Formation Document Initial Findings - Adversarial Review

Status (June 2026): Active; Rounds 1–3 complete; Round 4 (response/v2, after corpus expansion) reserved. Round 1 (April 2026) opened the adversarial challenge; Round 2 (June 2026) answered it against the now-20-source corpus and reclassified the synthesis to modern rights-constitutionalism convergence, not cross-tradition convergence; Round 3 (June 2026) ran the cross-lineage adversarial pass — two clean independent lineages (Grok 4.3, GPT-5.5) plus a third (Gemini 3.1 Pro) discarded as contaminated (it was an agent that wrote into the exchange, defeating the reduced-context discipline). Round 3 did not falsify the synthesis but wounded it: R3-2 (concentration) and R3-3 (modern-not-cross-tradition) → contested; R3-5 ("peace is the only gap") → contested / weakened (education-civic-formation is a co-candidate); R3-1 and R3-6 narrowed; R3-4 (expression→truth-infrastructure over-read) confirmed and now safe to enact. The decisive finding: the still-absent non-liberal / pre-modern / Indigenous families (Iran, PRC, Saudi, a Confucian order, rights-of-nature texts) are load-bearing, so corpus expansion is now the gating move. Steward decision (June 12, 2026): the peace discussion is held until corpus expansion — reinforced by Round 3, and widened to re-evaluate peace and education/civic-formation together.

Why this exchange: The project now has enough source coverage to start making stronger claims about overlap, divergence, and possible gaps in Civic Blueprint's principles. That is exactly the stage where false confidence becomes most dangerous. Before the corpus is treated as evidence that the project's principles are broadly aligned with global formation traditions, the early synthesis needs structured adversarial challenge.

Central question

What are the strongest counterarguments against the early synthesis emerging from the formation-document corpus, and which current findings are robust versus artifacts of source selection, translation, or interpretive generosity?

Dependencies

• Formation Document Comparative Analysis - Exchange
• Comparative Alignment Protocol
• Adversarial Review Protocol
• Alignment Matrix
• Gap Analysis
• Uniqueness Report

Early synthesis claims to challenge

The current corpus suggests several tempting conclusions:

1. Civic Blueprint is less normatively unusual than it first appears. Many modern constitutions and rights documents already align strongly on dignity, equality, anti-exclusion, welfare, openness, and constrained power.
2. The most unusual Civic Blueprint principles are technology-, ecology-, and moral-expansion related. AI governance, truth infrastructure, biosphere centrality, and open moral consideration remain less common than dignity or justice.
3. Structural constitutions and rights declarations need to be compared differently. The original U.S. Constitution, for example, is strongest on institutional architecture and weakest on substantive moral commitments, while the UDHR or South African Constitution carry more direct moral content.
4. The corpus may pressure the project more through subprinciples and tension notes than through wholly new principles. Education, peace, civic formation, and social-rights language recur, but may fit inside the current principles set rather than demand a full rewrite.

These may all be true. They may also be premature.

Adversarial challenge prompts

1. Source-selection bias

The current corpus is not random. It includes several texts already famous for rights-expansive or dignity-centered language:

• South Africa
• UDHR
• Canada
• India
• Montana

That could make Civic Blueprint look more aligned with the world than it really is.

Adversarial question:

• Are we choosing sources that flatter the principles while excluding founding texts built around order, religion, hierarchy, nationalism, or economic liberty in ways that would expose deeper misalignment?

2. Form mismatch

The corpus mixes unlike genres:

• constitutions
• rights charters
• international declarations
• organizational identity statements

Adversarial question:

• Are we treating a rights declaration and an institutional constitution as if they were solving the same problem, then reading both through one moral template?

3. False-overlap inflation

The comparative protocol tries to prevent false overlap, but the risk remains:

• "expression" becomes "truth infrastructure"
• "popular sovereignty" becomes "open design"
• "welfare" becomes "essential needs should not be held hostage to avoidable scarcity"

Adversarial question:

• Which current alignments are actually only loose family resemblance?

4. Corpus modernity bias

Many current findings rely on modern postwar texts. Older documents often look much worse on dignity, equality, or positive rights.

