{
 "kind": "fall-spore-prereg",
 "v": 1,
 "written": "2026-10-05",
 "approvedBy": "Simon, relayed: build the SEED-CELL / SPORE organ (step 3 of 4). Spec, verbatim: transmit the SEED/DNA (tiny) not the grown build (big); the femtoLLM + compressed genome is a cell that, on arrival, reads its DNA and grows the build locally; seal and measure the real claims (seed-smaller-than-forest ratio, the grown build works on held-out, it germinates offline, differentiation), report whichever way each lands, and say plainly what is a femtoLLM vs deterministic assembly.",
 "statement": "Sealed, committed and pushed before any organ is germinated on the real sessions for grading. This fixes, in advance: the sessions, the ribosome (spore.mjs), the germination engine (organ.mjs), the runner, the spore shipped, the grow-config, the contexts, the five seeds, the rules and a prediction for each. The result is published whichever way it lands.",
 "question": "A 32-byte spore carries a stem genome and a grow-config — no built organ. On arrival a universal ribosome germinates it on the client's own sessions. Does the grown organ (a) weigh far less as a spore than as a grown body, (b) predict buying on two months it never grew on, (c) germinate with no network at all, and (d) grow differently, and better, for the context it is grown in?",
 "honesty": "Germination is DETERMINISTIC ASSEMBLY. The ribosome is a decoder plus the pattern-organs miner/evolver (grow, then fit) — it is NOT a language model, and no femtoLLM runs inside anything graded here. A femtoLLM (a real small model) is an optional, non-load-bearing soft step for sensing which context to express; it is kept out of this kernel and out of these rules. This mirrors pattern-organs: no language model is used in the measured build.",
 "sealed": {
  "data/online_shoppers_intention.csv": "b3055ee355f59134d851d32641183cb4a8b45def7124d2f50442a042f358e0d9",
  "organ.mjs": "d21b07da4a74fb9ee674446b17ee9a957eb05acfc136ffc6ff135b317fb724d2",
  "spore.mjs": "e4386af430637e6f838027f7c72793c670515b01eaf52f827212a6da450cd769",
  "tools/run.mjs": "a3c82a67e90bae607a4a778d67fa0a8e0362eec03a0d827b4d62402f59d7f2f7"
 },
 "sessions": {
  "source": "Online Shoppers Purchasing Intention Dataset — C. Sakar and Y. Kastro, UCI Machine Learning Repository (2018), DOI 10.24432/C5F88Q, CC BY 4.0; introduced in Sakar, Polat, Katircioglu and Kastro, \"Real-time prediction of online shoppers' purchasing intention using multilayer perceptron and LSTM recurrent neural networks\", Neural Computing & Applications (2019). 12,330 sessions of one online shop over a year, 1,908 of which bought.",
  "split": "by month: February–October train, November and December held out. No organ ever grows on a held-out session; each grown organ is graded on them once.",
  "contexts": "all visitors, returning visitors (Returning_Visitor), new visitors (New_Visitor) — the sub-populations a spore can differentiate into."
 },
 "spore": {
  "bytes": 32,
  "example": "◊ spore.v1 funnel 11 g:12.2.10",
  "format": "\"◊ spore.v1 <recipe> <seed> g:<population>.<elites>.<generations>\" — a konomi-family tag, the recipe the ribosome holds, the seed, and the grow-config. The genome it grows from is the STEM; a fixed-genome spore instead carries f:<base64url of 8 packed bytes> (58 bits: 8 bytes).",
  "recipe": "funnel — the universal ribosome holds the stem genome and the context map; the spore names it and ships the seed and grow-config. Powered by the Konomi architecture, created by Thomas Frumkin.",
  "stem": {
   "on": 127,
   "wiring": 1,
   "features": 2097151,
   "bins": 4,
   "book": 4,
   "minSupport": 2,
   "minLift": 2,
   "smooth": 3,
   "pairK": 3,
   "proofZ": 2,
   "halfLife": 3,
   "ownMin": 2
  },
  "growConfig": {
   "population": 12,
   "elites": 2,
   "generations": 10
  },
  "germination": "decode the spore → grow (differentiate) from the stem on the context's own training sessions for 10 generations of 12 → fit the champion on all that context's training sessions → the grown organ (its books). The \"grown body\" measured is that organ's books serialized as JSON."
 },
 "seeds": [
  11,
  12,
  13,
  14,
  15
 ],
 "grade": "Held-out AUC (ties half) of each grown organ on its context's November–December sessions. Differentiation: for each of returning and new, the organ grown for that context (matched) against the organ grown for the OTHER context (crossed), both graded on the same held-out sessions; a paired bootstrap of matched − crossed, 1000 resamples. The ratio: the spore's bytes against the grown body's bytes. Medians are over the 5 seeds.",
 "rules": [
  {
   "id": "seed-smaller-than-forest",
   "rule": "the grown body's bytes are at least 8× the spore's bytes, at the median seed"
  },
  {
   "id": "grown-works-held-out",
   "rule": "the grown organ's median held-out AUC over the seeds is above 0.5 (it predicts buying on months it never grew on)"
  },
  {
   "id": "germinates-offline",
   "rule": "every seed germinates with fetch, XMLHttpRequest and WebSocket trapped, touching none"
  },
  {
   "id": "differentiation-returning",
   "rule": "for returning visitors, the median matched held-out AUC is above the crossed one (the organ grown for them beats the one grown for new visitors)"
  },
  {
   "id": "differentiation-new",
   "rule": "for new visitors, the median matched held-out AUC is above the crossed one"
  },
  {
   "id": "reproducible",
   "rule": "re-running every seed from this seal gives an identical record — CI re-runs it on every push"
  }
 ],
 "predictions": {
  "said": "before any organ was germinated on the real sessions for grading, by Kar",
  "seed-smaller-than-forest": "pass, comfortably — a ~31-byte spore grows a body of roughly 0.9–2.2 KB; median ratio well above 8×",
  "grown-works-held-out": "pass — the pilot (seed 1) grew to about 0.81 held-out AUC, near PageValues alone (~0.807) and far above chance",
  "germinates-offline": "pass — the kernel makes no network call; the trap is a formality that proves it",
  "differentiation-returning": "pass — the pilot showed returning matched 0.820 over crossed 0.798 (+0.021); returning visitors are ~85% of sessions and carry their own signal",
  "differentiation-new": "fail — the pilot showed new matched 0.847 UNDER crossed 0.858 (−0.010); new visitors are ~14% of sessions, too few for a grown organ to beat the one grown on the larger, correlated population",
  "reproducible": "pass"
 },
 "disclosures": [
  "Before the seal the pipeline was run once on seed 1 (a pilot) at three grow-budgets, to size the grow-config and sanity-check the claims; that pilot graded seed 1 on the held-out months and its numbers are quoted in the predictions above. The sealed grading seeds are [11,12,13,14,15] — fresh, never graded before this seal.",
  "The germination engine (grow, fit, the stem, the held-out protocol) and the real sessions are reused verbatim from pattern-organs (sjgant80-hub, sha256 of organ.mjs sealed above). fall-spore adds only the spore: packing the genome to bytes, the konomi-family codec, germinate/differentiate, the offline probe and the ratio.",
  "The seven stem elements are the capability router's seven stages, whose basis is Thomas Frumkin's MACCubeFACE lattice; what each element means inside a funnel organ is pattern-organs' design.",
  "A femtoLLM (an optional, real small model) is offered on the live page for one soft step only — reading a free-text description of a client's domain and suggesting which context to express. It is not part of any rule here and nothing graded depends on it."
 ]
}