Peer Review 1 — final-answer.md (09_comparative/3_synthesis)

Reviewer independence: I authored nothing in this vault. Method: full read of the target; verification against AGENTS.md, 00_meta/Methodology.md, BiasAudit-2026-06-11.md, and the OPEN_QUESTIONS register; full read of 22 of the 29 cited notes (all seven named load-bearing notes plus qafzeh, sungir, dyeus-cognate-set, both rigveda notes, israelite-origins, kuntillet-ajrud, shamanism-baseline, dreaming, aboriginal-flooding, oracle-bones, shinto, eridu, sumerian-religion, osiris-cycle) and frontmatter verification of the remaining seven; independent web stress-tests of five load-bearing empirical claims.

Overall: strong. Every per-note confidence rating I checked is carried over correctly, including the hard cases: the dyeus/sky-father two-level scoping (name-formula high in 09 vs. deity-reconstruction medium in 04) is reproduced exactly as the two notes reconcile it; el-yahweh is cited as "high presence / medium mechanism," which actually repairs the flattening BiasAudit #15 complained about; shamanism-baseline and the Nunn-Reid claim are correctly carried at low; the Rigveda dual anchor (c. 1040 CE reported / 1464 CE secure) matches the reconciled statement in both rigveda notes verbatim. The atrahasis claim is cited as "high — wave-2 re-sourced," which matches that note's documented sourcing upgrade. Dates check out across all spot-checks (Qafzeh 92 ka, Sungir 34–30 ka, Eridu XVIII c. 5500 BCE, Unas c. 2350 BCE, KN Fp 1 c. 1400–1375 BCE, Mitanni c. 1380 BCE, Kuntillet Ajrud c. 800 BCE, oracle bones 1254–1197 BCE, Kojiki/Nihon Shoki 712/720 CE).

Overreaches found (quote pairs):

• O1 (the serious one). Answer, Thesis: "All three mechanisms are proven for at least one major story each." The cited notes prove descent (sky-father: sound laws) and contact (great-flood: dated corridor + sequence parallels) at that grade, but no note "proves" convergence for a major story. dying-rising-god says Osiris is "best read as convergent" — a probabilistic reading; great-flood says global convergence is "the default for distant occurrences" within an unresolved verdict (Q5); sky-father says non-IE sky fathers are "convergent on present evidence," with steppe-contact alternatives explicitly open (Q11). The body of Part 4 is properly hedged; the Thesis sentence is not, and a thesis is what gets quoted.
• O2. Part 2, trait 1: "treating the dead as still mattering — still owed labor, goods, ritual attention — is the oldest and most universal documented religious behavior." paleolithic-mortuary-religion: the inference from mortuary archaeology to "religion" "is a 5th-evidence-class inference... not directly readable from the stratigraphy." claim-qafzeh-ochre-earliest-symbolic-behavior: the evidence supports "symbolic behaviour in mortuary context, not the specific content of belief." Part 1 carries this hedge faithfully; Part 2 drops it and upgrades the same evidence to "religious behavior."
• O3. Part 2, trait 1: "Osirian afterlife judgment." osiris-cycle-earliest-attestation attests murder, reconstitution, and kingship over the dead — it says nothing about judgment of the dead (weighing-of-the-heart material belongs to later corpora the vault has not gathered). Unsupported by the cited note.
• O4. Feynman section: "That's most dying-and-returning gods: farmers everywhere watch crops die and return... without ever meeting each other." The note's finding is the opposite of a convergence majority: the best-dated strands are descent-within-lineage (Dumuzi/Tammuz) and contact (Tammuz→Jerusalem, Adonis's loan-name); only Osiris is read as convergent. "Most" is an invented quantifier.
• Trivial. "written down by c. 1635 BCE" (following great-flood) vs. "c. 1646–1626 BCE" (atrahasis-antedates-gilgamesh-flood-tablet) — an inherited intra-vault inconsistency, both within Ammi-ṣaduqa's reign; harmonize or cite the reign range.