Adversarial question:

• Is the corpus discovering genuine cross-tradition alignment, or simply the convergence of late-20th-century rights constitutionalism?

5. Translation and excerpt risk

Several future sources will rely on translations and all long texts rely on selected excerpts.

Adversarial question:

• How much of the current synthesis depends on excerpt choices and English-language accessibility rather than the deeper document as a whole?

What would count as a serious objection

The corpus should not be treated as strong evidence for broad alignment if any of the following turn out to be true:

• the strongest overlap claims collapse when less rights-friendly sources are added
• multiple current mappings are better described as different-resolution than explicit-alignment
• peace, education, or social-rights language recur so strongly that the current principles genuinely look under-specified rather than merely differently phrased
• future multilingual review materially changes the meaning of high-confidence mappings

Initial adversarial hypotheses

Next steps for this exchange

• add multilingual and organizational sources before treating any first-wave synthesis claim as stable
• reclassify any suspiciously generous alignments from explicit to implicit or different-resolution where warranted
• use this adversarial review to decide which current "gap" candidates deserve a later principles-level discussion

Round 2 — Constructive response (June 2026)

Everything above is Round 1 — the April 2026 adversarial challenge. This round answers it. The key fact that makes a substantive response possible: when Round 1 was written, the challenge named five rights-forward texts (South Africa, UDHR, Canada, India, Montana) and worried the corpus was selected to flatter the principles. The corpus has since grown to 20 sources, and — decisively — several of the additions are exactly the order-, liberty-, structure-, and power-skeptical texts Round 1 said were missing. So Round 2 does not merely rebut the worries on paper; for two of the five it can check whether adding the missing kind of source actually moved the synthesis.

The corpus as it now stands (per the Alignment Matrix)

Grouping the 20 sources by the genre/era axis Round 1 cares about:

• Rights-forward modern (the set Round 1 said flatters the principles): South Africa, UDHR, Canada, India, Montana, Brazil, Germany Basic Law, EU Charter, California, Japan.
• Structural / liberty / power-skeptical / pre-modern (the set Round 1 said was missing): US Constitution (1787), US Bill of Rights (1791), Declaration of Independence (1776), France Declaration of Rights (1789), Massachusetts (1780), Texas (rights-skeptical toward concentrated state power).
• Regional / sovereignty / collective-security (not individual-rights instruments): UN Charter, African Union Constitutive Act.
• Organizational (non-state, member-governed): Mondragon, ICA cooperative identity, B Corp Declaration.

This is the material fact Round 2 turns on: the corpus is no longer all of one kind.

Verdicts on the five challenges

1. Source-selection bias — partly relieved, and the specific worry is now testable and confirmed-in-part. Round 1's sharp version of the worry — "we picked rights-forward texts, so broad alignment is an artifact" — can now be checked, because the power-skeptical/structural/pre-modern texts have been added (US 1787, Bill of Rights, Declaration, France 1789, Massachusetts, Texas, UN Charter, African Union). Adding them did not raise the alignment; it produced systematic asymmetry. Per the matrix's "major absences" column, every text in that subset is strong on constrained/reversible power and procedure and weak-or-absent on dignity-as-universal, essential needs, biosphere, and truth infrastructure. So the honest reading is: the broad dignity-and-rights overlap is concentrated in the rights-forward modern subset; it is not a property of the corpus as a whole. That both relieves the artifact charge (we did add the unflattering texts) and vindicates its substance (the alignment is narrower than "broad"). Residual worry that stands: the corpus still contains no theocratic, monarchical, explicitly hierarchical, or single-party founding text, and no Confucian- or Islamic-constitutional order in original form — the families most likely to expose genuine values divergence. The selection worry is retired for the order/liberty axis and live for the hierarchy/non-liberal axis.

2. Form mismatch (genre) — confirmed as a real hazard; partly mitigated, with one concrete discipline added. The matrix mixes constitutions, a rights charter, international declarations, a grievance/independence document, and organizational identity texts. Reading all through one moral template inflates apparent overlap. Existing mitigation is real: Exchange #18 already established the structural-vs-rights distinction, and the per-source memos record asymmetry rather than a single fused score. Added discipline from this round: never compute a cross-genre aggregate "alignment score," and treat the three organizational texts (Mondragon, ICA, B Corp) as evidence about governance/ownership form — member-governed organizations, not coercive states — relevant to Principle 5 and the (closed) ownership-form work, not as evidence for civilizational values convergence. Folding them into a convergence count is a category error.

