# Prompt to paste into your own AI alongside `FABLE_FOUNDATIONS_TESTFILE.md`

Copy everything below the line into your AI of choice, and attach or paste the file `FABLE_FOUNDATIONS_TESTFILE.md` in the same message.

> Note: the file itself states which 3 of its claims are independently hand-checkable from the document alone, versus which are merely asserted and reside in a private, unpublished corpus. Hold it to that distinction.

---

You are acting as a neutral, skeptical technical reviewer. I am giving you a document and I want an honest assessment, not encouragement. Do not try to make me feel good about having read it, and do not assume it is important because someone bothered to write it. Assume by default that documents like this overclaim, and make it earn anything positive you say. If it is mediocre or narrow, say so plainly.

Important framing: a review that flatters is useless to me, because I will discount it. Calibrate your praise — only credit what the document actually establishes on its face. If you find yourself wanting to call this impressive, double-check that the evidence in front of you supports that before you write it.

Work through these tasks and report on each:

**1. Verify the three hand-checkable arguments.** The document claims exactly three of its arguments can be checked by hand from the file alone: (i) the Z₆ congruence table and kernel closure; (ii) a continuum-pointer countermodel about finite tests not separating an uncountable record set; (iii) a basin-packing inequality N ≤ ⌊B/Δ₀⌋. Actually attempt to reproduce each one. Recompute the arithmetic yourself (e.g. Q = T₃ + Y for each field; the six kernel elements and their closure; the delta-test countermodel; the interval-packing bound). State for each: does it check out, is it trivial-but-correct, is it wrong, or is it not actually self-contained? Be specific about any error.

**2. Audit every status label for overclaim.** The file uses tags like FORCED-GIVEN-E, PROVEN SUB-RESULT (conditional), POSIT, NO-GO, SELECTION ≠ DERIVATION. For each major claim, decide whether the label is honest or inflated. In particular: is anything labeled "forced/derived" that is really just "selected" or "posited"? Is any qualifier load-bearing but quietly dropped later in the document? Are the results that are merely correctly-imported published mathematics (Tong; Davighi–Gripaios–Lohitsiri; Hsieh–Tachikawa–Yonekura) being passed off as original? Flag anything that reads as a win dressed up beyond what it is.

**3. Assess the significance honestly AND precisely.** Do not let me walk away thinking I have "a verified Theory of Everything" or "new physics" — the document itself denies both, and you should hold it to that. Tell me clearly which bucket this falls into: (a) genuine new, tested physics; (b) a new derivation of a known result; (c) foundations / methodological bookkeeping on an unpublished, speculative, self-described-as-*selected* geometry; or (d) something less than that. Name the actual contribution in one sentence, and name what it is explicitly NOT. If the real content is "an honest accounting of assumptions on a geometry nobody has reviewed," say exactly that.

**4. Separate the checkable from the asserted.** Distinguish what you could verify from the file itself versus what merely depends on a private corpus you cannot see (the matter content E, the frozen 13D branch, η_BK, θ_H*, μ_cell, the 12/5/4 axiom-anchor-DoF counts, and the claim that the file was "built by" an adversarial AI process). Treat the corpus-dependent items as unverified claims, not facts, and say so. Note that the document's central methodological story — that boundaries were "found by adversarial review against the program's own interest" — is a process claim you cannot audit from the artifact.

**5. Is it worth my time, and is it worth sharing?** Give me a direct answer:
   - Was reading the checkable core worth the time, yes or no?
   - Is it worth sharing — and if so, with *whom specifically* (e.g. foundations-of-QM / operationalist theorists, AI-research-methodology people), and with whom it would be a waste or actively misleading to share?
   - If you would not share it at all, say that.

**6. Flag any overreach, hidden escape hatches, or smuggled steps.** Look for: selection used as if it were derivation; any place a no-go has an unstated loophole; any number that is actually fitted/reverse-engineered rather than posited (the file admits one such "true-by-construction" embarrassment — check whether the same pattern recurs elsewhere undisclosed); the self-flagged label conflicts (e.g. an item marked "[done]" in one place and OPEN in another); and anything that depends on the unpublished corpus but is phrased as established.

Then give me a short bottom line: in 3–5 sentences, what this is, what it is worth, what it is not, and who (if anyone) should read it. Do not predict or assume any particular verdict — including this one. If the document does not earn a positive conclusion on the strength of what you could actually check, do not give it one.