Discovery Principle / Develop-Leg — Research Grounding Digest

What this is. A neutral evidence catalog for the develop-leg claims of the Discovery Principle and Moral Architecture Riff (v1.2). It is a Research Protocol T2 pass (output form (b), Research grounding) lifted out of an exchange and into a standalone file, so it can be handed to an independent reviewer as-is. It answers the riff's own §9 Q6 (does the §4.1 literature survive tiering, and does any of it already state the civic implication?) and §9 Q11 (does the correlated-collapse claim survive tiering?). It is now the shared evidence base for Exchange #27 — the digest every Round-1 lineage was given, and the catalog the reserved (independent-lineage) Round 2 should re-verify rather than inherit. The reserved Round 2 has since run cross-lineage (GPT, Grok, Gemini) and returned a HOLD (Exchange #27 §2); after Round 4 opened a v2, the eight sources those reviewers surfaced beyond this catalog were grounded as Cluster D (§9).

What this is NOT. It is not an exchange — it runs no rounds, decomposes nothing into claims, and reaches no verdict. It carries no confidence numbers and no adversarial findings; those belong to Round 2 — now run cross-lineage by independent model families (GPT, Grok, Gemini) per the Adversarial Review Protocol lineage-independence default, and recorded in Exchange #27 §2, not here. This digest itself remains a neutral catalog and is not registered in _EXCHANGE_INDEX.md.

Provenance caveat — read first. This catalog was assembled by the same model lineage (Opus) that wrote the riff. A research sub-agent gathered and tiered the sources; a second sub-agent re-retrieved every load-bearing source to check citation integrity — but both sub-agents share that lineage. So the citation-integrity layer is double-checked in-lineage, while the independent read — does this evidence actually support the claims, and what did a non-shared prior surface that we missed? — is exactly what remains to be done. Treat every entry below as pre-vetted material to re-verify, not settled grounding. The verification ledger (§5) tells you what was confirmed verbatim, what was confirmed only via secondary access, and what was corrected.

1. How to use this digest (verification handoff)

If you are the independent lineage running Round 1 (or a human reviewer):

1. Do not inherit the tiers or stances on trust. Re-retrieve each URL, confirm the source exists at the cited venue/date, and assign your own tier under Research Protocol §2. Where we flag "confirmed via secondary," the primary text was not directly opened by us — treat those as unconfirmed until you open them.
2. Re-check the corrections in §5. Three citation errors were caught and fixed in our pass (Mezirow's venue; a "personal vs. social transformation" quote misattributed to Fleming rather than Taylor; Hood et al.'s tier and the location of its intratextual quote). Confirm our fixes are right; we may have introduced new ones.
3. Form your own read. The per-cluster "what the sources collectively bear on" notes in §2–§3 are deliberately neutral (what each source argues, not whether the claim survives). Our same-lineage interpretive read is quarantined in §7 and labeled do not inherit — read it last, or not at all, if you want a clean look.
4. Mind the gaps (§6). The three dose-response literatures originally deferred are now grounded in Cluster C (§4); the eight reviewer-surfaced sources are now grounded in Cluster D (§9). What remains open is the matched-pluralist comparison (Cluster B) and collective-scale dose-response. If your Round leans on dose-response conditions, start at Cluster C; if it leans on the relabel (S18) or cross-scale-transfer (S17) questions, start at Cluster D.

Clusters A and B map to the riff's two candidate transferable claims (the only parts the riff argues are its own contribution, riff §2.4); Cluster C grounds the make-or-break dose-response condition (riff §4.2, §8); Cluster D grounds the adversary-surfaced sources that bear on the relabel and transfer claims:

• Cluster A → the demand-driven discovery claim (riff §2.1) and the dose-response / make-or-break condition (riff §4.2, §8), plus the relabel/redundancy gate (riff §6 position D, §8) and the cross-scale transfer fault line (riff §2.4).
• Cluster B → the authority-coupling / correlated-collapse claim (riff §2.5).
• Cluster C → the dose-response / productive-vs-harmful condition (riff §4.2, §8; exchange S6/S7) — the named gaps this digest originally deferred, closed in a June 2026 gap-close pass.
• Cluster D → the reviewer-surfaced sources (exchange §2.7): relabel antecedents (Piaget → S1/S2/S18; Rogers → S23; Kohlberg/Rest → S2) and counters (Bandura → S2; Latané/Durkheim → S12; Kahan → S10; Buchanan & Powell → S17). Grounded in a June 2026 Round-4 pass.

