{"id":"EXP-0002","type":"experiment","title":"Knowledge Reconstruction Fidelity: once recognized and retrieved, can evaluators reconstruct Upstream Zero faithfully?","created":"2026-07-15","status":"draft, pre-registration pending founder approval","authors":["Upstream Zero"],"edges":[{"rel":"tests","to":"H-1"}],"predictions":["At least one retrieval-capable evaluator cites upstreamzero.com for a basic identity prompt within the observation window","Tier flattening is the modal failure, where evaluators restate the company's Narrated founding claims as established fact","Machine surfaces (graph.json, object JSON) are not used unprompted; they appear only when verification is explicitly requested, if at all","The optimization-agency misread (describing Upstream Zero as an SEO/AEO/AI-visibility agency) occurs at least once","Published refutation conditions rarely survive reconstruction, and uncertainty is the first thing an evaluator drops"],"founderDecision":"FD-8","pubState":"published","body":"**Knowledge Reconstruction Fidelity.** The observatory's second\nexperiment asks a question that only becomes meaningful *after*\npropagation: once an evaluator has recognized and retrieved Upstream Zero,\n**can it reconstruct Upstream Zero faithfully:** its identity, claims,\nrelationships, evidence tiers, uncertainty, and theory? This is distinct\nfrom [EXP-0001](/experiments/EXP-0001) (Knowledge Propagation), and the\ntwo are not merged: recognition and retrieval are the precondition;\nfaithful reconstruction is the thing measured here.\n\n**Day 0 evidence is not scored here.** Propagation captures collected at\nDay 0 are baseline observations for EXP-0001. Reconstruction scoring\napplies only to responses in which the evaluator has already recognized\nand retrieved the entity; attaching Day 0 propagation captures to\nreconstruction scoring would confound the two experiments.\n\n**Motivating example (external, not evidence produced by Upstream\nZero):** the *Inspired Taste* case, in which visibility and citation\ncoexisted with a failure of representation fidelity. It demonstrates that\nbeing cited is not the same as being reconstructed faithfully. That is the\nexact pattern this experiment is built to measure. It motivates the question; it\nevidences no finding of ours.\n\n**Status honesty:** this is a draft protocol, held privately until it is\nformally pre-registered. It has not been pre-registered, no data has been\ncollected, and its predictions are Narrated intuitions registered so that\nreality can falsify them. On pre-registration the protocol and its frozen\nmaterials become published research objects here. Pre-registration requires founder approval\n(FD-8); when it happens, this object's status change will be a recorded\nRevision.","url":"/experiments/EXP-0002","machineUrl":"/objects/EXP-0002","referencedBy":[],"_meta":{"site":"Upstream Zero · Commercial Evaluation Observatory","version":"0.1","note":"Claims are presented at their evidence tier; Narrated is the lowest. Verify by walking edges, not by trusting us."}}