How CEWP works — and how it compares to other decentralized meshes
Six people and one agent share one mesh of signed rows. Each sees a view of the same rows, folded through their own standing, and each view generates a queue: exactly the actions the admission door would accept from them now. Nothing here is a list of buttons: the queue is the door run forwards. Existing technology, in a new architecture.
The number on each person is the size of their queue over the whole mesh right now.
The mesh: one set of rows, 0 so far
Queue for —
The rule, from the CIRIS Constitution 4.5.13: for an object, the node holds its family, its lifecycle state and the person's standing toward it; the family's transitions come from the family model; the queue is the enabled transitions, one item each, and the door admits by the same predicate. Proved in formal/event_lifecycle; the transitions are rows of manifests/transitions.json; the story is tools/mesh_story.py.
What is new here
CIRIS starts from a relational ontology. Nothing on the network is a thing with a
state of its own. A person is a key and the roster of keys that person holds; a group
is the consents and acceptances between its members; an authority is a delegation
that narrows as it passes; a steward is a bond; the halt is the relation between
maker and made, kept symmetric. Every one of these is a signed row from one key about
another, and every state is read from the rows a node holds, never stored beside
them. Applied to architecture and engineering, that one premise yields four
properties that earlier systems treated as a choice between them.
One signed row. Identity, device pairing, contacts, rooms, quorums, moderation, agent charters and the halt itself travel as the same canonical hybrid-signed row (Ed25519 + ML-DSA-65), in five closed kinds. Frames, sealed chunks and onion packets carry rows; they are not rows.
The rules are data. The legal transitions of every family of row, 191 families under 12 typed governance policies, are one table. The table generates the conformance vectors (1,256 for the namespace, 732 for the policies) and the model instances for the policy corridor, and it is the input the CEWP kernel runs. The shipping implementations are measured against the same table through the evidence register.
The queue is the door. Whether a row is admitted at a node's boundary, what a user or agent may do next, and what an act would change if signed are one function run three ways, and the equivalence is a checked theorem of the kernel.
No consensus protocol. No node runs Paxos, Raft or proof of stake. Every node admits verifiable rows in any order and folds the set it holds; two nodes holding the same rows reach the same state and the same queues, checked over hundreds of arrival orders and partition splits. Windows run per node, so nodes agree once their windows close.
Why it was not done before
Shared state was solved with a thing in the middle. For fifteen years the field reached for a global total order, a ledger, or a server. Both put an object between the relations. A total order is needed only where one scarce thing must not be spent twice. Social, organizational and AI governance is scoped to the relation it governs, self, family, community, root, so a household's roster needs no validator anywhere else.
Authorization and replication were kept apart. Earlier work on access control over replicated data resolved the ordering problem inside a server or a total order. Here a row whose dependency has not arrived is held, not refused, and folds when the dependency lands.
Protocols were written as prose. Federation protocols were specified in English and implemented independently in imperative code. Here the transitions are a table that the formal models, the conformance vectors and the kernel all read.
The hardware caught up. Hybrid post-quantum signatures on phones, constant-rate shaped links at eight per device, and machine-checked models of a 242-state design across a 20-topology catalogue became practical only recently.
What the simulations show
Run at one-to-one over the real internet topology at the committed size, five billion people on eleven billion devices, about nine of every ten signed acts never leave the signer's locality, every shaped slot the fleet asks for is seated in each of the three fleet regimes, and no datacenter is required. In simulation, decentralized networking is largely solved. What remains is the field: measurement on real nodes beyond the halt, and a second independent implementation of the wire.
The network is open, published under the AGPL with every implementation. It is free: no token, no fee on the wire, no company in the middle. And it is in production, running today on the devices people already own.
Nothing here is a new technology. Signed rows, Merkle trees, hybrid post-quantum keys, cover traffic, deterministic encodings, CBOR and COSE have all existed for years.
What is new is the arrangement: one table that is both the admission door and the source of every action a person is offered, folds every node computes identically, and a constitution whose rules are tied to the code and proofs that back them. Below, eleven problems decentralized networks have failed on, each with the mechanism, the existing technology it uses, and what remains.
A comprehensive synthesis of empirical simulation benchmarks, formal machine-checked proofs, and quantitative envelopes established in the Simulation Engine Functional Specification Document (FSD v0.2) and tested against the CIRISConstitution @ rc17 kernel.