3. False-overlap inflation — confirmed for specific mappings; this is the challenge that survives most intact. The Uniqueness Report's own watch list flags Principle 14 (truth/evidence) as thinly represented — consistent with the worry that free-expression clauses are being over-read as truth-infrastructure commitments. Verdict and concrete action: reclassify expression/information mappings from explicit to implicit or different-resolution unless a source pairs expression with institutionalized knowledge duties (public education mandates, disclosure/right-to-know, public-media or research provisions). Montana ("right to know") and California (open-governance/initiative norms) survive as closer-to-explicit; a bare speech clause does not. Open-design (Principle 10) mappings drawn off popular-sovereignty or representation default to different-resolution.

4. Corpus modernity bias — confirmed, and now sharply visible precisely because the older texts were added. The dignity/social-rights cluster is strongest in postwar texts (UDHR 1948, Germany 1949, Japan 1947, India 1950, Canada 1982, Brazil 1988, South Africa 1996, EU 2000). The pre-modern texts (US 1787, Bill of Rights 1791, Massachusetts 1780, France 1789) are strong on liberty and structure and weak on dignity-as-universal and positive rights. So a meaningful part of what reads as "alignment" is the convergence of late-20th-century rights constitutionalism, not a trans-historical human consensus. This directly tempers the convergence framing inherited from Exchange #18 and leaned on by Exchange #20: until pre-modern and non-Western-in-original sources are added, the synthesis should say "modern rights-constitutionalism convergence," not "cross-tradition convergence."

5. Translation and excerpt risk — stands, largely untested, but narrower than feared for the current claims. Most of the present 20 are English-original or carry authoritative English texts (the US set, Canada, UDHR, UN Charter, South Africa, India's English text, EU, ICA, B Corp), so the current synthesis is less translation-exposed than Round 1 anticipated. But Germany, France, Japan, Brazil, the African Union, and Mondragon rest on translation/excerpt and have had no native-language review, and every long constitution here is read through selected articles. Verdict: unresolved; any high-confidence claim resting on the translated subset stays provisional and is gated on native-language review.

Verdicts on Round 1's three initial adversarial hypotheses

• H1 — corpus over-represents rights-forward modern constitutionalism. Round 1 named its own test: "adding mixed/conservative/structurally thin sources without changing the synthesis much." That test has partially run (US Constitution, Bill of Rights, Declaration, France, Massachusetts, Texas, UN Charter, AU were added). The synthesis did change — from "broad overlap" to "concentrated overlap + systematic asymmetry." So H1 is upgraded toward confirmed, but its consequence is a reframed synthesis (concentration, not breadth), not a collapse. Confidence: from working hypothesis to established by corpus structure for the narrow claim "alignment is concentrated in the rights-forward modern subset."
• H2 — over-reading openness/truth into free-expression clauses. Supported by the uniqueness watch list; it is the driver behind the challenge-3 reclassification. Confidence: working hypothesis (modestly strengthened).
• H3 — peace is more underdeveloped than the synthesis admits. The Gap Analysis independently surfaces peace/anti-war from three sources (UN Charter, Japan, African Union) as the single strongest recurrent gap. That is enough independent recurrence to earn an explicit principles-level discussion, rather than silent absorption into Principle 17. Confidence: from speculative to working hypothesis.

