Peer Review 2 — final-answer.md (methods & epistemics referee)
Reviewer role: independent methods-and-epistemics referee. I authored nothing in this vault and deliberately did not read PeerReview-1 before forming these findings. Review basis: full read of the target, AGENTS.md, Methodology.md, BiasAudit-2026-06-11.md, OPEN_QUESTIONS.md, the three motif notes, what-religions-do.md, and spot-verification of load-bearing cited claims (including the wave-2 re-sourcing of the Atrahasis priority claim, which I confirmed in 02_mesopotamian/2_notes/atrahasis-antedates-gilgamesh-flood-tablet.md — the "wave-2 re-sourced" annotation in the References section is accurate, with George 2003 and Lambert & Millard 1969 now independently anchored). Web search was used for the steelman audit (dimension 3).
The trichotomy is applied consistently in everything the synthesis itself wrote. The 09 motif notes keep descent | contact | convergence | unresolved to cross-tradition verdicts, gloss intra-lineage cases explicitly ("descent within its own lineage" for Dumuzi, with the cross-cultural category kept separate), and describe Gilgamesh-XI-from-Atrahasis as textual dependence without forcing the enum. One inherited inconsistency: the cited supporting note 02/atrahasis-antedates-gilgamesh-flood-tablet.md still carries transmission: descent for that intra-cultural literary dependence — BiasAudit #13, flagged but unfixed in wave 2. The final answer's own prose handles it correctly; the frontmatter of its source does not.
Scoping is maintained almost everywhere — better than I expected to find. The flood verdict is scoped to the Near Eastern → biblical lineage in the frontmatter thesis, in Part 4's heading, in the braided-river paragraph's global-distribution carve-out, in counter_evidence, and even in the Feynman section's "not that we can find." Part 2's six traits are framed as "observed regularities in the vault's documented sample." The only slippage is stylistic shorthand inside the braided-river paragraph ("the flood was borrowed, the sky father was inherited") two paragraphs after scope was established — acceptable in context, but it is exactly the sentence a skimming reader will quote unscoped (Suggested #6).
The n=3 problem is real, and it bites in one specific place. The braided-river conclusion needs all three mechanisms demonstrated. The thesis paragraph claims: "All three mechanisms are proven for at least one major story each." Audit of that claim against the motif notes:
| Mechanism | Best case | Status in the vault's own terms |
|---|---|---|
| Descent | sky-father cognate set | proven — high confidence, sound-law discriminator |
| Contact | flood, Near Eastern lineage | proven — high confidence, attested corridor + sequence parallels |
| Convergence | Osiris strand of dying-rising; global flood; non-IE sky-father concepts | not proven anywhere — Osiris is "best read as convergent" inside an unresolved verdict with Q7 open; global flood is "convergence-default, unresolved" with Q5 open; Tiān/Tengri carry Q11 open |
Convergence is everywhere the inference-to-best-explanation in a question the vault itself classifies as open. That is a perfectly respectable epistemic position — but it is not "proven," and the symmetric three-legs framing ("three demonstrably different routes" in the frontmatter thesis; "demonstrably operate" in Part 4) overstates one leg. This is the single most consequential finding of this review (Required #1). Note the irony: the steelman's fourth prong already concedes the braided river "gives the three mechanisms equal billing" — the synthesis knows the legs are unequal and says so in the counter-argument, then asserts equality in the thesis.
Pass, with distinction. I hunted for a dated claim resting on class 3/4/5 evidence and found none:
- Rigveda composition dates are flagged "by linguistic inference" with the Mitanni treaty as the honest class-2 anchor, and the three-way manuscript contradiction from BiasAudit #3 is resolved in the prose ("c. 1040 CE (reported) / 1464 CE (secure)").
- Composition-vs-attestation is maintained at every occurrence: Genesis ("composition inferential"), Śatapatha Brāhmaṇa, Shinto's 712–720 CE gap, the Dreaming's ethnographic-only anchor.
speculativelabels are carried into prose, not laundered out: "every deeper reconstruction in the vault... sits at low confidence or is explicitly speculative" (Part 4); the Aboriginal coastal-flooding claim and the shamanism-baseline hypothesis are both named at low confidence in Part 1.- The frontmatter confidence (medium) takes the lower bound per BiasAudit §4 practice and says so.