"Bears on" is neutral: a source listed under a claim may cut for or against it. The "What it argues" column records the source's own stance toward the claim, not our judgment of the claim.

2. Cluster A — Does framework-breaking tension develop values-reasoning capacity?

Bears on the riff's demand-driven discovery claim (§2.1), the dose-response condition (§4.2), the relabel gate (§6 D / §8), and the cross-scale transfer guard (§2.4).

Supplementary (retrieved and tiered, used for context): Kohlberg's cognitive-conflict mechanism (W4/W5, secondary); Gilligan, In a Different Voice, 1982 (W1, care-ethics critique); Vygotsky, ZPD, 1978 (primary is W1; the linked explainer is a W5 secondary — re-source to the primary); Wood, Bruner & Ross, "The role of tutoring in problem solving," 1976 (W1, origin of "scaffolding"); and the backfire trio — Nyhan & Reifler, 2010 (W1, original backfire effect) versus Wood & Porter, 2019 (W1) and Swire-Thompson et al., 2020 (W1/W2), which find the backfire effect largely fails to replicate.

What the sources collectively bear on (neutral). A century-spanning chain (Festinger → Kohlberg's cognitive conflict → Vygotsky → Bjork → Kapur) addresses whether cognitive/skill growth is driven by encountering the limits of one's current framework, with Kapur 2008 the strongest experimental leg and Mezirow supplying the "disorienting dilemma" construct. Two sources pull the other way on the values and collective versions specifically: Haidt (moral judgment may be intuitive, reasoning post-hoc) and Sunstein (collective deliberation can polarize). A standing observation for the reviewer to weigh independently: every retrieved primary source operates at the individual/pedagogical scale; none tests the project's actual civic claim (that a society or institution develops members' values-reasoning by designing dilemmas), and the one retrieved source studying the collective level (deliberation) reports polarization. Whether that is fatal, bounding, or merely an open gap is a Round-1/2 judgment, not one this digest makes.

3. Cluster B — Authority-coupling and correlated collapse

Bears on the riff's authority-coupling / correlated-collapse claim (§2.5) and the cross-scale transfer guard (§2.4).

Supplementary: Rokeach, Beliefs, Attitudes, and Values, 1968 (W1, central/peripheral beliefs — content confirmed via secondary, no verbatim); Altemeyer & Hunsberger, Amazing Conversions, 1997 (W4, fundamentalist apostates leaving over ~3 years — details via secondary).

What the sources collectively bear on (neutral). The structural half of the claim has strong support: the series/parallel framing is textbook-settled, Hood et al.'s intratextuality is a near-exact substantive match to "a single upstream authority," and Boutyline & Vaisey plus Rokeach give belief-centrality as a measurable property. The dynamics half divides the sources: the riff's vivid "sudden, total cascade" timing runs against the best direct deconversion evidence (Streib; Altemeyer & Hunsberger find a gradual, multi-year exit). The robust-yet-fragile transfer is rigorous for engineered and financial networks (Carlson & Doyle; Doyle et al.; Gai & Kapadia) but no retrieved source applies it to belief systems — that transfer is a hypothesis imported from systems theory. The comparative prediction that would make the claim transferable — single-authority believers lose more beliefs more totally than pluralistically-warranted believers under the same disconfirmation — is untested against a matched pluralist comparison group in anything retrieved. That is the highest-value target for Round 2 / a future research sweep.

4. Cluster C — The dose-response boundary (productive vs. harmful tension)

Bears on the riff's make-or-break dose-response condition (§4.2, §8) and the exchange's S6 (dose-response exists) / S7 (the design gate). Added in a June 2026 T2 gap-close pass that closed the three literatures §6 originally named.

