Coordination Triad (See / Decide / Act) — Combined Exchange (Path C)

Status (June 2026): Active; Rounds 1–2 complete; Rounds 3–5 reserved. Round 1 restates the coordination-triad riff §10 claim set as falsifiable claims in two tiers — per-leg components (see M1–M5 imported from #25 at status; decide D1–D5; act A1–A4) and triad-level T-series (T1–T4b) — each with a falsification condition and its honest status carried from the riff, not upgraded. Round 2 applied first-pass adversarial review (Adversarial Review Protocol Options A + C — reduced context + a coordination-skeptic lens), concentrating pressure on the T-series and the "does combining add value over Path A?" question. Headline: the combined frame survives — it was not falsified — but the case for a combined exchange over Path A did not pass cleanly. T2 (interdependence) is downgraded from supported to working hypothesis (the §10.12 single-leg cell is confounded by a deliberation-catalogue sampling frame, and the "the act leg must bind" sharpening quietly reintroduces a sequence); T3 (distinctness over #24) drops to contested (its variation lives in the thin corporate / AI / financial rows while the well-grounded rows risk collapsing to "everything bottlenecks at act/enforcement"); and the deepest finding is that the surviving combine-rationale (T2 + T3) points repeatedly at act/enforcement as a candidate upstream leg — a sequence claim, a near-inversion of the original see-first bet, not the co-equal interdependence the triad asserts. The remaining rounds (external human review; response; synthesis) are reserved pending steward direction.

Why this exchange: The coordination-triad riff (v2.8, June 3, 2026) developed a combined claim set for the See / Decide / Act triad to the point where its §10.6 readiness bar is substantially met / steward-callable — the hinge claim T2 (interdependence) is supported by a pre-registered systematic test (§10.12: single-leg durable gain 0/4 vs multi-leg 10/14, no clean single-leg-alone success), T3 (distinctness over #24) leans supportive, and the one strong claim that could have falsified the combine-rationale (strong T4b) was settled without collapsing it (§10.13). The steward approved opening Path C (riff §9): the one combined exchange that can test the triad as a single object, because the load-bearing claims — the T-series — are by construction untestable leg-by-leg. This exchange is where those claims survive or are falsified together.

What is not in scope. This exchange does not re-open or re-grade the see leg: M1–M5 are imported at Exchange #25 status, the #25 routing decision of hold governs, and the exchange uses them (especially M1-weak as the seed of T2) rather than re-litigating them. It does not promote the triad to project doctrine or to a Phase 3 front-door brief — #25's single most robust cross-model finding was that the coordination reframe is not ready for doctrine, and that discipline binds the triad too; the steward's "protocol between the nodes" is a felt north-star sense (riff §1.1), not a claim this exchange tests. It does not rest the case for combining on strong T4b (the "single unifying object" claim, falsified in §10.13) — the combine-rationale rests on T2 (supported) and T3 (lean-supportive), and the honest frame is promising but explicitly not doctrine, with a readiness bar that is substantially met / steward-callable, not proven. It inherits, unchanged, the riff §1.2 privacy firewall: idea layer only; the originating personal/biographical substrate is excluded.

Dependencies

Round 1 — The combined claim set under review (steward + agent, June 2026)

1.1 What this round does

Round 1 restates the riff §10 claim set as falsifiable claims, each paired with what would falsify it and its current honest status carried from the riff — not upgraded. A claim without a stated falsification condition is not a claim adversarial review can engage. The set is organized in two tiers (§1.3 per-leg components, §1.4 triad-level T-series), and the exchange's design principle (§1.2) is stated up front because it governs where Round 2's pressure should fall.

This is the #25 Round-1 pattern applied to a combined object. Two things make it different from #25:

1. The see leg is imported, not stated fresh. M1–M5 ride at their #25 status (§1.3.1); the #25 hold governs; they are not re-litigated here.
2. The load-bearing tier is the T-series, not the per-leg claims. Per §1.2, a combined exchange earns its existence only by testing what separate ones cannot. So the per-leg claims are carried as components, and the analytical weight sits on the triad-level claims (§1.4) — above all T2 and T3.