Reclassifying the four early synthesis claims

1. "Civic Blueprint is less normatively unusual than it first appears" — partly true, materially narrowed. True for the dignity / accountable-power / openness / justice cluster within the modern rights subset; false for the AI / biosphere / open-moral-consideration / competence-not-theater cluster (genuinely unusual, per the uniqueness report); and overstated against the structural, pre-modern, regional, and not-yet-sampled non-liberal traditions. Reframed claim: Civic Blueprint's rights-and-dignity core is well-companioned by modern rights constitutionalism; its technology-, ecology-, and moral-expansion core is not.
2. "The most unusual principles are technology-, ecology-, and moral-expansion-related" — confirmed (uniqueness report: P3 AI, P11 biosphere, P15 moral circle, with P9 competence-not-theater close behind). No change.
3. "Structural constitutions and rights declarations need different comparison" — confirmed. The asymmetry is now visible across the whole structural subset, not just the US proof-of-concept; the method holds and should be kept.
4. "The corpus pressures more through subprinciples and tension notes than wholly new principles" — mostly confirmed, with one earned exception. Peace is the one recurrent gap that has cleared the bar for a principles-level discussion — even if that discussion lands on a Principle 17 tension note rather than a new principle.

Which gap candidates deserve a principles-level discussion

Epistemic status of Round 2's claims

Honest limits of this round. Both Round 1's challenge prompts and this response were produced in the same authoring lineage, so this is a same-lineage constructive round, not the cross-lineage adversarial pass the Adversarial Review Protocol §2 makes the default. Every confidence above is an upper bound pending (a) an independent-model-lineage or comparative-politics/constitutional-law human pass on the reclassified synthesis, and (b) corpus expansion into the non-liberal and pre-modern families that are still absent. No principle is edited on the strength of this round; the gap verdicts are routed for discussion, not enacted.

What Round 2 hands forward

• To the corpus synthesis artifacts: apply the challenge-3 reclassification (expression/sovereignty mappings → implicit/different-resolution) and record the "modern rights-constitutionalism convergence, not cross-tradition convergence" framing. Cross-reference notes added to the Alignment Matrix, Gap Analysis, and Uniqueness Report.
• To Round 3 (queued June 12, 2026): a cross-lineage adversarial pass whose sharpest targets are the "concentrated, not broad" reframe and the absent non-liberal families. The self-contained, copy-paste hand-off packet is below (Round 3 — Adversary hand-off packet).
• To the steward: one decision was genuinely theirs — whether to open the peace discussion now or hold it. Steward decision (June 12, 2026): HOLD. The peace discussion is deferred until the corpus adds non-liberal and pre-modern sources, so it is grounded in a less skewed corpus rather than the current modern-rights-forward one (which is exactly the bias Round 2 surfaced). The other three gaps remain articulation/subprinciple work, not new principles. The hold is logged as a standing item, not a closed question — peace re-enters the queue when corpus expansion delivers the missing families.

Round 3 — Adversary hand-off packet (self-contained — copy-paste)

How to run this (run June 12, 2026 — results in the Round 3 results section below). Round 1 (challenge) and Round 2 (response) were both produced in the same authoring lineage, so the convergence problem the Adversarial Review Protocol exists to fix is only caveated, not removed. Round 3 is the cross-lineage pass. Run the packet below on a different model family (Grok, GPT, or Gemini — Protocol §2 Option D) and/or give it to a human comparative-politics / constitutional-law / global-intellectual-history reviewer. The packet is self-contained — it carries the corpus list, the reclassified claims as assertions to test (Option B), reduced context (Option A — the reviewer gets the claims and corpus, not this whole thread), and a domain lens (Option C). Paste everything inside the rule below as the prompt, in a context that does not include this exchange. Record each lineage's verdict, blind to the others. (This packet is retained as the reusable artifact; the June 2026 run is synthesized below.)

You are reviewing a synthesis claim set from Civic Blueprint, a project that maintains a 17-principle civic document and compares it against the world's formation documents (constitutions, rights charters, international declarations, and organizational founding texts) to test, sharpen, and challenge its own principles.

Your role is adversarial. Do not extend or refine these claims. Find what is wrong, overstated, or artifact. Attack the strongest version of each claim (steelman first, then break). Agreement is not the goal.

Your lens: a comparative-politics / constitutional-law scholar who studies cross-tradition constitutional borrowing and is skeptical of "values convergence" narratives — alert to selection effects, Whig history, and the difference between textual overlap and functional equivalence.