Supplementary: Allport, The Nature of Prejudice, Addison-Wesley, 1954 (W1 canonical; origin of the four contact conditions — content confirmed via Pettigrew & Tropp; primary not opened, named not linked); Diamond, Campbell, Park, Halonen & Zoladz, "The Temporal Dynamics Model… and the Yerkes–Dodson Law," Neural Plasticity 2007:60803 (PMC, W2) — the 1908 data are thin (few mice, coarse shock gradient, the inverted-U clean in only one of three experiments); the law's robustness comes from later (esp. WWII human-performance) work, not the mice; Janoff-Bulman, Shattered Assumptions, Free Press, 1992 (W1; the assumptive-world construct PTG builds on — named, via Tedeschi & Calhoun; primary not opened).

What the sources collectively bear on (neutral). The dose-response conditions the riff leaves unspecified (§4.2, §8 names "psychological safety, scaffolding, voluntariness, scaled dose"; exchange S6/S7) have established literatures, and they converge on a reframe: the productive-vs-harmful boundary looks governed more by conditions than by dose magnitude alone. Edmondson supplies the safety precondition; Allport / Pettigrew & Tropp supply the contact conditions under which cross-difference tension reduces rather than entrenches — and find them facilitating, not strictly necessary; Tedeschi & Calhoun supply the high-dose case (framework-shattering can rebuild) while showing growth co-occurs with distress; Frazier et al. warn that reported growth may not equal measured growth; Yerkes–Dodson supplies the inverted-U shape, but its weak origin (Diamond et al.) cautions against treating "the curve" as settled. Standing observations for the reviewer: (a) every result is individual / small-group scale — none specifies dose-response conditions for designed civic exercises at population scale (the cross-scale transfer fault line, exchange S17, recurs here); (b) the convergence on conditions over magnitude would recast S6 from "find the right amount of tension" toward "establish the conditions (safety, equal footing, shared goals, voluntariness) under which tension is metabolized" — the more buildable, and more falsifiable, form of S7's design gate; (c) Frazier et al. is a measurement caution for the whole develop leg, landing directly on the exchange's new measurement open question (§1.9).

5. Verification ledger (citation-integrity)

Per Research Protocol §4.1: a research sub-agent produced the grounding with a self-check, then a separate sub-agent re-retrieved every load-bearing source rather than trusting the citation. (Both same-lineage — see the provenance caveat.) Results:

• Confirmed verbatim at the cited work (9): Kapur (2008); Haidt (2001); Bjork & Bjork (2020); Nyhan & Reifler (2010); Wood & Porter (2019); Swire-Thompson et al. (2020); Carlson & Doyle (1999); Doyle et al. (2005); Gai & Kapadia (2010).
• Confirmed without a direct quote (2): Boutyline & Vaisey (2017); the series/parallel reliability result.
• Confirmed only via secondary access — re-open the primary before relying on these (5): Festinger (1957); Hood et al. (2005); Streib et al. (2009); Rokeach (1968); Altemeyer & Hunsberger (1997).
• Corrections applied in this pass (re-confirm them):
  1. Mezirow's 1978 venue was Adult Education (renamed Adult Education Quarterly in 1983) — not the later name.
  2. The "personal > social transformation" critique is Taylor (1998, p. 25), surfaced via AERC proceedings — not Fleming (2018), whose venue (IGI Global) is not a W1/W2 outlet. The cluster table reflects the fix.
  3. Hood et al. (2005) is a commercial-scholarly-press monograph — W2-equivalent, not W1 — and its intratextual verbatim sits in the review, not the book excerpt.
• Residual flag (no quote affected): Sunstein's vivid wording ("predeliberation tendencies," "indeed fanaticism") is verbatim in Sunstein, Deliberative Trouble? Why Groups Go to Extremes, Yale Law Journal 110 (2000) and later books — not at the linked 1999 working paper (which reads "judgments" / "even fanaticism"). This digest's Cluster A entry uses only paraphrase of Sunstein, so no quotation is affected; the linked source is the WP, not the origin of that exact phrasing.
• Cluster C additions (June 2026 gap-close) — re-confirm these: Confirmed verbatim at the cited work (4): Edmondson (1999, ASQ — MIT PDF); Pettigrew & Tropp (2006, JPSP — Wharton PDF); Tedeschi & Calhoun (2004, Psychological Inquiry — ResearchGate PDF); Frazier et al. (2009, Psychological Science — SAGE/PubMed). Confirmed only via secondary — re-open the primary (3): Yerkes & Dodson (1908, primary not opened; shape confirmed via Wikipedia / The Learning Scientists); Allport (1954, via Pettigrew & Tropp); Diamond et al. (2007, via reviews). No corrections were required in the Cluster C pass.
• Cluster D additions (June 2026 Round-4 grounding) — re-confirm these: Citation confirmed at venue/catalog with author, venue, year (8): Kahan et al. (2007, JELS — Yale OA PDF); Bandura (1986, Social Foundations — APA catalog, book/no OA full text); Piaget (1985, Equilibration — archive.org); Latané (1981, American Psychologist — OA PDF); Durkheim (Suicide 1897 / Division of Labor 1893 — primary via U-Chicago academic summaries, primary not opened); Rogers (1957, J. Consulting Psychology — ScienceOpen); Schlaefli, Rest & Thoma (1985, RER — SAGE); Buchanan & Powell (2018 OUP book; 2016 Ethics precursor — OA PDF, NDPR review). Named but not re-retrieved this pass (flag before relying): Bandura, Ross & Ross (1961); Piaget (1932); Blatt & Kohlberg (1975); Kahan et al. (2017, BPP); Appiah (2010); Pinker (2011). No corrections were required in the Cluster D pass.
• Open verification step: steward (or independent-lineage) spot-check at this version boundary remains outstanding (§4.2).

6. Named gaps — status

The three dose-response literatures this digest originally named were closed in a June 2026 T2 gap-close pass — now grounded in Cluster C (§4):

• Psychological safety — Edmondson (1999): the safety precondition for learning-through-discomfort. ✔ grounded.
• Intergroup-contact theory — Allport (1954) / Pettigrew & Tropp (2006): the conditions under which cross-difference tension reduces vs. entrenches prejudice (facilitating, not gating). ✔ grounded.
• The trauma / productive-difficulty boundary — Tedeschi & Calhoun (2004) with the Frazier et al. (2009) measurement counter and the Yerkes–Dodson (1908) inverted-U (Diamond et al. caveat). ✔ grounded.

The eight reviewer-surfaced sources Round 2 named were grounded in a June 2026 Round-4 pass — now Cluster D (§9).

Still open (named, not closed), per the §2 W8 discipline:

• The matched-pluralist comparison for the correlated-collapse claim (Cluster B) — single-authority vs. pluralist believers under the same disconfirmation — remains untested in anything retrieved (the highest-value Cluster B target).
• Collective-scale dose-response and cross-scale transfer (S17) — every Cluster C result is individual / small-group scale, and the one structural account of collective moral change now catalogued (Buchanan & Powell, Cluster D) runs against the individual-capacity transfer. No retrieved source supplies a matched collective-scale test of the transfer (the Exchange #27 §4.3 test design names this as the make-or-break experiment); grounding sources is not the same as running the test.

7. Same-lineage interpretive read (DISCLOSED — DO NOT INHERIT)

This section records what our lineage concluded from the catalog above, disclosed for transparency. It is not evidence and not a verdict you should adopt. It is included so that an independent reviewer can see — and discount — the prior the source-gatherers were operating under. If you want a clean read, skip it. It carries no confidence numbers (those were Round-2 outputs of a rolled-back same-lineage adversarial pass and are deliberately omitted).