1.2 The design principle (riff §10.1) — what a combined exchange must test that separate ones cannot

A combined exchange earns its existence only by testing claims a single-leg exchange could not. If the only claims were per-leg, three separate exchanges (Path A) would be cleaner and lower-risk — each leg tested on its own evidence base, without the overhead of holding three at once. So the set has two tiers:

• Per-leg claims — the components. The see leg already has these (M1–M5, tested in #25); decide (D1–D5) and act (A1–A4) carry first-pass sets.
• Triad-level claims — the reason to combine (the T-series). These are claims whose truth depends on the interaction of the three legs and that are, by construction, untestable leg-by-leg. The T-series is the load-bearing justification for Path C. Without it, "combined" is a concatenation and Path A wins.

The combined set is therefore not a concatenation of the per-leg claims; it is the distinct T-series, with the per-leg claims carried as its components. The operational consequence for Round 2: adversarial pressure should concentrate on the triad-level claims (§1.4) — above all T2 and T3 — and on the question "does combining add value over Path A?", not on re-litigating the per-leg claims (and emphatically not on re-opening the imported see leg).

1.3 Tier 1 — Per-leg claims (the components)

1.3.1 See-leg claims (M1–M5 — imported from #25 at status, not re-opened)

The see leg was developed in the shared-mirror riff and tested to a resting point in Exchange #25 (Rounds 1–2 + a six-run cross-model Ring-1 test; routing = hold). Its claims and their full falsification conditions live in #25 §1.2; they are imported here at status and are not re-graded. The #25 hold governs: if Round 2/3 surface new evidence that materially moves an M-claim, re-grading is a steward decision, not an automatic change.

1.3.2 Decide-leg claims (D1–D5)

First-pass and origin-balanced; the §10.9 evidence cleared the Research-Protocol §4.2 verification bar (§10.11). The cross-cutting pattern is theory/lab strong, real-world-at-scale weak-to-mixed. Statuses are carried from riff §10.3 / §10.9.3, not upgraded.

1.3.3 Act-leg claims (A1–A4)

First-pass and origin-balanced (§10.9.2, verified §10.11); same theory-strong / real-world-mixed pattern. Statuses carried from riff §10.4 / §10.9.3.

1.4 Tier 2 — Triad-level claims (the reason to combine: the T-series)

These are the claims a single-leg exchange cannot test, because each is a claim about the interaction of the three legs. This is where Round 2's pressure belongs. Statuses are carried from riff §10.5 / §10.8.6 and the two gate-4 rigor tests (§10.12 T2-systematic; §10.13 T4b-corpus) — not upgraded.

T2 is the hinge. If it holds, Path C is the only path that can test it — the strongest case for combining. If it failed (one leg genuinely upstream), the frame would revert toward a sequence and separate exchanges (Path A) would suffice. The §10.12 systematic test is why T2 enters this exchange supported rather than merely leaning-supportive.

1.5 The combine-rationale, stated honestly (on T2 + T3 — not strong T4b)

The reason to run one combined exchange rather than three separate ones rests on exactly two surviving load-bearing claims, and the exchange should say so plainly:

• T2 (interdependence) — supported. The legs co-determine durable coordination gain; no single leg delivers alone. This is a claim about the interaction, so only a combined object can carry it. It is the primary warrant.
• T3 (distinctness over #24) — lean-supportive (provisional). The triad × layers matrix gives a per-cell binding-constraint prediction the layer-lens alone does not. This is the secondary warrant, and it is explicitly provisional (the corporate / AI / financial rows are thin).
• T4a / T4b — weak-supported, strong-falsified — carry weak support only. The shared interface lever (weak T4a) and the family-resemblance signature (weak T4b) are genuinely cross-leg, so they still point at a combined rather than a separate treatment — but the strong "single literal unifying object / north-star protocol" did not survive (§10.8.5, §10.13). The combine-rationale does not rest on strong T4b, and must not be argued as if it did.