The corpus under analysis (20 sources)

• Rights-forward modern: South Africa, UDHR, Canada Charter, India, Montana, Brazil, Germany Basic Law, EU Charter, California, Japan.
• Structural / liberty / power-skeptical / pre-modern: US Constitution (1787), US Bill of Rights (1791), Declaration of Independence (1776), France Declaration of the Rights of Man (1789), Massachusetts (1780), Texas.
• Regional / sovereignty / collective-security: UN Charter, African Union Constitutive Act.
• Organizational (member-governed, non-state): Mondragón, ICA cooperative identity, B Corp Declaration.

The corpus contains no theocratic, monarchical, explicitly hierarchical, or single-party founding text, and no Confucian- or Islamic-constitutional order in its original form.

Method note

Each source has its own alignment memo recording where it overlaps the 17 principles, where it diverges, and what it adds. No single cross-genre "alignment score" is computed. Civic Blueprint's principles most relevant below: P1 dignity, P2 essential needs, P3 AI augments (not replaces) democratic accountability, P9 institutions designed for competence not theater, P10 the future built in the open, P11 functioning biosphere, P14 truth/evidence as public goods, P15 the moral circle stays open, P17 principled constraints on public power.

Claims to test (evaluate each: supported / weak / contested / falsified, with reasoning)

• R3-1. Adding the structural / pre-modern / power-skeptical texts (US 1787, Bill of Rights, Declaration, France 1789, Massachusetts, Texas) and the regional texts (UN Charter, African Union) produced systematic asymmetry — strong on constrained/reversible power and procedure, weak-or-absent on dignity-as-universal, essential needs, biosphere, and truth infrastructure — rather than raising overall alignment. Falsified if those texts in fact carry strong dignity / positive-rights / ecological / epistemic content that the "absences" reading missed.
• R3-2. The dignity-and-rights overlap is concentrated in the rights-forward modern subset, not a property of the corpus as a whole. Falsified if overlap is roughly uniform across eras and genres, or if a defensible aggregate measure shows broad overlap.
• R3-3. A meaningful part of the apparent "alignment" is the convergence of late-20th-century rights constitutionalism, not trans-historical cross-tradition convergence. Falsified if pre-modern or non-Western-in-original texts show the same dignity / positive-rights commitments — i.e., the convergence predates and transcends postwar constitutionalism.
• R3-4. Free-expression→truth-infrastructure (P14) and popular-sovereignty→open-design (P10) mappings are over-read and should be downgraded to implicit / different-resolution unless a source pairs expression with institutionalized knowledge duties (public education, disclosure / right-to-know, public media). Falsified if sources routinely pair expression with such duties, making the explicit mapping earned.
• R3-5. Peace / anti-war is the only candidate gap that has earned a principles-level discussion (recurring independently across UN Charter, Japan, African Union); the other three candidates — social-rights articulation, education / civic formation, international cooperation — are articulation/subprinciple work, not new principles. Falsified if one of the other three recurs as strongly and independently, or if peace is already adequately carried by P17.
• R3-6. The technology / ecology / moral-expansion cluster (P3 AI, P11 biosphere, P15 open moral circle, P9 competence-not-theater) is genuinely unusual and survives challenge. Falsified if the broader constitutional tradition carries these commitments in substance under other names.

Required adversarial probes

1. Absent families. For each of R3-1…R3-6, state whether the omission of theocratic / monarchical / hierarchical / single-party / Confucian / Islamic founding texts makes the claim an artifact of who was left out. Name the specific source(s) whose addition would most change each verdict.
2. Steelman the rival. Construct the strongest version of "Civic Blueprint's core does reflect a broad cross-tradition human convergence" and assess whether it survives. If it does, R3-2 and R3-3 are in trouble.
3. Falsifiability under revision (Protocol §5, Q6). Round 2 revised the synthesis from "broad alignment" to "concentrated alignment + asymmetry." Did that revision expose new falsifiers, or did it dissolve the old ones into an unfalsifiable hedge? Be specific.
4. Functional vs. textual. Where Round 2 read overlap or asymmetry off the text, identify cases where functional/constitutional practice would flip the reading.

Output format

For each claim R3-1…R3-6: a one-paragraph verdict + confidence (established / working hypothesis / speculative / contested / falsified). Then a short epistemic-status table, a ranked list of the absent sources most likely to change the conclusions, and any analytical tradition or perspective absent from the synthesis whose inclusion would change it.