No case found where the answer exceeds a cited note's confidence rating in its explicit ratings; the overreaches above are all prose-level, not frontmatter-level.

The four-part structure maps one-to-one onto the four questions, and the three per-motif verdicts are carried over exactly (contact/high/scoped; descent/high/scoped; unresolved/medium — including the subtle point that the dying-god medium qualifies the verdict, not the data). Q-numbers cited in the gap table and open_questions all resolve against the register (Q1–Q26 verified present).

Two internal tensions:

• T1. Part 4 calls Osiris "the earliest attested of all (c. 2350 BCE)" while dying-rising-god's own occurrence table puts Dumuzi's name in Early Dynastic god-lists mid-3rd millennium BCE — potentially earlier. The note resolves this implicitly (name-attestation ≠ motif-content attestation); the synthesis should resolve it explicitly in one clause.
• T2. Part 1's closing: "earliest evidence is always ritual before theology, archaeology before text." "Always" overstates the documented sample: the earliest Israelite-religion attestation is a textual toponym (Soleb), and PIE religion has no class-1 anchor at all — it is reconstruction throughout. "In every domain with archaeological coverage" would be true.
• Minor: Part 2 opens "Across nine domains with independent evidence trails" — eight domains carry evidence; 09_comparative is synthesis over them, not an independent trail.

No non sequiturs found. The braided-river conclusion follows from the three verdicts given O1's correction (it survives as "all three mechanisms operate, two provably, one as the best available reading").

Five load-bearing claims stress-tested against current literature; all held:

1. Oracle bones radiocarbon 1254–1197 BCE (Wu Ding): confirmed — Liu et al., Radiocarbon (Cambridge Core), 26 bones, exactly the range and probability the vault states.

2. Whitehouse et al. 2019 retraction: confirmed — retracted Nature, July 2021, over missing-data coding (Beheim et al. critique); the answer's "correlation survives, causal arrow unresolved" framing matches the post-retraction literature (including the 2022 "retake" reanalysis debate).

3. Atrahasis colophon + Gilgamesh XI dependence: confirmed — Ammi-ṣaduqa-reign colophon (1646–1626 BCE) and the hero called Atra-ḫasīs inside Gilgamesh XI are standard since Lambert & Millard 1969 / George 2003. (Older editions cite one occurrence at XI 187 vs. the note's two at 49/197 under George's numbering — an edition-numbering difference, not an error.)

4. Linear B di-we, KN Fp 1: confirmed — the tablet lists oil offerings to Diktaian Zeus. Knossos tablet dating is genuinely contested (c. 1450–1375 depending on destruction-horizon position); "c. 1400–1375 BCE" is a defensible mainstream span, and sky-father already pre-registers that dating revisions would be minor for the verdict.

5. Soleb "Shasu of yhwꜣ": confirmed as majority-but-not-universal reading (e.g., Fleming's Yahweh before Israel dissents) — which is precisely how the answer hedges it ("read as Yahweh by most specialists but not all").

Nothing in the answer contradicts current mainstream scholarship that I could find. Where scholarship is split (Genesis dependence direct vs. mediated; Knossos dating; asherah goddess vs. pole), the answer sits on the consensus center and names the dissent.

Overall confidence: medium is correct — neither timid nor bold. The scoped confidence comment (high for per-motif verdicts and dated anchors, medium for the integrated picture, lower-bound per BiasAudit §4 practice) is exactly the right structure and is BiasAudit-compliant.

Per section: Part 1 correctly mixes high anchors (oracle bones, Pyramid Texts, attestation gaps) with medium/low components and labels each inline — well-calibrated. Part 2 is presented as "observed regularities... not a law," which is right, but trait 1's "most universal documented religious behavior" is its one bold spot (O2). Part 3 at medium matches what-religions-do exactly, including the retraction caveat and the hindsight-machine concession. Part 4 high-for-flood and high-for-sky-father match the notes; the only calibration breach is the Thesis's "proven" for convergence (O1). The counter-argument section is genuinely adversarial, not decorative — the attestation-bias steelman correctly identifies the method's constitutional blind spot and the answer does not pretend to escape it.