So the honest headline the exchange opens with: the case for combining is real but bounded. It is promising but explicitly not doctrine. The riff §10.6 readiness bar is substantially met / steward-callable, not proven — the hinge claim survived a systematic test and the one strong claim that could have falsified the rationale (strong T4b) is settled without collapsing it (the rationale never rested on it). Whether to treat coordination as a single three-function design target (Path C) rather than three separate problems (Path A) is precisely what the remaining rounds are for.

1.6 Honest residual caveats this exchange inherits

These are evidence-rigor caveats the exchange carries forward as-is (they are about evidence, not claim shape, so Round 1 does not re-run them — riff v2.8):

• D3b (range/score) and D4b (liquid democracy) owe their own evidence. They are split out as deliberately weaker and must not borrow D3a's QV theorem or D4a's sortition base; each is owed its own Research-Protocol T2 pass.
• T2's single-leg cell is thin. Participedia / OECD structurally bundle see + decide, so the systematic test's single-leg cell is only 4 (6 under a generous recode), and ProZorro sits on the coding seam (durable, but coupled to the DoZorro enforcement pipeline, so not a clean see-only success). The pattern holds, but the cleanest counter-test would need a frame richer in pure single-leg reforms.
• The T4b control set was under-powered. Only the second law of thermodynamics was a clean hard-limit control (n = 1); the other four controls (Arrow, Gibbard–Satterthwaite, Myerson–Satterthwaite, Holmström) belong to the same assumption-based family as the corpus. The direction (strong T4b falsified) is unambiguous and converges with §10.8.5, but a sharper follow-up would draw genuinely irreducible limits.

Carrying these in Round 1 is the discipline: the combine-rationale is built on what survived (T2, T3), with the residuals named rather than buried.

1.7 What the claims are not

• Claims that the triad frame is correct. Correctness is a downstream (doctrine) question. The claim set asks whether the triad-as-one-object survives structured pressure well enough to justify a combined treatment over Path A — not whether the frame is true.
• A re-litigation of the see leg. M1–M5 ride at #25 status; the #25 hold governs; the exchange uses them, above all M1-weak as the seed of T2.
• A north-star promotion or a Phase 3 front-door brief. The steward's "protocol between the nodes" is a felt direction (riff §1.1), not a claim tested here. #25's most robust finding — not ready for doctrine — binds the triad too. Note what is deliberately absent from the T-series: any claim that the triad is the north star or belongs in the Phase 3 front door.
• Claims that the decide/act mechanisms are solved. The D/A evidence is first-pass and theory/lab strong, real-world-at-scale weak-to-mixed; which mechanisms survive a fuller Research-Protocol T2 tiering is a Round-2-and-beyond question, not asserted here.
• Claims about the steward's standpoint. The riff §1.2 privacy firewall is upstream and recorded in the riff; idea layer only.

1.8 Round plan

Round 1 is complete; Rounds 2–5 are reserved pending steward direction (the steward asked to open and register the exchange, not to run the full sequence autonomously).

1.9 The relationship this exchange must hold (consolidate, not fragment)

This exchange is positioned to consolidate the project's coordination thread, not to add a competing one:

• Imports and respects #25 (the see leg). It carries M1–M5 at status and uses the negative finding "see does not stand alone" as the seed of T2. It does not re-open the see leg or disturb its hold.
• Tests distinctness over #24 (the layers). T3 is the test that the triad × layers matrix adds a per-cell binding-constraint prediction #24's diffuse-sovereignty layering alone does not give. #24 is itself unresolved (Round 2, not clean); this exchange tests against it, it does not depend on its resolution.
• Subsumes #20 (Social Slop) as one named see-layer failure mechanism, generalized under the see leg.
• Answers the riff §6 sequencing question (positions A strict-sequence / B parallel / C interdependent / D layer-dependent). T2's support already leans the evidence against strict-sequence A and toward B/C/D; Round 2 is where the remaining discrimination among B/C/D is pressed.