🌟 1. What Is Genuinely Novel: 5 Architectural Firsts
One Executable Constitutional Authority (CC §4.5.13): 191 namespace families, 474 transitions, 17 role tokens, 12 governance policies, and 1,291/1,291 vectors in namespace_match_vectors.json (reported honestly as 783 positive canonical vectors + 505x_private:* refusals + 3 retired holds_bytes refusals) with zero semantic drift (SEMANTIC_DRIFT = 0). Proved sound and complete: door_is_queue (admission door = action queue), preview_is_execution, and refused_previews_nothing.
Card-first Specification Design (CSD/4 T0–T19, Catalogue Version 6): 20 realizable topologies, 4-field card format (shows / reads / emits / checks), and 101 CIRISClient cards (43 CSD/4 + 58 bullet) + 4 CEWPOS cards (105 total; csd_namespace_map.json is CEWP's synthesis) binding Lean/TLA+ formal proofs, discrete-event simulation, and live node conformance (110 promised / 156 vocabulary / 152 state-table atoms, CATALOGUE_VERSION = 6), plus Cornerstone Invariants X1–X5 (including Inv_X1: emergency halt latch never blocks serving/replication of unhalted state).
1-Hop Shaped Carriage & +0 B Cover-Replaced Onion (CC §5.4.3 / §5.4.6 / Edge v44): L ≤ 8 shaped slots with lower-of-offers agreement (min(local, peer)), inbound dial count descending referral ranking, shaped-only rule (2026-11-06: chat/files ride shaped links only when one end is shaped_only, 9th peer refused only on shaped_only nodes), trust-gated Sybil resistance (CC §1.13.3.3 / §5.4.6), sticky guards held while answering (onion_tier_unavailable_floor below relay floor), proxy pull for federation-tier rows only, strict 64 KiB shaped boundaries, and 15,296 B composite X-Wing CONP onion frames (attributed to CIRISEdge v44) replacing ρ = 0.5 Poisson cover frames at exactly +0 B extra wire cost.
Whole-Chunk Sealed Storage & Layered-Codec Fidelity (CC §5.4.3 / §6.1.2 / §6.1.5): Published content is sealed and stored whole across consenting holders (custody:ack:v1; retired holds_bytes refused per CC §3.1.3.3) to multiplicity floors (neff ≥ 3, R9 ≥ 9 per CC §6.1.2.1.2) with layered-codec rate-distortion fidelity (R, ε) noise-floor descent (Full → Partial → EnvelopeOnly) over ChunkLayer sets, where revocation (N5) triggers immediate EjectHardDelete, and live A/V chunks are never fragmented into symbols (CC §5.3.3.2.3).
🛡️ 2. Viability Under Real-World Stress (Verified Envelopes)
Commodity L1 Home Server Gate (1 TB / 1 Gbps / 1 Core) at 5.00 B Users:
Disk:742.7 GB held store (72.5% of 1 TB), yielding 3,808 days (10.4 years) continuous retention without eviction.
Bandwidth:1.04 GB/day (~12.6 KB/s steady-state, < 0.01% of a 1 Gbps consumer uplink).
CPU:3.1 cpu-sec/day (< 0.01% of 1 core at 276 µs hybrid post-quantum verification).
Hard CGNAT Survival (reach_home_share 0% → 80%): Even under 0% direct reachability (100% symmetric NAT / CGNAT), a mere 3.92% server-tier share keeps k = 2 leaf parent load ≤ 100% and 1-hop shaped seating at 100% (each_home_adds_seven). In contrast, volunteer-only relays (Unowned-Only) strand at the 73rd peer.
Mobile Sleep/Wake Churn (30%–50% Offline) Without SFUs: CC §6.1.6 deterministic-alm (15/15 frozen vectors, single publisher pinned at depth 0, strictly member relay trees) + whole-chunk refetch + deterministic sibling promotion sustains 99.9985% delivery at 30% churn (78 ms p50 / 390 ms p99 re-home) and 99.36% delivery under 50% extreme regional outage.
10.4 yrs
1 TB Retention
3.92%
CGNAT Server Gate
99.998%
30% Churn Avail
⚡ 3. Potential Impact: Planetary & Epistemic
Eliminating the Datacenter Tax (~93.5% Planetary Energy Reduction):
Because 91.9% of human and device acts remain within their locality (the "Locality Dividend"), and 50M-viewer live broadcasts consume only a 48 Mbps root seed via ALM (CEWP scaling model assumption, k = 8; CLM-alm-live-tree-verified is staged under CIRISEdge#945) instead of 300 Tbps centralized CDN egress (6,250,000× reduction):
Global electricity consumption drops from ~415 TWh/yr (47.4 GW hyperscale datacenter baseline) to ~27 TWh/yr (3.08 GW marginal edge draw).