All four questions are answered; none is dodged. The "where evidence ends" honesty is consistent across the thesis, Part 4's close, the gap table, and the steelman — the same boundary (pre-writing, beyond sound laws) is drawn the same way in all four places, which is the mark of a real epistemic position rather than a disclaimer. The sample limit (Mesoamerica, sub-Saharan Africa, Oceania ethnographic-only) is named three times. Two small gaps: Greek/Roman/Norse religion appears only through motif evidence and is not named as a coverage limit alongside the other three; and Christianity, Islam, Buddhism, and Jainism are dispatched in a sentence each — defensible for a vault scoped to origins, but worth one explicit sentence of scoping so "where does each religion come from" is not read as promising more than nine formation-histories.

Clean. Emic/etic discipline holds throughout — trait 5 handles the most dangerous material (every tradition's self-account of timelessness vs. the etic paper trail) with the emic register explicitly protected, and the closing paragraph correctly disclaims any bearing on truth-claims. No advocacy, no debunking tone detected; the "romantic extreme / reductive extreme" framing targets scholarly hypotheses, not traditions. No URLs appear in the synthesis at all, so the no-hallucinated-URL rule is trivially satisfied; both meta-references (Methodology, BiasAudit-2026-06-11) resolve to real files. One stylistic note: "the literate traditions all deny their own historical origins from the inside" — "deny" is mildly loaded; "do not narrate their origins as historical development" is the cooler etic phrasing.

REQUIRED

1. Thesis sentence (O1): replace "All three mechanisms are proven for at least one major story each" with a graded statement — descent and contact are demonstrated at dependence/sound-law grade; convergence is the best-supported reading for the Osiris strand and the default for the global flood distribution, but is not "proven" at the same grade anywhere in the vault. The braided-river image survives this correction intact.

2. Part 2 trait 1 (O2): restore the class-5 hedge — "religious behavior" → "mortuary/symbolic behavior" (or add the "belief content is not readable" caveat that Part 1 already carries).

3. Part 2 trait 1 (O3): remove "Osirian afterlife judgment" or re-source it; the cited note attests netherworld kingship, not judgment of the dead.

4. Feynman section (O4): drop "most dying-and-returning gods" — state instead that one strand (Osiris, and plausibly other agrarian cases) is read as independent invention, while the Tammuz→Adonis trail is borrowing. The simple explanation must not contradict the verdict it is simplifying.

5. Part 4 / Osiris (T1): add one clause distinguishing motif-content attestation (Osiris, c. 2350 BCE) from name-only attestation (Dumuzi in ED god-lists, mid-3rd millennium), since the cited note's own table otherwise contradicts "earliest attested of all."

SUGGESTED

6. Soften "always ritual before theology" to "in every domain with archaeological coverage" (T2).

7. Fix "nine domains with independent evidence trails" → eight evidence domains plus the comparative layer.

8. Harmonize the Atrahasis tablet date (c. 1635 vs. 1646–1626 BCE) or cite the Ammi-ṣaduqa reign range; ideally fix it upstream in great-flood / the 02 note simultaneously.

9. Name the Greek/Roman/Norse coverage limit alongside Mesoamerica/Africa/Oceania, and add one scoping sentence for the post-Israelite and post-Vedic daughter traditions in Part 1.

10. Reword "deny their own historical origins" to a neutral formulation in trait 5.

VERDICT: minor revisions

The synthesis is faithful to its evidence base to an unusual degree — 22 spot-checked citations, zero frontmatter-level confidence breaches, five external stress-tests passed. The required fixes are all prose-level: one overclaimed thesis verb, two trait-level slips past hedges the answer itself maintains elsewhere, and two one-clause clarifications. Nothing structural. After revisions 1–5, this is promotable to reviewed.