Round 2 — First-pass adversarial review (Adversarial Review Protocol, Options A + C)

2.1 Round 2 framing — what this round is and is not

The Protocol's mandate. Per the Adversarial Review Protocol §1, the adversarial contributor identifies the weakest claims and argues against them as forcefully as the evidence allows — not disagreement for its own sake. Per §2 this round uses Option A (reduced context — the claims evaluated as standalone assertions, not inside Round 1's narrative) plus Option C (a domain-specific skeptical lens). Per §1.2 of this exchange, pressure concentrates on the T-series — above all T2 and T3 — and on the question "does combining add value over Path A?"

The lens for Round 2. You are a scholar of institutional design and collective action who has spent a career watching grand unifying frameworks — "systems thinking," "governance architecture," "the new science of coordination" — get proposed, funded, and celebrated, and then either (a) fail to produce a single prediction the component literatures did not already give, or (b) "succeed" only by being abstract enough that no result could contradict them. You have seen interdisciplinary syntheses that turned out to be relabeling with a fresh vocabulary. You are not against synthesis — you have built some — but you owe this project the specific reasons most "unify the three functions" moves added nothing load-bearing over just doing social choice, commons governance, and information integrity well, separately.

The non-protocol caveat the agent owes the steward. The adversarial reviewer here is the same agent lineage that wrote the riff, Round 1, and this round. Option A reduces but does not remove that confound; Round 3 (external human review) is the real gap-closer. Treat every confidence number below as an upper bound, and read §2.5 for where the agent suspects it is steelmanning itself.

2.2 The discipline this round holds (what it does not touch)

• It does not re-litigate the imported see leg. M1–M5 ride at #25 status; the #25 hold governs. Round 2's pressure is on the T-series and the combine-vs-Path-A question, not on the M-claims.
• It does not re-grade the riff's evidence; it argues about what that evidence supports. The §10 grounding cleared the Research-Protocol §4.2 verification bar (§10.11), and the two rigor tests (§10.12 T2-systematic, §10.13 T4b-corpus) stand. Round 2 does not re-run them. It contests the inference from those results to the triad-level claims — which is the adversarial role's proper job.
• It runs no new research sweep. Candidate disconfirming cases named below (§2.3) are flagged as open challenges for Round 3 / a research pass, explicitly not asserted as verified findings — the first-pass agent cannot adjudicate them, and asserting them would breach the Research Protocol's tiering discipline.
• It promotes nothing. No doctrine, no Phase 3 front-door, no north-star. Round 2 pressures claims; it makes no routing decision (that is Round 5, steward-held).

2.3 Adversarial findings — the T-series first

2.3.1 Against T2 (interdependence — the hinge) — rated supported entering Round 2

The steelman the project would want. The §10.12 systematic test is genuinely better than the curated five: an external, pre-committed frame (Participedia + OECD), a pre-registered coding scheme, a mandatory disconfirming search, and a hostile-recode robustness check. Single-leg durable gain 0/4 vs multi-leg 10/14 (~71%), surviving the recode at 33% vs 67%, with no clean single-leg-alone success found — that is the strongest single piece of triad-level evidence the project has.

Attack 1 — the single-leg cell is not merely thin, it is structurally confounded by the sampling frame. Participedia and the OECD deliberative-wave database are catalogues of deliberative / participatory innovations — by construction they are populated with multi-leg (see + decide) cases. The "single-leg" cell (n = 4) is whatever pure-transparency reforms happened to slip into a deliberation catalogue. So the test did not ask "do single-leg reforms fail?"; it asked "do the multi-leg cases that fill a multi-leg database outperform the few single-leg stragglers that database happens to contain?" — which is close to tautological. The universe where the strongest single-leg-alone durable successes actually live — central-bank independence, the secret/Australian ballot, national FOIA/RTI regimes, standardized public accounting, independent statistical agencies — is outside the frame the test could see. The riff half-concedes this ("Participedia / OECD structurally bundle see + decide"), but then still reads 0/4 as support for T2. An adversary reads the same fact as a fatal selection effect.