- The one prior critical violation in this chain (Atrahasis high-on-Wikipedia, BiasAudit #2) was genuinely repaired in wave 2 — verified.
One hairline item: Part 4 states descent is "only provable as far back as language families reach (for Indo-European, roughly the fourth millennium BCE)." The descent verdict is class-2-anchored, but the fourth-millennium date is class-3 reconstruction — the sky-father note itself labels PIE "speculative as a dated object by vault rules." The sentence quietly imports a class-3 date into the "provable" zone (Suggested #7).
The four existing prongs (attestation bias, trichotomy non-transfer, functionalism-as-hindsight-machine, ANE sample thinness) are genuinely strong — prong 1 in particular correctly identifies that the method converts absence of evidence into evidence of absence, and prong 2 concedes the framework's "flagship victories are one linguistics case and one philology case." The prompt-level candidates are partially covered: functionalism unfalsifiability is prong 3; convergence-explains-more is half-covered by prong 4.
What is missing — two published objections stronger than anything currently in the section:
1. The category-formation critique (the strongest unengaged objection). Brent Nongbri, Before Religion: A History of a Modern Concept (Yale UP, 2013), with Talal Asad, Genealogies of Religion (1993), and J. Z. Smith's "religion is solely the creation of the scholar's study" (Imagining Religion, 1982): "religion" as a distinct sphere separable from politics, law, and economy is a 16th–17th-century European construct, and applying it to Shang divination, the Dreaming, or pharaonic kingship is anachronistic projection. This attacks the question itself — all four parts presuppose a stable cross-cultural object that has origins, traits, and functions. The vault's emic/etic discipline does not answer this, because the critique targets the etic category, not emic/etic blending. Notably, the synthesis's own data flirt with the rebuttal (Part 2 trait 3 shows "religion" doing sovereignty work everywhere, i.e., not a separable sphere — which is Nongbri's point) without ever naming the problem. Nongbri's own resolution (redescriptive use of an avowedly modern category) is available and would cost the answer one paragraph, but the objection must be engaged, not absorbed silently. (Surfaced via web search 2026-06-11; book and reviews at Yale UP via review coverage and criticalreligion.org.)
2. Lincoln's discipline critique — named but never engaged. The source_diversity field lists "Lincoln's discipline critique" as a represented school, yet no sentence in the answer or its steelman engages it. Bruce Lincoln, Theorizing Myth: Narrative, Ideology, and Scholarship (Chicago UP, 1999): comparative mythology — specifically the Indo-European reconstruction the sky-father benchmark sits inside — has a documented political genealogy (Aryanism, Dumézil's entanglements), and "if myth is ideology in narrative form, then scholarship is myth with footnotes." This is sharper than the existing prong-2 trichotomy critique because it attacks the benchmark case's pedigree, not just the framework's transferability. Q23 (Dumézil retrojection) already half-files this; the steelman should cite it. (University of Chicago Press; BMCR review.)
Listing a school in source_diversity while never engaging it is the precise pattern BiasAudit §3 called "steelman written from memory of the opposition" — here it is steelman-by-citation-only.
The language analogy is apt, the dates in the simple explanation match the technical anchors (Eridu ≈ 7,500 yr; Pyramid Texts ≈ 4,400 yr; oracle bones ≈ 3,250 yr), and the gaps table maps cleanly to real register questions — I verified every Q-number cited (Q22, Q4, Q5, Q6/Q7, Q3/Q1, Q12/Q26, Q2/Q8/Q9) against OPEN_QUESTIONS.md; all exist and match.
Two honesty failures:
1. The dying-god simplification inverts the technical weighting. Simple explanation: "That's most dying-and-returning gods: farmers everywhere watch crops die and return, and tell god-stories about it without ever meeting each other." The technical section says the documented strands are descent-within-lineage and contact (Tammuz mourned in Jerusalem; Adonis carrying his Semitic loan-name), with only Osiris read as convergent — inside an unresolved verdict. A bright 12-year-old leaves the simple explanation believing convergence explains "most" dying gods; the technical reader learns the opposite. This is exactly the failure mode the Feynman gate exists to catch (Required #2).
