FxDAG

Update log

• 7 September 2026 - First edition. Drawn from the DAG and finality specifications, the runtime-engine documentation, the architecture programme's chapter on causal topology, the Yellow Paper's consensus semantics, and the papers' account of isochronic settlement. Edge-class and object names are given as the shape of the design; parameters are not.

Short answer

FxDAG is the fifth organ and the one that classic chains sell. Its promise is to order causality without pretending conflicts disappeared. Where a conventional chain treats ordering as a race for inclusion - leaders, priority fees, builders, private order flow - FxDAG treats it as a constitutional surface: which cells can commute, which must wait, and which conflicts must be made explicit instead of being buried as timing accidents.

Today the organ is precisely what its specification says and nothing more: a local, deterministic ordering. It orders the cells it is given by their explicit dependency constraints, breaks ties with an ordered ready set keyed by the canonical bytes of cell identities, ascending, and refuses empty input, duplicate identities, missing dependencies in strict mode and dependency cycles. It does not implement consensus, voting, Byzantine fault tolerance, finality, networking or block production. The finality specification is, in its own words, a placeholder. The local block assembler preserves FxDAG's order exactly and sorts nothing.

The direction has two layers. The architecture programme takes FxDAG from dependency ordering to a causal transition graph with typed edges, effect-aware conflict detection, bounded traversal, and a strict separation between causal necessity - a safety constraint - and policy preference, which is not. The Yellow Paper places that graph inside CapsuleBFT, the native consensus semantic: cells propagate in a causal DAG, independent cells commute, conflicting cells form conflict sets, and epoch capsules crystallise finality. Above both sits the organ inversion the papers call isochronic settlement: windows in which order does not exist, so that front-running and sandwich games lose their substrate rather than being policed.

What is established, what is in development, what is not claimed

Where FxDAG sits

   [1] FxWire -> [2] FxIntent -> [3] FxCore -> [4] FxCell -> [5] FxDAG -> [6] FxVM
                                                                  ^        -> [7] FxState
                                                                  |        -> [8] FxBlock
   after FxCell (units with declarations), before FxVM (execution in the order decided)

Figure 1 - Step five of eight. Everything before it produces units; everything after it consumes an order.

The native semantic: an order is evidence

Three of the architecture programme's laws describe what FxDAG must be before any mechanism is chosen.

Consensus sees deterministic committed inputs only. The final commit order must be reproducible from committed inputs and the protocol version. Nothing a validator observes outside the committed bytes may make it disagree with another validator about the same cells.

Determinism has an explicit firewall. The Yellow Paper's obligations for the organ are absolute: eligibility, the conflict key, deterministic rank and equivocation detection are pure functions of the DAG view - no wall clock, no oracle, no artificial-intelligence input. Time evidence, when it exists, is a claim to be judged, never an authority.

Causal necessity is not policy preference. A causal dependency is a safety constraint: a cell that requires another cannot run before it. A fee preference or a latency policy is not a constraint; it may influence which window a cell enters, never whether a dependency holds. The programme names the separation as a law precisely because classic chains blur it.

   what an order must preserve            what an order must not sell
   causal dependency  (safety)            position inside a window
   declared conflicts (surfaced)          priority over a peer's claim
   commit order reproducible from         a leader's private view
   committed inputs + protocol version    an opaque local score

Figure 2 - Two columns the organ keeps apart. The left is structure; the right is the market classic chains run on, and the direction removes it.

What exists today

Ordering, not consensus. FxDAG orders the cells it is given by their explicit dependency constraints. When several cells are ready, ties are broken by an ordered ready set keyed by the canonical bytes of their identities, ascending - a rule with no clock, no randomness and no opinion. The result is the same on every machine that receives the same cells.

A grammar of refusals. The ordering refuses empty input, duplicate cell identities, missing dependencies when configured as strict, and dependency cycles. Each refusal is deterministic; none is an exception.

Exact preservation downstream. The local block assembler consumes FxDAG's result and preserves its order exactly. It does not sort cell identifiers and does not implement consensus. The finality specification is a placeholder for a future finality rule, validator model and finality receipt; it defines none of them and says so.

   cells + explicit dependencies
        |
        v   refuse: empty . duplicate identity . missing dependency (strict) . cycle
   dependency constraints  ->  ready set ordered by identity bytes ascending
        |
        v
   deterministic order  --->  local block assembly (order preserved exactly)
        |
        x   no consensus . no voting . no BFT . no finality . no network

Figure 3 - The local ordering as implemented. The tie-break is the whole of its opinion, and it is a byte comparison.

The runtime-engine documentation also records a docs-only plan for time evidence and fairness concepts under a reserved plane, and states that it adds no ordering, scheduler, anti-extraction, consensus or finality behaviour.

The direction

The causal transition graph

The architecture programme evolves FxDAG from local dependency ordering toward a causal transition graph. Its candidate edge classes name relationships, not fields: a cell may require, observe, consume, conflict with, supersede or derive from another. Not every class must become consensus-visible in the first network; the atlas reserves the semantics while implementation proves which are load-bearing.

Conflict detection is effect-aware: two reads of the same immutable resource may be compatible, while write-with-write or read-with-write may conflict depending on the object's semantics. The conflict grammar must be deterministic and bounded, and no graph traversal is allowed to scale without an explicit ceiling. The long-horizon advantage the programme claims is that the protocol can reason about a causal transition graph rather than a flat list: deterministic parallelism, explicit conflict evidence, object-aware scheduling and a clearer explanation of settlement, without each application reinventing its own dependency semantics.

   cell A  --requires-->  cell B         A cannot run before B
   cell C  --conflicts--  cell D         same declared write: one conflict set
   cell E  (no shared declaration)       commutes with all of the above
   cell F  --supersedes-->  cell G       G is withdrawn, never rewritten

   every edge derived from declarations; every traversal bounded

Figure 4 - Typed edges of the target graph. The order is computed from what cells declared, and the graph can explain why any two cells were parallel, waited or contended.