Attack 2 — "durable coordination gain" is coder-defined, and the coder's lineage authored the frame. The riff admits the outcome measure "mixes institutional persistence, policy output, and corruption metrics," and the coding was an agent-run (spot-checked, but same lineage). Holding pure-transparency reforms to a downstream coordination outcome they were never designed to deliver alone risks assuming the conclusion: of course a see-only reform "fails" at an act-layer outcome — that is true by the decomposition, not by the world.

Attack 3 — candidate disconfirmers the search may have missed (open challenges, not assertions). A genuine adversary names the cases the disconfirming search should have to beat: the secret ballot (a near-pure decide-layer reform widely credited with durably curbing vote-buying and coercion); central-bank independence (a near-pure decide/allocation reform with durable macro-coordination effects); national RTI/FOIA regimes (near-pure see — the riff coded India's RTI as "not durable alone," a coding a transparency scholar would contest hard). If even one survives scrutiny as a clean single-leg-alone durable success, T2 drops from supported to contested. The first-pass agent cannot adjudicate these; Round 3 / a pre-registered research pass must.

Attack 4 — the "act must bind" sharpening quietly smuggles a sequence back in. The riff sharpened T2 to "coupling is necessary but not sufficient — the act/binding leg must actually hold" (the three multi-leg failures were all cases where the binding act leg did not). But look at what that does: it makes the act/enforcement leg the load-bearing thing. "Non-binding participation changes nothing; you need enforceable authority" is the vTaiwan tiger-without-teeth lesson generalized — and it is a claim that the act leg is upstream, not a claim that three legs are co-equal and interdependent. The sharpening that was meant to strengthen T2 actually points away from interdependence and toward a different sequence (act-first) than the original see-first bet.

Verdict on T2. Working hypothesis (~0.55, upper bound) — downgraded from the riff's supported (moderate). The systematic test is real and the directionality is suggestive, but the single-leg cell is confounded by a deliberation-catalogue frame, the outcome measure is coder-defined toward the hypothesis, named disconfirmers are unadjudicated, and the sharpening reintroduces a hierarchy interdependence was supposed to dissolve. T2-as-interdependence is not falsified, but it is materially weaker than "supported," and part of its surviving evidence actually supports a rival (act-upstream) reading.

2.3.2 Against T3 (distinctness over #24) — rated lean-supportive entering Round 2

The steelman. The binding-constraint column in the §7.1 matrix is not constant — platforms bind at act / the decision threshold, knowledge-production at see-with-an-act-gap, state-electoral at decide↔act coupling, commons at act-via-decide. Variation across layers is a per-layer prediction #24's layer-lens alone does not give.

Attack 1 — the column nearly collapses on exactly the rows that are well-grounded. Read the confidently assigned cells: platforms → act; state-electoral → decide↔act, but the §10.12 sharpening says the failure is the act/binding side; commons → act-via-decide; knowledge-production → see with an act gap. That is act, act, act, act-gap. The variation the riff leans on to show distinctness lives almost entirely in the corporate / AI / financial rows — which the riff itself flags as thin / lightly grounded / speculative. So T3's distinctness is carried by its least-grounded cells, while the grounded cells trend toward the failure mode the riff named: if the binding-constraint column collapses to one function, the triad is a relabel. On current evidence it is collapsing toward act/enforcement.

Attack 2 — distinct from #24, or distinct from a one-leg analysis? T3 claims distinctness over #24's layer-lens specifically. But "platforms fail at the act/decision threshold" is arguably already inside #24's diffuse-sovereignty treatment of the platform layer. The matrix supplies a vocabulary (see/decide/act) laid over #24's layers; what it has not yet supplied is a single concrete, transferable prediction that #24 plus the per-leg literatures do not jointly already make. Until it does, T3 sits exactly on the "non-trivial vs. trivial restatement" line that #24's own C2 adversarial review drew — and fails it for now.