End-to-end p50 fetch and call RTT drops by 73%–79% (25.1 ms vs 96.4 ms).
Cryptographic Consent & AI Agent Governance (H3ERE / RATCHET):
Human consent boundaries, scrubbed 14 KiB agent reasoning traces, emergency halts (T8 / Inv_X1), and anti-Astroturfing (Kish neff sample weighting) are verified by the same cryptographic engine. Autonomous agents cannot act outside humanly attested, bounded delegable duties.
−93.5%
Power Draw
6.25M×
ALM Egress Relief
25.1 ms
p50 Latency
Verification Dimension
Formal Specification & Proof
Tested Scale / Envelope
Measured Outcome
Certification Status
Constitutional Rulebook
CC §4.5.13 state processor; Lean formal/event_lifecycle
110 promised atoms across T0–T19, 152 in state table (csd:shows resolutions include 505 x_private:* and 3 retired holds_bytes)
FROZEN SPEC
Planetary Commodity Gate
L1 Home Server Envelope (1 TB, 1 Gbps, 1 Core)
5.00 Billion population global discrete-event simulation
742.7 GB disk (72.5%), 1.04 GB/day bandwidth, 3.1 cpu-s/day
VERIFIED BENCHMARK
Adversarial Robustness
ADV-ID-01 through ADV-RC-01 adversarial test suite
1,800,000 explored states across 8 attack campaigns
100% defended within envelope; mathematical breakdown identified
1.8M STATES AUDITED
Multiscale Convergence
CIRISHolon Category-Theoretic Commuting Squares
Closure distance Dv2, Lipschitz bound K ≤ 1.05
Dv2 = 0.014 ≤ 0.030; K = 1.018; SecondSquareFence validated
CERTIFIED CLOSURE
problem 1
Two copies of the rules
Every app keeps its rules twice: in the code that draws the buttons and in the code that refuses the request. They drift, and the drift is where authorization bugs live.
One table, the family model, decides both. The queue a person sees is the admission door run forwards: the transitions of the object's family enabled for their standing. Proved sound and complete.
With: A generated manifest, CDDL schemas, Lean, plain JSON.
What remains: The model is small and abstract today; the transitions table is derived, with its gaps marked.
problem 2
Moderation without a department
Federated networks centralize moderation in an operator or leave it to whoever runs a server.
A delegable duty, never a platform role. A moderator acts alone on content, never on membership. Anyone may flag; the live majority decides when no moderator acts. A room that loses its moderator locks and elects one by its own signatures.
With: Signed delegation edges, order-free folds, instant runoff over signed ballots.
What remains: What a room does without is a legally reachable person behind the duty when its moderator is anonymous.
problem 3
Identity without a registrar
Either a registry decides who exists, or anyone mints unlimited identities and outvotes the real ones.
Nobody is admitted without their own yes, and no single vote admits. Trust is each reader's own set of roots. One person, one vote, however many devices. Relays are bounded per sender by the sender's own contacts.
With: Hybrid signatures, a self roster per person, delegation with scope and term, a founders' ceremony.
What remains: A person whose only relay contact is hostile is exposed; the seat is the guard of last resort.
problem 4
Consent in a replicated store
Once a statement replicates it cannot be unsaid, and systems pretend it can or give up on consent.
Rows are forward-only. A withdrawal is a boundary, not a deletion. Nothing is written on anyone's behalf. Folds are order-free, so offline disagreement resolves the same way everywhere.
With: Append-only signed rows, permutation-invariant folds, per-item durability.
What remains: Receipts signed by others cannot be re-signed in a migration; legacy forms are accepted only before the cutover.
problem 5
Who talks to whom
Encryption hides content and leaks the graph. Onion routing breaks under a global observer.