Round 3 results — Cross-lineage adversarial pass (June 2026)

Run on independent model families per Adversarial Review Protocol §2 Option D, against the hand-off packet above. Two clean passes are recorded (Grok 4.3 and GPT-5.5). A third (Gemini 3.1 Pro) was discarded as contaminated — see the process note. Statuses below take the more demanding defensible reading wherever the two clean passes diverge, per the protocol's purpose.

Process note — the discarded Gemini pass (a reusable lesson)

The Gemini pass was run as a file-writing agent and wrote its review directly into this exchange. That defeats the packet's core discipline: Option A (reduced context) only works if the adversary sees the claims and corpus, not the thread. An agent with read/write access to the exchange has the Round 1 framing, the Round 2 reasoning, and the steward's decisions in context — exactly the bias the packet was built to withhold — so its output cannot count as a clean blind pass. It was removed from the file and excluded from the count. Reassuringly, no signal was lost: every substantive point it raised (functional-vs-textual; the unfalsifiable-hedge worry; education as a rival to peace; Indigenous/religious texts for the ecology/moral-circle cluster) was independently raised by Grok and/or GPT below. Lesson: run packet adversaries in a context that does not include the exchange (a fresh session, or a model without repo access), and have the steward paste the result back — never hand the packet to an agent pointed at the exchange file. (Candidate for promotion into Protocol §2; steward's call.)

Per-claim synthesis (Grok 4.3 + GPT-5.5)

The two clean passes diverge sharply in temperature — Grok found all six claims survive as working hypotheses or stronger (zero falsified); GPT pressed harder and contested three. Taking the more demanding reading:

• R3-1 (asymmetry from the structural/pre-modern additions) — Supported but narrowed. Both agree the asymmetry is real (strong on constrained power/procedure, weak on the rest). GPT's correction stands: "weak-or-absent on dignity-as-universal" is too broad — France 1789, the Declaration, the UN Charter, and Massachusetts carry equality / worth / education language. Refinement: the asymmetry is against Civic Blueprint's full dignity-plus-material-stability-plus-biosphere-plus-truth package, not against dignity per se. Live vulnerability (both flag): functional vs. textual reading.
• R3-2 (overlap concentrated in the rights-forward modern subset) — Contested (downgraded). Clean split: Grok supported, GPT contested. The claim holds only if "dignity-and-rights" means postwar liberal/social rights; under a broader definition (natural rights, equality-before-law, anti-domination, divine dignity, socialist social rights, subject-rights), the overlap is not concentrated. The still-absent Iran ("the exalted dignity and value of man") and PRC (equality, work, welfare, education, ecology — inside a party-state frame) texts are named falsifiers. This is the round's sharpest hit.
• R3-3 (modern rights-constitutionalism convergence, not cross-tradition) — Contested. Both passes flag a denominator/Whig problem: the claim risks making cross-tradition convergence disappear by definition. Many traditions plausibly converge at a higher abstraction (human worth, anti-arbitrariness, public welfare, education, constrained rule); what is distinctively modern is Civic Blueprint's particular resolution (liberal-democratic, pluralist, ecologically constrained, accountability-heavy). The constitutional-borrowing literature adds that the modern convergence may itself be an artifact of postwar US/European influence and decolonization-era borrowing, not organic discovery. Defensible residue: the explicit textual convergence on CB's package is modern.
• R3-4 (expression→truth-infrastructure and sovereignty→open-design over-read) — Supported (the strongest claim). Both passes endorse it; the clean reasoning is the negative-liberty/positive-duty distinction (a right to speak is not a state duty to maintain an epistemic commons). Established for the free-expression→truth-infrastructure downgrade on the current corpus; working hypothesis for the popular-sovereignty→open-design downgrade. This is the one reclassification safe to act on now.
• R3-5 ("peace is the only earned gap") — Contested / weakened. GPT (and the discarded Gemini pass) reject the "only": education / civic formation recurs independently across Massachusetts, UDHR, ICA, and Mondragón, and social-rights articulation recurs strongly across South Africa, Brazil, UDHR, and India — both at strength comparable to peace, and both would recur more once Iran/PRC/Saudi are added. P17's existing nonviolent-theory-of-change also keeps the peace falsifier live. Net: peace has earned a discussion but is not uniquely earned — education/civic formation is a co-candidate. This reinforces the steward's hold and retires the "only."
• R3-6 (tech/ecology/moral-expansion cluster is genuinely unusual) — Supported but split within the cluster. P3 (AI) and P15 (open moral circle) remain genuinely unusual across both passes — robust. P9 (competence-not-theater) and especially P11 (biosphere) have functional analogues — modern environmental constitutions, PRC-style ecological provisions, and Latin-American rights-of-nature / Indigenous stewardship (Ecuador, Bolivia, the Haudenosaunee Great Law of Peace) — so "unusual" for those two narrows to "more central in CB than in the corpus," not "absent elsewhere."