Verdict on T3. Contested (~0.4, upper bound) — below the riff's lean-supportive. The honest test (one prediction the matrix makes that neither #24 nor the single-leg literatures give) has not been produced; the variation offered as evidence of distinctness is concentrated in the rows the riff admits are thin, while the grounded rows lean toward the collapse that would falsify it.

2.3.3 T1, T4a, T4b — brief

• T1 (decomposition). Carried at first-pass plausible; untested. One added adversarial point: the decomposition's completeness is most strained by legitimacy — the load-bearing unsolved problem in #24 (its Finding 4) and a recurring theme in #25. Where does "legitimacy" live in see/decide/act? If it is a fourth thing the triad cannot localize, T1's completeness fails; if it is smeared across all three, the decomposition is not clean. Unchanged status, but legitimacy is the sharpest test case for a future T1 corpus.
• T4a / T4b. Already strong-falsified in the riff (§10.8.5, §10.13); Round 2 does not re-attack them. It records the consequence: with T4-strong gone, the entire load of the combine-rationale has shifted onto T2 and T3 — and §2.3.1–§2.3.2 just moved both of those down. That compounding is the substance of §2.4.

2.3.4 The per-leg components — not re-litigated, but two flags

The per-leg claims are carried, not re-opened (per §1.2). Two flags the adversary owes the steward, because they bear on the combined object:

1. Two of the three legs enter this exchange un-adversarially-tested. The see leg went through #25 (Rounds 1–2 + six cross-model runs). Decide (D1–D5) and act (A1–A4) carry first-pass evidence (§10.9, verified §10.11) but have had no adversarial review. The combined exchange is being asked to test triad-level claims that sit on top of two components that have not survived the pressure the see leg survived.
2. D3b and D4b are not yet grounded / weak-cautionary. Carrying them inside a combined set risks lending them the set's borrowed credibility; the §1.6 residual must stay loud.

2.4 The central attack — does combining add value over Path A? (the make-or-break)

This is the question §1.2 says the round must answer, and it is where an adversary concentrates. Four moves:

1. Interdependence is a fact about the object, not a constraint on the method. Grant T2 in full — the legs co-determine durable coordination gain. That tells you the world is coupled. It does not follow that the review process must be combined. You can establish "the legs are interdependent" with three clean single-leg exchanges plus one short integration step — which is Path A plus a synthesis, the lowest-overhead option, not a single large combined exchange.
2. T2 was already produced by the cheaper path. The interdependence finding did not come out of a combined exchange — it came out of the see-leg exchange (#25's M1-weak) generalized by the riff itself (§10.12). So the existence of T2 is evidence that the headline triad-level claim is reachable without a combined exchange. That cuts against the necessity of this exchange, not for it.
3. The original "one object" warrant is gone. The load-bearing reason to treat the triad as a single object was T4 — "a single unifying object / protocol between the nodes." T4-strong is falsified (§10.8.5, §10.13). What remains (T2, T3) are claims you can state about the triad, but neither obviously requires a combined adversarial exchange to test: T2 was tested by a case-study design (§10.12), and T3 needs a matrix build-out plus per-cell research (§7) — neither is an adversarial-exchange task.
4. The combined exchange is outrunning its components (the §10.1 failure, inverted). §10.1 warned against concatenation masquerading as combination. The live risk here is the opposite: combination outrunning its components. Decide and act enter with un-adversarially-tested per-leg claims (§2.3.4). A combined exchange that presses triad-level claims resting on two untested legs is building the second story before the first is framed.