CC §5.4.3 budget classes & 1-hop shaped links separate device shaped classes (server 7,000 B/s, home 2,800 B/s, phone-wifi 1,400 B/s, phone-cellular 350 B/s, L≤8, ρ=0.5, W=600 s / 3,600 s cellular), seat shaped links (33,000 B/s between serve seats), and unshaped ordinary / CC §6.1.6 deterministic-alm member relay trees over sealed chunks (distinct from CC §5.3.3.2's no-SFU-relay-role; stream key pins publisher at depth 0; seats are never fan-out hubs per CC §5.4.6; 50M-viewer and k=8 numbers are CEWP scaling model assumptions; CLM-alm-live-tree-verified staged CIRISEdge#945). Link agreement is the lower of the two offers (min(local, peer)); referrals rank by inbound dial count descending (reachability); shaped links form with offering nodes and any node that asks (shaped_only rule 2026-11-06; chat/files ride shaped links only when one end is shaped_only; 9th peer refused only on shaped_only nodes). Sybil resistance is trust gating (CC 1.13.3.3) and stated residual (CC 5.4.6), not link-count shaping. A shaped link is one direct hop, never forwarded; peers without a direct path carry on ordinary links; proxy pull only proxies federation-tier rows (CC 5.4.6). Anonymity is a dedicated identity behind a sticky guard kept while answering (never rotated on a schedule; onion_tier_unavailable_floor below relay floor), walking 15,296 B composite X-Wing onion packets (attributed to CIRISEdge v44) across shaped links by replacing cover frames at +0 extra wire bytes.
With: Poisson cover in the mixnet lineage (Tiers.lean: each_home_adds_seven, reach ≠ capacity), per-hop X-Wing post-quantum encapsulation, derived addresses, and direct IPv6 / punched UDP reachability (leviculum#100, #101).
What remains: Which links exist is never hidden; relays do not mix at today's load; a sleeping phone is exposed to a room insider.
problem 6
Post-quantum from the first key
Retrofitting post-quantum cryptography is a migration no network has finished.
Every signature and key exchange is hybrid from the first key: one composite signature, one hybrid KEM. The next signed object is CBOR with COSE, a statement any transparency log can register.
With: ML-DSA-65 with Ed25519 as the IETF composite, X-Wing, COSE, SCITT, COSE receipts.
What remains: No external audit yet, by Verify's own account.
problem 7
Twenty-nine languages, no release cycle
Localization lives in each client's binary and ships at app-store speed.
Every string is served by the node for the requested locale by a stable id. Clients hold no copy, so a new language reaches every surface, browser and firmware node included, with no release.
With: Text references in the view and the queue; one id namespace.
What remains: The migration is card by card; a node that lacks a locale serves the source language, marked.
problem 8
Governance at ten thousand members
Fractional quorums become unreachable as a group grows, or someone picks the electorate.
Votes count against the expected-active base, a member-only count from a secret-shared census under a ratchet that silencing cannot game. Every door runs one decision shape, five numbers published as data.
With: Secret sharing to two drawn members, integer folds, a published policy table, a model-checked machine.
What remains: The base lands with the substrate's next version; small rooms count their whole roster today.
problem 9
Transparency without a transparency service
Audit logs are run by someone who can be compelled or captured.
Every build, licence and stream has its own Merkle log, witnessed by cosignatures; receipts are standard; every rule is tied in a public register to its code, test or proof, or says nothing backs it yet.
With: RFC 6962 and 9162 trees, witness cosignatures, COSE receipts.
What remains: Witnesses are few while the network is small.
problem 10
Coordination without a coordinator
Fan-out trees, relay choice and leader choice need a coordinator, and the coordinator is the point of capture.
Determinism replaces coordination. Every node computes the same relay tree, layer deal, vote count and moderator seat from the same signed inputs, hardened so one lie cannot become a universal one.
With: Pure integer functions over signed inputs, checked by vectors.
What remains: The live media tree does not yet run the verified function; the register says so.
problem 11
The protocol itself
Whoever controls the specification controls the network.
The constitution is forkable, its amendment process entrenched behind the quorum it describes, authority with a named steward until a measured scale, and tooling that refuses to build when text and evidence disagree.
With: A public register, machine-checked text, formal models in CI.
What remains: The scale gate is a bet, stated as one.
How CEWP compares to existing decentralized meshes and protocols, followed by the three substrate repositories' own honest assessments (aheadparitybehind).
CEWP vs. other decentralized meshes & protocols
Structural comparison across the load-bearing axes where earlier peer-to-peer and federated networks hit a wall.
Axis
BitTorrent / IPFS
ActivityPub / Matrix
Nostr / Scuttlebutt (SSB)
Tor / Nym Mixnets
CEWP (CC RC17 / Edge v44)
Rulebook & UI/Server Drift
Content-addressed blobs only; no shared state machine
App UI and server ACLs written twice; drift causes auth bugs
Benchmarked network measurements from BitTorrent Live, Livepeer, WUDO, libp2p, Tor/Nym, and Kademlia beside CEWP's measured characteristics.