Cross-cutting findings (robust across both clean passes)

1. The absent non-liberal / pre-modern / Indigenous families are load-bearing, not a residual caveat. Both passes independently rank them as the top falsifiers and name specific content (Iran's dignity language; the PRC's welfare/ecology; Ecuador/Bolivia rights-of-nature). This converts "corpus expansion" from a background task into the gating move for hardening the synthesis — and strongly vindicates the steward's June 12 peace-hold.
2. Functional-vs-textual is the deepest unexamined seam. The synthesis reads overlap and asymmetry off text; reading off constitutional function/practice could flip R3-1, R3-2, and R3-5. The asymmetric danger (GPT): authoritarian constitutions may textually promise rights, welfare, ecology, and competence while functionally violating accountability; liberal constitutions may textually protect rights while functionally failing material stability. New required discipline: tag every mapping textual vs. functional, and never compare one source functionally against another textually.
3. The convergence framing needs a borrowing/influence lens (constitutional-migration critique — Frankenberg, Hirschl). "Concentration in modern rights texts" may be a predictable outcome of postwar hegemony and decolonization borrowing incentives, not an organic cross-tradition discovery. This must be settled before any convergence claim goes public.

Updated epistemic status after Round 3

Net verdict: Round 3 does not falsify the synthesis, but it wounds R3-2, R3-3, and R3-5, narrows R3-1 and R3-6, and confirms R3-4. The synthesis is directionally sound but over-claimed in two places (concentration; "peace only"), and its biggest exposure is now precise and named: the absent families and the textual-vs-functional seam.

What Round 3 hands forward

• Corpus expansion is now the gating move (not background): prioritize the named non-liberal / pre-modern / Indigenous families — Iran, PRC, Saudi Basic Law, a Confucian-influenced order (Meiji operation / Singapore), and rights-of-nature / Indigenous texts (Ecuador, Bolivia, Haudenosaunee). These are where R3-2, R3-3, R3-5, and R3-6 will be decided.
• Adopt the functional-vs-textual discipline in the Comparative Alignment Protocol and the per-source memos before the next synthesis pass. Protocol half enacted June 14, 2026 — codified as the §4 functional-vs-textual tagging discipline (tag every mapping textual/functional; never compare one source functionally against another textually). The per-source-memo half is folded into the next synthesis pass (after corpus expansion).
• R3-4 enacted (June 12, 2026): the expression→truth-infrastructure downgrade has been applied to the per-source memos — six bare-expression Principle 14 mappings (Japan, India, Canada, Texas, US Bill of Rights, France) moved implicit-alignment → different-resolution, while the nine paired-knowledge-duty sources R3-4 itself excepts (South Africa §32, California, Montana, UDHR, Massachusetts, Germany, EU, ICA, Mondragon) were preserved and annotated. The working-hypothesis half — popular-sovereignty→Principle 10 ("built in the open") — is deferred to the corpus-expansion / functional-vs-textual round (logged in the alignment matrix).
• Peace-hold reinforced, and widened: when corpus expansion delivers the missing families, re-evaluate peace and education/civic-formation together (the "peace only" framing is retired), not peace alone.
• Round 4 (response/v2) reserved: integrate these wins into a v2 synthesis once the corpus is expanded; an external human comparative-constitutional reviewer remains the largest unmet input.