Our same-lineage reading was that the develop-leg frame survives but does not pass cleanly: the demand-driven claim looks strongest recast as demand-then-supply sequencing (not demand instead of supply) and is weakest on its values-reasoning reading (Haidt); the correlated-collapse claim looks strongest recast as more-total-not-more-sudden (the suddenness is contradicted by gradualist deconversion evidence, and the robust-yet-fragile rescue is an unvalidated analogy); the dose-response condition looks answerable at individual scale and unanswered at civic scale; and the single most decision-relevant pattern we saw was that the individual→collective transfer is the fault line — the evidence is strong for how a person develops and silent-or-reversed for how a society does. We read what survives as heuristic-shaped, not principle-shaped. A reviewer with a non-shared prior may cut this differently — that divergence is the point of the handoff.

8. Provenance and register

Idea layer only (riff §1.2 privacy firewall): the originating personal/biographical substrate is excluded. Working-claim register; named uncertainty; both sides of each contested claim cited per Research Protocol §2.3. The full pre-rollback Round 2 (the same-lineage adversarial pass these sources were gathered for) is preserved out-of-tree on git branch wip/round2-rollback-snapshot-20260605 for anyone who wants to see how the lineage used this evidence — but that read is reserved, not authoritative.

9. Cluster D — Reviewer-surfaced sources (grounded June 2026)

The reserved cross-lineage Round 2 (GPT, Grok, Gemini; see Exchange #27 §2) named eight sources not in this digest. After Exchange #27 Round 4 opened a v2 (reframing the develop-leg as a testable civic design hypothesis), those eight were grounded in a June 2026 T2 pass — full citations, tiers, and what each argues. They cluster around the two claims Round 2 found decisive: S18 (relabel — most entries are named antecedents of the riff's individual-scale mechanisms) and S17 (cross-scale transfer — Buchanan & Powell supply a competing, structural account of collective moral change). Most cut against or bound the claims; that is expected, since they were surfaced by adversaries.

Supplementary: Kahan, "Ideology, Motivated Reasoning, and Cognitive Reflection: An Experimental Study," Judgment and Decision Making 8(4), 407–424, 2013 (SSRN, W1) — higher cognitive reflection amplifies identity-protective polarization rather than reducing it (the "motivated numeracy" line, Kahan et al., Behavioural Public Policy 1(1), 2017, extends this); a sharp reinforcement of the Cluster A Haidt caution that more reasoning capacity can entrench rather than repair. Also named (not retrieved this pass): Appiah, The Honor Code: How Moral Revolutions Happen, Norton, 2010 (esteem/honor dynamics, not individual capacity, drive moral revolutions); Pinker, The Better Angels of Our Nature, Viking, 2011 (long-run violence decline attributed to institutions — Leviathan, commerce, cosmopolitanism) — both structural-driver companions to Buchanan & Powell on S17.

What the sources collectively bear on (neutral). This is the adversary-surfaced cluster, and it sorts into three groups. (a) Relabel antecedents — Piaget (equilibration ⊃ the demand-driven discovery mechanism), Rogers (client-centered conditions ⊃ the S23 "therapist pole"), and Schlaefli/Rest/Thoma (Kohlberg/Rest dilemma discussion ⊃ S2's "moral exercises") — each names a well-precedented individual-scale mechanism, sharpening S18: the psychology is not new, so the project's novelty, if any, is the civic-design transfer, not the mechanism. (b) Counters — Bandura (modeling builds architecture without tension, vs. S2), Latané (number and immediacy of sources do matter, vs. S12's correlation-only move), Durkheim (plural authority yields anomie absent re-integration, vs. S12), and Buchanan & Powell (collective moral change is condition-driven, the decisive counter to S17). (c) A mechanism reframe — Kahan relocates S10's "coupling" into identity-protective cognition and, with the motivated-reflection finding, reinforces the Haidt caution that added reasoning capacity can deepen rather than dissolve the lock. Standing observation for the reviewer: the cluster lands hardest on exactly the two claims Round 2 found decisive — it supplies named antecedents for the relabel charge (S18) and a competing structural theory of collective moral change (S17, via Buchanan & Powell). Both push toward the Round 4 reframe: carry the develop-leg as a civic design hypothesis with a falsifiable transfer test, not as a psychological claim.