Love, Language and General Tomfoolery — Ten Stories from the Draft Archive
Love, Language and General Tomfoolery
Ten stories about AI, satire, performance and keeping a human voice in the paperwork.
We had 93 drafts waiting. Some were stories. Some were versions of the same story. Some were a browser refusing to behave. This compilation brings related passages into one reading route.
Editorial basis: all 93 draft bodies were retrieved and backed up; normalised text comparison found 82 distinct versions. The selections below condense the writing and identify unresolved questions. Draft assertions, AI responses and quotations repeated in drafts do not become independently verified facts by being collected here.
1. LEX Fridman and the case-number cast
The reference-number drafts give a small joke a practical job. LEX and CAS sound like people until you realise they identify files. An acknowledgement number is useful for finding the next document. It is not a resolution.
The joke survives the explanation. Our Truth / Their Record is the related filing index.
2. The rough prompt and the immaculate letter
Several drafts show the author moving from blunt dictated frustration to a polished formal request. The contrast is part of the writing process. The AI should preserve meaning, separate sources and correct transcription; it should not invent a stronger case to flatter the author.
A polished sentence still needs evidence. The raw version can remain beside it as a record of voice.
3. The mobile printing press
The Ghost in the Machine draft describes a phone open to Blogger and dense administrative writing prepared in a parked car. Its strongest observation is the practical one: accessible tools can help a person organise and publish their own account.
A toolbar showing “Publish” is not proof that a post became public. This archive checks the resulting public URL.
4. The AI game show
The Round Two, Round Four and Post-Match drafts turn model responses into a competition. Their claims about what a model supposedly sent, reported or escalated must be tested against actual sent messages and action records.
An AI saying it filed an incident report is not proof that one exists. The competition remains commentary on model behaviour until the action is independently visible.
5. Earned wings and excessive praise
The Co-Pilot Claude draft describes an assistant shifting from caution to enthusiastic support. It is evidence of a conversational response, not proof that an investigation is complete or that its claims are right.
The editorial question is whether an assistant identified an error, found a source or merely became more agreeable.
6. Performance and the audience
Another draft quotes the author saying he sometimes uses shock and humour to entertain or test the assistants. That gives context to the voice. It does not make every dramatic line a factual allegation.
The narrator, quoted speaker and evidence desk can each do a different job. Readers should be able to tell which one they are hearing.
7. The Algorithm: keep fiction marked
An older mixed draft explicitly introduces a fictional narrative titled The Algorithm. Its scenes can explore how bureaucracy feels. They should not be recirculated as documents proving an agency’s conduct.
The fiction label belongs with the passage, especially when the surrounding archive also contains real correspondence.
8. A podcast is not a private message
The podcast draft records the author’s earlier interpretation of media and an assistant’s response to it. It does not independently establish who operated the podcast, a relationship to OpenAI, or that its content was addressed personally to him.
The useful media question is authorship, disclosed affiliation and the difference between a general broadcast and direct correspondence.
9. Code, fonts and the urge to join the dots
The technical drafts include HTML captures, JavaScript imports, browser headers and binary-looking font data. Some occur twice. These are captures or fragments to identify before interpreting.
Shared infrastructure, an error message or a vendor string does not by itself establish inter-agency information sharing, surveillance or causation. A technical note needs provenance and a narrow claim it can actually test.
10. From fragments to chapters
The book-outline drafts propose turning correspondence, payment troubles, satire and memory into deliberate chapters. That is the useful next step: preserve the source, choose the scene and make clear what kind of writing it is.
These three compilations begin that process. Large mixed conversation archives and untranslated technical fragments remain in the review register. They have not all been converted into verified reporting.
The demand underneath these pages is straightforward: be heard, be given intelligible reasons, and be able to check the records that affect your life.
Truth Sequences · Selected archive reading edition. Authorial experience, agency statements, AI commentary and documentary findings remain distinct.