2. "We can even trace the route the story traveled" — overstated. The corridor is attested at two candidate periods; the actual route and date are the open question (Q4), as the answer's own gaps table says one screen later (Suggested #5).
3. (Minor) "the oldest temple in Iraq" states flatly what the technical section holds at medium confidence as an interpretation ("the 'temple' label is an interpretation") (Suggested #9).
One workflow breach: the gaps table's last row (sample coverage) maps to "(scope limit, named in counter_evidence)" — i.e., no register entry. AGENTS §3.5(c) is unambiguous: every gap becomes an entry in OPEN_QUESTIONS.md. The vault's most consequential scope limitation is currently the only gap exempted from the loop that exists to fix gaps (Required #4).
"Unresolved" is used honestly — neither use is a dodge. Audit per use:
- Global flood distribution: the evidence as gathered leans convergence, and the answer says so plainly ("convergence is the default there"), keeping "unresolved" for the deep-descent alternative that is unfalsified rather than supported. The lean is stated, not hidden. Honest.
- Dying-rising category: "unresolved" here does not mean "we can't tell which mechanism" — it means "the category fragments into three strands with three different verdicts." The note delivers the per-strand verdicts as the real finding and says so. Honest, and actually the methodologically interesting result.
- Moralizing-gods causal arrow: "genuinely unresolved since the flagship 2019 study was retracted" — correct; the correlation is reported as surviving, the causal claim as open. Honest.
One conceptual refinement available: the single enum value unresolved is doing two different jobs — insufficient evidence (global flood) and ill-posed category (dying-rising). The dying-rising note's own counter_evidence concedes the honest verdict might be "not a motif," which is a third state the enum cannot express. Worth one clarifying sentence in the final answer and eventually a Methodology note (Suggested #8).
Angles absent from all 51+ notes, ordered by how directly they threaten or strengthen the current conclusions:
1. Category-formation critique (Nongbri, Asad, J. Z. Smith) — see dimension 3; doubles as steelman repair and wave-3 gathering.
2. A provable convergence case — the braided river's weak leg (dimension 1) is repairable with evidence rather than rewording: gather pre-contact Mesoamerican/Andean primary material (Popol Vuh with its colonial-transmission caveats, Dresden Codex, Chavín archaeology) where no transmission route to the Old World exists, and run the motif machinery there. A documented Old/New-World motif parallel with zero contact route is the cleanest available convergence demonstration — and simultaneously fixes the sample-thinness steelman prong.
3. Axial Age frameworks (Jaspers; Bellah, Religion in Human Evolution, 2011; Eisenstadt) — the vault independently dates the canonical Axial pivots (Upanishadic karma c. 700–600 BCE, exilic monotheism 6th c. BCE, Zhou Mandate) and never asks whether their clustering is a real synchrony, a diffusion effect, or a historiographic artifact (Mullins et al. 2018 tested this against Seshat). The vault already owns all the ingredients for a tier-3 essay here.
4. Ritual-before-belief theory (Rappaport, Ritual and Religion in the Making of Humanity, 1999; the Göbekli Tepe feasting/aggregation literature) — Part 1's closing pattern claim ("ritual before theology, archaeology before text") is currently an empirical observation with no theoretical engagement; ritual-first theory is its natural frame and its strongest test. Also resolves BiasAudit #12 (Göbekli Tepe's miscategorization) constructively.
5. Gender — the vault's evidence is saturated with women doing religion (Asherah at Kuntillet Ajrud, Inanna/Ishtar, the women weeping for Tammuz in Ezekiel 8:14, Mycenaean di-u-ja) and contains zero gender analysis. Most glaring single omission: Enheduanna (c. 2300 BCE), en-priestess of Ur and the earliest named author of religious literature in human history — absent from a vault whose organizing obsession is earliest attestations.
6. Ancestor cult as candidate deepest layer — Part 2's trait 1 ("the dead remain socially present") is the vault's most universal documented trait, yet it has no motif note and no transmission verdict of its own; it lives only as synthesis prose. It is the best candidate for a fourth motif, and (unlike the current three) one where class-1 archaeology dominates.
7. Neuroarchaeology and altered states beyond Lewis-Williams — Q24 covers the critique side; nothing covers the constructive program (music/trance archaeology, the psychoactive-substance claims that need skeptical pairing before they enter by popular osmosis).
