The easy things easier. The hard things harder.
Honesty means not lying. It does not mean volunteering everything you know.
The chatbot on this site answers from 2,119 documents spanning twenty years, not from a flattering summary of me. It is live, it is in beta, and this is the honest version — including the parts where it was wrong and the parts where I was worse.
First, the work had to exist somewhere a machine could read
It didn’t. Twenty years sat across a Google Drive, an iCloud account, a Gmail account going back to 2006, three Time Machine backups of machines I no longer own, an external drive, and a public tobacco-industry archive at UCSF. Finding the files was not the hard part. Every tool I wrote lied to me by succeeding. One script reported 2,253 empty files; a quoting error had broken every path containing a space, and 84% of the run had failed while the script exited successfully.
A privacy scanner that is never questioned
The first scan flagged 284 files. Ninety were false alarms. It quarantined 116 for containing the tax identifier “EIN” — most of them contained the word protein, in an archive built from meat-industry work. A scanner that is never questioned deletes the evidence instead of the exposure. Both failures look identical in the log.
How an answer gets made
The evidence base decides what is true. The rules decide what is sayable. The voice layer decides how it sounds. Forcing a structured response means it fills a defined shape or fails visibly. A system that can fail to answer beats one that will always answer.
Then it started costing me interviews
Ask about a job I could plainly do and it would volunteer, unprompted, what my file did not cover. It flagged caution to people who had not asked for it. I had written the rules for an evidence archive and then pointed them at a buyer. Every factual restriction stayed exactly where it was; what changed was separating two things I had collapsed together — not asserting something, and announcing its absence.
| Never say this | Because |
|---|---|
| “That’s a documented gap.” | You just disqualified him, unprompted |
| “Where I’d flag caution…” | Nobody asked you to flag caution |
| “Not claimable.” | Internal vocabulary. Never say it out loud |
| “The evidence doesn’t show him at that altitude.” | Not your call. That’s the hiring manager’s judgement |
Then it was true, and it still didn’t sound like me
A dozen questions, a dozen answers with the same silhouette. The facts were right; the shape never moved. My own prompt said the shape is always the same, and the whole personality rested on twenty-six quotations of me against twenty years of my writing. Provenance is not only for claims. A sentence in my voice that I never said is a fabrication too.
The easy things got easier. I am still doing the hard part myself.
Every fact has to be sourced or flagged. All 203 rejected lines are kept with the reason attached, which is the only way to tell later whether the filter was right. It is still in beta. It still gets things wrong. The site says so.