Publication history: 2026-06-29

Details

• New /llms.txt context file for AI / answer engines (ChatGPT, Perplexity, Claude, Gemini), generated at build time from the same data layer as the site (papers.ts, predictions.ts, version.ts) so it never drifts. It carries the one-line summary, the “claims / does not claim” split, the document set with PDF links, the headline predictions, and an explicit disambiguation from the unrelated Brouwer–Lefschetz “topological fixed point theory” of mathematics.
• robots.txt now lists the major search and AI crawlers (Googlebot, Google-Extended, Bingbot, GPTBot, ChatGPT-User, OAI-SearchBot, PerplexityBot, ClaudeBot, anthropic-ai, …) explicitly as allowed; the site-wide meta description and keywords are reworded toward natural search queries (“derivation of the fine-structure constant”, “Standard Model from first principles”, “where does 137 come from”) and carry the same physics/mathematics disambiguation.
• Backlinks (the missing piece). The root README.md and the Zenodo deposit description (zenodo_description.html) previously contained no outbound links; both now link to fixpoint-theory.com and the source repository, and the README gains a citation block with the archived deposit DOI 10.5281/zenodo.20846087.
• Website / documentation infrastructure only — no theory, suite, ledger or status change.

Details

• New module v436 (OVERDET.FLOOR.02) — it hardens OVERDET.FLOOR.01 (v432) so the numerology verdict no longer rests on any subjective chance assignment. v432's nominal {\sim}10^{-10} multiplies the counting census by hand-chosen plausibilities (input-“8” independence 0.1^3, seed 0.012, kaon 0.15) a skeptic can contest; v436 shows the conclusion survives without them. itemize
• [E]assumption-minimal floor. The \alpha cubic-class census (v100 layer D) is pure counting: of N=94500 complexity-matched F_{U(1)} variants only one lands in the 4\times10^{-8} CODATA window — the TFPT instance, reproducing \ainv=137.0359992 — so p_\alpha\le1/94500=1.06\times10^{-5}\sim4.40\sigma with no subjective probability, prior or multi-observable grammar.
• [E]monotone concession ladder. -\log_{10}p from generous to adversarial: full scorecard+\alpha\sim30.7\to scorecard-only \sim25.8\to\alpha-census-only \sim4.98; each rung uses strictly less evidence and the “not chance (>4\sigma)” verdict holds at the most adversarial rung, so it never rests on one contestable piece. Stripping all of v432's subjective factors (\sim1.8\times10^{-6}) leaves the floor at 4.40\sigma — they only sharpen it.
• [C]honest 5\sigma gap. The conventional 5\sigma line (p=5.7\times10^{-7}) is crossed unconditionally by the census times exactly one further independent factor \le0.054 (the input-“8” four-fold over-determination, or one foreign witness) — stated, not hidden. [C]Firewall unchanged: bounds only “is it numerology?”, not the physics bridge (v187).

Details

• New module v435 (ALPHA.QUILLEN.PROGRESS.04) — it attacks the single remaining [O]left after v434 (residual (2), the cubic \alpha^3 Chern/Maxwell moment) and sharpens it; it does not close it. The target ALPHA.QUILLEN.EXACT.01 (v382) stays [O]; \ainv=137.0359992 stays [E]. itemize
• [E]\pi-power / metric-independence test. The three EM-closure coefficients carry \pi-powers exactly \{0,3,6\}, equal to their \cthree-orders: Maxwell k_0=1\to\pi^0, Calderón -2\cthree^3=-1/(256\pi^3)\to\pi^3, transport -8b_1\cthree^6=-41/(327680\pi^6)\to\pi^6. So the Maxwell a^3 is the unique \pi^0/\cthree^0 (metric-independent) term. A heat-kernel boundary coefficient always carries the (4\pi)^{-d/2} measure; a metric-independent term carries none — the signature of a topological term. The test thus partitions the closure into one topological level (a^3) + two heat-kernel boundary terms, confirming v342/v434's “a^3 is a level, not a Seeley–DeWitt curvature integral” from a test rather than a claim.
• [C]conditional level quantisation. If the a^3 term is the seam U(1) Chern–Simons/\eta boundary level (the EM1 reading, v48), large-gauge invariance quantises the level to \Z, so k_0=1 is the unit (minimal) level — an integer, not a tunable real. This converts “the coefficient happens to be 1” into “the coefficient is an integer level and 1 is the unit”.
• [O]not closed: the EM1 lemma itself — the from-first-principles Chern–Simons boundary proof that \partial_\tau\log Z_{\mathrm{Maxwell}}=a^3 — stays external math (type A, v384); the seam F-normalisation stays the one [C](residuals 1 & 3, v434); \ainv stays [E].