8. (Lower priority) Child-cognition teleology findings (Kelemen) for the class-5 layer; secular-substitution longitudinal data (already half-tracked at Q9); writing-technology effects on attestation bias as its own methods note.
1. Fix the convergence overclaim. In the frontmatter thesis ("three demonstrably different routes") and the thesis paragraph ("All three mechanisms are proven for at least one major story each"): descent and contact are proven at high confidence; convergence is the best explanation in cases the vault itself classifies as open (Osiris/Q7, global flood/Q5, Tiān-Tengri/Q11). Either reword to the honest asymmetry ("two proven, the third strongly indicated but nowhere yet proven — see wave-3 item 2") or hold the claim until a provable convergence case is gathered. The braided-river image survives this fix; the word "proven" does not.
2. Fix the Feynman dying-god paragraph. Replace "that's most dying-and-returning gods... without ever meeting each other" with a version matching the technical verdict: the one well-evidenced convergence candidate is Osiris; the documented trail (Tammuz → Jerusalem, Adonis's loan-name) is contact; the category as a whole is fragmented. The simple explanation must not teach the opposite weighting of the technical section.
3. Engage the two missing steelman prongs. Add (a) the category-formation critique (Nongbri 2013 / Asad 1993 / J. Z. Smith 1982) — the strongest published objection the answer currently ignores, since it attacks the question's presupposition that "religion" is a stable cross-cultural object; and (b) Lincoln's Theorizing Myth (1999), which the source_diversity field claims as a represented school but no sentence engages, and which targets the political pedigree of the comparative project behind the benchmark sky-father case. Cross-link Q23.
4. File the sample-coverage gap as a register question. The Feynman gaps table's last row currently exits the open-questions loop ("scope limit, named in counter_evidence"), violating AGENTS §3.5(c). File a wave-3 Q for Mesoamerican/African/Oceanian primary gathering — which also services Required #1.
5. Soften the Feynman flood sentence "we can even trace the route the story traveled" to match Q4 (corridor attested; route and date open).
6. In the braided-river paragraph, restore one scope word to the shorthand ("the Near Eastern flood was borrowed") — it is the most quotable sentence in the document and currently the least scoped.
7. Rephrase "provable as far back as language families reach (for Indo-European, roughly the fourth millennium BCE)" so the class-3 status of the PIE date is visible ("conventionally dated, by reconstruction, to...").
8. Add one sentence distinguishing the two senses of unresolved (insufficient evidence vs. ill-posed category); consider a Methodology amendment or a fifth enum value (not-a-motif) in a future wave.
9. Hedge "the oldest temple in Iraq" in the simple explanation to match the medium-confidence interpretive status of Eridu Level XVIII.
10. Wave-3 housekeeping on inherited debts the final answer cites: the transmission: descent enum misuse in 02/atrahasis-antedates-gilgamesh-flood-tablet.md (BiasAudit #13, still unfixed), and the tradition-profile single-prong counter-evidence pattern (BiasAudit §2) that Part 1's per-domain paragraphs sit on.
In priority order: (1) category-critique tier-1 sources (Nongbri, Asad, J. Z. Smith) + Lincoln 1999 — steelman repair; (2) a New World convergence test bed (Mesoamerica/Andes primary material) — repairs both the convergence leg and the sample-thinness prong; (3) an ancestor-cult motif note with its own transmission verdict — the vault's most universal trait deserves the vault's own machinery; (4) Axial Age clustering essay from already-gathered dates + Bellah/Mullins sources; (5) ritual-first theory (Rappaport) + Göbekli Tepe recategorization; (6) gender of early religion, starting with an Enheduanna claim note (earliest named religious author, c. 2300 BCE — an attestation-class fact squarely in this vault's wheelhouse).
The synthesis is the best-disciplined document in the vault: evidence classes clean, scoping near-perfect, "unresolved" used honestly, falsifiability section exemplary, and the wave-2 repairs it claims actually happened. Its faults are concentrated and prose-level: one overclaimed word ("proven") at the load-bearing joint, one Feynman paragraph that inverts its own technical finding, two missing steelman prongs, and one gap exempted from the register loop. All are fixable without re-gathering.
VERDICT: minor revisions