘭𝘰𝘨𝘪𝘤 𝘢𝘯𝘥 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘈𝘐 𝘪𝘯 𝘢 𝘱𝘳𝘰𝘥𝘶𝘤𝘵 𝘵𝘩𝘢𝘵 𝘶𝘴𝘦𝘳𝘴 𝘯𝘦𝘦𝘥 𝘵𝘰 𝘵𝘳𝘶𝘴𝘵?

1:52

0 comments

Where should deterministic logic stop and generative AI begin?