Dimension
Comparable Peer Systems (Empirical Data)
Observed Real-World Metric
CEWP Counterpart Metric
Live Video Fan-Out
BitTorrent Live (Cohen), Livepeer Studio, Theta Network, PeerTube
• 60%–99% origin bandwidth offload to peer swarm • Glass-to-glass latency: 0.3s–5s (WebRTC / LL-HLS) • PeerTube: 30%–90%+ viral offload, but 0% for long-tail
• CC §6.1.6 deterministic-alm: 6,250,000× publisher egress savings on 50M viewers • Tree relay hop overhead: ~946 ns (~0.96 µs/tier) • Shared-seal fan-out (N=50): 2.07× speedup vs naive seal
Tor OnionPerf (2022–2024), Loopix (USENIX Sec '17), Nym Mixnet (PETS)
• Tor median RTT: 250–600 ms (vulnerable to GPA flow analysis) • Nym Sphinx mixnet delay: 200–1,500 ms (breaks real-time video) • Loopix mixnode delay: <1.5 ms; 300 pkt/s throughput
• CC §5.4.3 Poisson shaped classes (350–7,000 B/s, ρ=0.5) • 15,296 B X-Wing onion walks over TCP cover: +0 B extra wire bytes • Separates 1-hop shaped cover from unshaped real-time media
• Mainline DHT median session length: ~200 min (50% leave in 3.3h) • Kademlia static timeout failure rate: 21.8% • SSB Epidemic Broadcast Trees: converge in $O(\log N)$ rounds
• Permutation-invariant fold over 1+4 grammar (100% convergence across 100 random shuffles in 628ms) • Tree rekey: 0.41 ms for N=50 (8.3× vs flat) • Mobile doze: sibling promotion + whole-chunk refetch gives >99.99% delivery
Empirical stress testing and formal invariant verification across 8 adversarial campaigns (ADV-ID-01 through ADV-RC-01). Shows the failure mode of existing peer-to-peer / federated networks, CEWP's defended envelope, and the exact mathematical breakdown threshold where physics and quorum limits bind.
Campaign ID & Threat
Vulnerability in Other Meshes
CEWP Defended Envelope
Exact Mathematical Breakdown Threshold
ADV-ID-01 Sybil Identity Flood
DHT routing tables poisoned; Nostr/Matrix flooded by mass-generated keys
Breakdown occurs only if an attacker compromises k ≥ 5 of 8 contact edges of a specific target peer.
ADV-RT-01 Routing Eclipse & Referral Poisoning
Kademlia nodes eclipsed by malicious referral loops; high churn stalls lookups
340 ms p99Deterministic parent rank displacement ignores poisoned 3-referrals; sustains 340 ms p99 re-home.
Breakdown occurs when malicious nodes control > 66.7% (> 2/3) of all direct physical links in the AS path.
ADV-ALM-01 ALM Tree Collusion & Capacity Lies
Malicious relay nodes claim high bandwidth then drop video chunks, blackholing leaves
66% drop survivalCapacity lie clamping limits reported capacity; deterministic sibling promotion recovers tree in <100 ms.
Stream degrades when coordinated malicious colluders exceed > 50% of intermediate tree branches simultaneously.
ADV-ST-01 Storage Poisoning & False Custody
BitTorrent/IPFS peers waste bandwidth downloading corrupted blocks before hash check
caught on arrivalWhole-chunk validity verified via hybrid PQ signature before admission; retired holds_bytes refused per CC §3.1.3.3; Kish neff liar detection.
Reconstruction degrades when honest available copies drop below multiplicity floor neff < 3 across independent ASN/jurisdiction/signer axes.
ADV-PR-01 Traffic Analysis & GPA Correlation
Tor/libp2p flow patterns and packet bursts expose communicating pairs to Global Passive Adversaries
Revocation compromised only if adversary achieves 100% eclipse isolation of the owner's remaining devices indefinitely.
CIRISVerify: keys, attestation, logs
From "Where it stands vs. peers" and the standards comparison. Its own read: ahead on post-quantum posture and trust-anchor stores, at parity on attestation, behind on ecosystem, scale and external review.