CapsuleBFT

The Yellow Paper names the native consensus semantic and freezes its priorities: safety over liveness over latency. Cells propagate in a causal DAG. Independent cells commute; conflicting cells form conflict sets. Epoch capsules crystallise finality: they commit the four roots - state, receipts, data availability and denial - plus coherence roots, and carry a quorum certificate. After a capsule certificate, history is irreversible: a correction is a new claim, never a rewrite.

Before any live network exists, adversarial schedules - partitions, equivocation, reordering, delay, crash and recovery - are to exist as frozen byte transcripts, and any implementation must produce identical capsule roots across all of them. The staged path is a simulated deterministic network, then permissioned sockets, then Byzantine chaos, and no stage may claim the next stage's readiness. The paper's provider note is the demotion principle applied to consensus: certified DAG-BFT research may implement CapsuleBFT; it does not define it.

Isochronic settlement

The organ inversion removes the value of intra-window order instead of policing it. Within a capsule window, execution semantics are order-invariant by construction: independent cells commute, and contested cells clear uniformly - one clearing outcome per conflict set per window. Position inside a window is worthless, so front-running, sandwiching and priority games lose their substrate. The block, mempool and leader trilogy is replaced by a window that opens, claims that are sealed, a window that closes and a deterministic, commutative settlement. Users may price window compression - urgency buys shorter windows - but nobody can buy position, because position does not exist.

   window opens -> claims sealed -> window closes -> commutative settlement

   for every permutation p of the cells of window W:
        settle(W) == settle(p(W))        a machine-checked invariant

Figure 5 - The window replaces the block as the unit of ordering. The invariant is stated as a conformance family to be tested exactly like the line's existing vectors.

The Yellow Paper adds the two lanes that keep the invariant honest: legitimately dynamic access patterns use typed, bounded namespace claims priced by state congestion; claims that cannot be bounded at all route to a serialised, single-writer lane and pay for their opacity instead of poisoning parallelism.

The refusal that defines it

The White Paper's grammar of organs replaces the mempool with sealed claim windows and refuses pre-power leakage and order-flow sale. FxDAG's own refusals follow.

• Silent reordering. An order that cannot be reproduced from committed inputs and the protocol version is not an order; it is an opinion, and the organ refuses to hold one.
• Extraction-prone ambiguity. Ties are broken by bytes, windows are order-invariant, and an opaque score can never change canonical ordering. There is nothing to extract from a position that does not exist.
• Conflicts buried as timing accidents. A conflict is an explicit object - a conflict set with a deterministic rank and one uniform clearing - not a race someone happened to win.

What FxDAG does not do

• It does not implement consensus, voting, Byzantine fault tolerance, finality, networking or block production on the local line. The finality specification is a placeholder.
• It does not discover dependencies. It orders the dependencies it is given; on the local line cells carry an intentionally empty list.
• It does not use a clock, an oracle or any external input to order. Time evidence belongs to a reserved plane and, in the direction, is a claim to be judged.
• It does not let policy - fees, latency, scoring - masquerade as a causal constraint.
• CapsuleBFT and isochronic settlement are design. No stage of the staged path has been claimed.

Verdict

FxDAG is the organ where the direction is most radical and the implementation most modest, and the dossier says both. What exists is a deterministic local ordering with a byte-level tie-break and a precise grammar of refusals - enough to prove that an order can be reproduced by anyone from the same inputs. What is designed is a causal transition graph inside a consensus semantic that seals windows rather than blocks, in which independent cells commute, conflicts are explicit objects, and the position that classic chains sell does not exist. The distance between the two is the distance to a network, and no stage of it is claimed.

Frequently asked questions

Is FxDAG a consensus protocol?

Not today; it is a local ordering. In the direction it is the causal graph inside CapsuleBFT, the native consensus semantic, which existing DAG-BFT research may implement but does not define.

How can ordering be "worthless" and still deterministic?

Because within a window the outcome does not depend on the order: independent cells commute and contested cells clear uniformly, one outcome per conflict set. Determinism is about reproducing the same settlement from the same committed inputs, not about who came first.

What is a conflict set?

Cells whose declared writes intersect. They are ranked deterministically and cleared uniformly within a window. The set is an explicit object with evidence, not a race resolved by timing.

Where does time come in?

Only as evidence, and only in the direction. A reserved plane, FxChronos, is designed to produce time evidence for ingress fairness and ordering audits; the consensus obligations forbid any wall clock inside the order itself.

Why byte-level tie-breaking?

Because it is the only rule with no opinion. Two machines that receive the same ready set compare the same bytes and reach the same order, with no clock, no randomness and no preference to sell.

Sources and methodology

Drawn from the DAG, finality and block-assembly specifications in the protocol repository, the runtime-engine documentation, the architecture programme's chapter on cells, the DAG and causal topology and its design laws, the Yellow Paper's consensus semantics and execution chapters, and the papers' account of isochronic settlement. Implementation claims are limited to what the specifications describe as implemented; design claims are attributed and carry the status research, planned or reserved.

The specifications and papers are not published on this portal and are not reproduced here; edge-class and object names are reported as the shape of the design, parameters are not. Figures shown anywhere on this portal come from a single dated snapshot; none are introduced by this dossier.

Related reading: FxCell, the organ before; FxVM, the organ after; Three planes and one small kernel for the laws this organ enforces.