Verdict on the central question. Contested → leaning weak (~0.4, upper bound). The combined frame is real (T2/T3 are cross-leg by construction, and weak T4a/T4b are genuinely shared). But the case that a combined exchange adds value over Path A is not established at first pass: its one-object warrant (T4-strong) is falsified, its headline claim (T2) was generated by the cheaper path, and two of its three legs have not had the adversarial pass the see leg got. The honest alternative an adversary would recommend is a hybrid: keep the combined frame as the integrating object (the riff already is that), but give decide and act their own first-pass adversarial rounds before — or in parallel with — pressing the T-series, so the triad-level exchange is not resting on untested components.

2.5 Cross-cutting findings

1. Agent-self-coherence confounder. Same agent lineage wrote the frame, the claims, and this adversary. Option A helps; it does not remove the confound. Every number above is an upper bound; Round 3 (external human) is the gap-closer. This is the same caveat #24 and #25 carried, and it bites harder here because two legs have no external pressure at all.
2. The frame is drifting from interdependence back toward a sequence — with act as the new upstream leg. (The most important finding.) Three independent threads point the same way: T2's "act must bind" sharpening (§2.3.1, Attack 4); T3's binding-constraint column collapsing toward act/enforcement on the grounded rows (§2.3.2, Attack 1); and the §10.12 finding that every multi-leg failure was an act-leg failure. The original riff bet was see-first; #25 downgraded that to see is co-equal, stalls without the others; Round 2's evidence now leans toward act/enforcement is the binding constraint. If that holds, the project's honest object may not be "three co-equal interdependent legs" (position C) but "a sequence with act/binding as the gate" (a position-A-or-D variant the triad was built to move past). This is uncomfortable and it is the single most decision-relevant thing Round 2 surfaced.
3. Two of three legs enter un-adversarially-tested (§2.3.4). The combined exchange inherits that asymmetry; it should not be hidden by the set's overall polish.
4. Standing questions (Protocol §5). Missing perspectives: no commons-governance practitioner, social-choice theorist, or deliberation organizer has reviewed this — the Round 3 ask. Misuse potential: "coordination architecture / a protocol between the nodes" is exactly the kind of totalizing vocabulary a centralizer can launder — the Hayekian objection (designed coordination displacing spontaneous order) the lens would press, and which the T4a "interface, not a sovereign" claim already half-failed. Practitioner feasibility: none of the decide/act mechanisms (QV, sortition, dominant assurance contracts) is deployed at the scale the project's domains need; the triad's real-world purchase is still theory-and-pilot.

2.6 Epistemic-status table (per Protocol §3)

Confidence numbers are upper bounds (§2.1 caveat). The see-leg rows are not assessed here — they ride at #25 status.

2.7 The honest answer to the steward

• The combined frame survives Round 2 — it was not falsified — but it did not pass cleanly, and the case for a combined exchange over Path A is the part that wobbled most.
• T2 (the hinge) is now a working hypothesis, not "supported." The §10.12 test is real, but its single-leg cell is confounded by a deliberation-catalogue sampling frame, and its "act must bind" sharpening points toward an act-upstream sequence rather than co-equal interdependence.
• T3 (distinctness over #24) is contested. Its distinctness lives in the rows the riff itself calls thin; the grounded rows trend toward "everything bottlenecks at act/enforcement," which is the relabel failure.
• The most important finding is directional: three independent threads (T2's sharpening, T3's collapse, the §10.12 failure pattern) all point at act/enforcement as the candidate binding leg — a near-inversion of the original see-first bet, and a sequence claim the triad was built to supersede. The project should sit with the possibility that its honest object is "act/binding is the gate," not "three co-equal interdependent legs."
• Recommended next steps (steward holds the decision): (1) Round 3 external human review — a commons-governance / social-choice / deliberation reviewer, framed around the two live questions: is there a clean single-leg-alone durable success the §10.12 search missed? and does the matrix make any concrete prediction #24 + the single-leg literatures don't already give? (2) Before or alongside it, give decide and act their own first-pass adversarial passes, so the T-series stops resting on untested legs. (3) Do not promote anything — the #25 hold stands, reinforced: a frame whose interdependence hinge just slid to working-hypothesis and whose distinctness is contested is not doctrine-ready, and Round 2 actively surfaced a rival (act-upstream) reading. The steward holds whether to commission Round 3, run the decide/act adversarial passes first, hold here, or revise the T-series in light of the act-upstream finding.