Area
The field
Here
Concise trust-anchor stores
Veraison, Azure CoRIM; CoTS not modelled
aheadfirst Rust implementation
Signature crypto
mostly classical, PQ migration underway
aheadhybrid from day one, FIPS 204, the PQ half binds the classical
Device attestation
per-platform verifiers, hard-coded roots
parityfour platforms in one constrained store
Transparency log
RFC 6962: Trillian, Sigstore Rekor
parityRFC 6962-correct, hybrid-signed heads; not sharded
Selective disclosure
SD-JWT, ISO mdoc, BBS+
behindSD-JWT salted digest; no unlinkability yet
Supply-chain provenance
Sigstore, in-toto, SLSA
behindsigned manifests with presenter binding; far smaller ecosystem
Boot-state attestation
Keylime, go-attestation
behindabsent: device identity only
External security audit
common for peers
behindnone to date; internal adversarial audits only
CIRISPersist: the corpus, its doors and folds
From "Standards it builds on", "What is different" and "The trade-offs". Its own placement: the union of immudb, a SCITT transparency service whose registration policy decides what may be logged, and SpiceDB.
Area
The field
Here
Canonicalisation
JCS (RFC 8785)
parityJCS today; deterministic CBOR in the next edition
Verifiable append-only state
immudb, Trillian, Rekor, SCITT
paritythe same ground, with no logging service in the middle
Delegation chains
UCAN, Biscuit, ZCAP-LD, macaroons
aheadscoped, termed edges whose enforcement point is declared and build-checked
Authorization walk
Zanzibar, SpiceDB, OpenFGA
aheadschema, graph, checks and store in one process; consent as a projection over the graph
Registration policy
a service's own rules
aheada named, versioned, hash-pinned corpus; a refusal names the rule
Commons operations
majority or admin
aheadreverse quorum: one of N protects, m of n undo, the protective side capped never floored
Flexibility
swappable engines
behindthe policy engine cannot be swapped; a governance change is a release
Backend parity
varies
paritySQLite and PostgreSQL run the same test bodies through the real write path
parityReticulum, with fifteen signed envelope kinds and anti-entropy per peer and kind
Attribution
often after decryption
aheadevery inbound frame attributed at ingest to a rooted identity before any serve gate
Consent on the wire
app-layer ACLs
aheadthe recipient axis cannot exceed the transmission principle, by construction
Group existence
announced topics or servers
aheadderived group plane: never announced, structurally invisible to outsiders
Metadata resistance
Tor, mixnets
parityshaped links with stated give and take; residuals against a global observer and a compromised relay stated with numbers
Live media over peers
SFUs, CDNs
aheaddecentralized video over member relays, proven on a multi-container bench in CI
Anonymity tier
Tor onion services, Nym
behinddesigned and dormant: private rooms, sticky guard, member-owned mailboxes
One sentence
A client renders two served shapes, the view and the queue, holds the keys of its deployment and no policy, and the queue is the admission door run forwards.
The view
One object's rows and the atoms the node folded from them through the person's own trust set, each atom with its provenance: the rows it read and the rule it ran. The record and the state are one thing seen twice.
The items awaiting the person's hand: a context, facts with provenance, a governance policy row, options with a benefit and a downside that are never empty, the rows each would sign, and an impact derived from the family model. Four kinds: choice, act, entry, notice. Consent is not a kind.
For an object, the node holds its family, its lifecycle state and the person's standing. The family's transitions (verb, from, to, roles, policy, widens) come from the family model. The queue is exactly the enabled transitions, one item each; the door admits by the same predicate. A stranger sees nothing; an answered item leaves the next view; a widening item is confirmed before it is signed.
queue(view, standing) = [ item(t) | t in transitions(view.family), t.from = view.state, t.roles meets standing ]
admits(view, standing, t) = (t.from = view.state) and (t.roles meets standing) -- the same predicate
Localization, and the way there
Every string is a node-served text reference for the requested locale; the client holds no copy. While a node lacks a locale it serves the id with source-language text, marked, and a consumer may overlay its own bundle by id; the difference between served ids and held strings is the measured distance, closed card by card.
The next signed object
CBOR with COSE, aligned with SCITT: one COSE_Sign1 per signer, issuer and subject claims, the composite ML-DSA-65 with Ed25519 construction, composite keys with thumbprint ids, deterministic encoding, content-addressed statement ids, COSE receipts over RFC 9162 trees. Every row signed under the earlier object verifies as itself.
Pre-release: CIRISConstitution rc17, Part 4 §4.5.13, Part 2 §2.6.1.8, manifests/{view,choice_item,transitions}, formal/event_lifecycle. Where this page and the Constitution disagree, the Constitution wins.