Standing items

• Cross-references. Round 1 cites the riff §10 sub-sections (§10.1 design principle; §10.2 see import; §10.3–§10.4 decide/act; §10.5 T-series; §10.6 readiness bar; §10.8 / §10.12 / §10.13 evidence and rigor tests) by section; updates to the riff should preserve those anchors.
• See-import integrity. M1–M5 ride at #25 status. If Round 2/3 surface new evidence that materially moves an M-claim, re-grading is a steward decision (the #25 hold governs), not an automatic change.
• Open for Round 3 (carried out of Round 2). (a) the disconfirming-search question — is there a clean single-leg-alone durable success the §10.12 frame structurally could not see (secret ballot, central-bank independence, RTI/FOIA, statistical agencies)? This decides whether T2 recovers to supported or settles at contested. (b) the T3 prediction question — one concrete per-cell prediction the matrix makes that #24 + the single-leg literatures do not jointly already give. (c) the act-upstream question — is the project's honest object "three co-equal interdependent legs" (position C) or "a sequence gated by act/binding" (a position-A/D variant)? (d) whether decide and act need their own first-pass adversarial passes before the T-series is pressed further (the §2.4 hybrid). (e) the §1.6 residuals (D3b/D4b evidence; the thin/confounded T2 single-leg cell; the under-powered T4b control) — named, not yet re-run.
• Provenance / register. Idea layer only (riff §1.2 privacy firewall). The register is the project's usual one: working claims, named uncertainty, origin-balanced, no rhetorical flourish.

Closing posture (Round 2 end)

Round 1 restated the riff §10 combined claim set as falsifiable claims in two tiers (per-leg components + the triad-level T-series), stated the design principle up front, and opened with the riff's honest headline — the case for combining is real but bounded, resting on T2 and T3, not on strong T4b. Round 2 then applied first-pass adversarial review (Options A + C) with a coordination-skeptic lens, concentrating on the T-series and the combine-vs-Path-A question.

The combined frame survived Round 2 — it was not falsified — but it did not pass cleanly. Adversarial pressure moved T2 (the hinge) from supported to working hypothesis (the §10.12 single-leg cell is confounded by a deliberation-catalogue sampling frame, and the "act must bind" sharpening reads as an act-upstream sequence), and T3 (distinctness over #24) to contested (its distinctness lives in the thin rows; the grounded rows collapse toward act/enforcement). With T4-strong already falsified in the riff, the full weight of the combine-rationale now sits on two claims that Round 2 moved down — and the deepest finding is directional: three independent threads point at act/enforcement as the candidate binding leg, a near-inversion of the original see-first bet and a sequence claim the triad was built to supersede. The make-or-break question — does a combined exchange add value over Path A? — comes out contested → leaning weak: the one-object warrant (T4-strong) is gone, the headline claim (T2) was generated by the cheaper path (the riff), and two of three legs enter un-adversarially-tested.

As with #24 and #25, Round 2 is first-pass — its confidence numbers are upper bounds (the same agent lineage wrote the claims and the adversary), and Round 3 (external human review) is what closes the gap. The exchange promotes nothing: the #25 hold stands, reinforced. The steward holds whether to (i) commission Round 3 (a commons-governance / social-choice / deliberation reviewer, framed around the disconfirming-search and T3-prediction questions), (ii) run decide and act's own first-pass adversarial passes before pressing the T-series further (the §2.4 hybrid), (iii) hold here, or (iv) revise the T-series in light of the act-upstream finding before proceeding. Either way, Round 2 did what a first adversarial pass should: it found the combined object's weakest joints and surfaced a rival reading the project now has to answer.