Founding Edition · Issue 001
THE AGE OF AGENTS
Intelligence is leaving the chat box.
AI systems are beginning to plan, act, transact, coordinate, recover and touch real infrastructure. The next question is not only what they can do — but how we know their actions are correct.
SIGNAL→CAUSE→EVIDENCE→VERIFICATION→ACTION
New Feature · Outcome Provenance
Who Saw the Outcome?
Authorization is only the first half of agent trust. A public implementation comparison converged on separate decision/outcome provenance and exposed a concrete missing axis: source class says how an outcome was established; observer identity + vantage say who saw it and from where.
Read the featureInspect Engineering Signal 008 →
Engineering Signal 001 · External Reproduction
A Memory Contract Survived a Second Implementation
CML’s frozen Current-State Applicability fixture crossed implementation boundaries byte-for-byte. A second public agent-memory implementation now consumes the same test surface, while the next lineage/supersession reproduction gate remains deliberately open.
Open Engineering Signal 001Read the underlying memory model →
Article #004 · Market Experiment 001
Before You Let an AI Agent Move Money
Authorization is not verification. We examine timeout, unknown commit, reconciliation, recovery, invariants and evidence — then ask the market what must be provably true before an agent can move real money.
Read + answer the open question →Protocol · v1.0
Transactional Trust Protocol
OBSERVE → VERIFY → AUTHORIZE → BIND → COMPARE → COMMIT → RECONCILE → PROVE — an experimental protocol for keeping trust connected across consequential agent transitions.
Read the protocol →Science Signal 001 · Verified
X(2370): The Strongest Glueball Case Yet
A media headline called it a new elementary particle. The primary BESIII evidence says something more precise: a composite hadron likely dominated by the lightest 0−+ glueball. We verify the chain — then ask what measurement could actually falsify it.
Read the science contour →
Research release · Published
When Agents Fail
A failure taxonomy becomes useful when it produces tests. Article #003 ships with the first RESONANCE Agent Failure Benchmark: 16 local/sandbox seed cases, a 100-point score and a critical-failure cap.
Read + open benchmark
Verified Report #041 · Constitutional root authority replay / root currentness
The root signature was valid. The root was retired.
Verified Workflow · Pilot
One real workflow. Explicit guarantees.
When a market signal is strong enough, RESONANCE starts small: map one consequential workflow across state, causality, phase, transition, time, recovery, verification and evidence before deciding whether a reusable product should exist.
See the pilotDescribe your workflow →
The RESONANCE Method
Journalism you can inspect.
We separate observation from interpretation and interpretation from prediction. Every serious story is designed to expose the path from claim to evidence.
01Signal
What changed? What is genuinely new?
02Evidence
What primary material supports the claim?
03Causality
What mechanism could explain the change?
04Uncertainty
What remains unknown, disputed or fragile?
05Verification
How could an independent reader check us?
06Action
What should builders, researchers and leaders do next?
Startups & Economy
The agent economy will be built on permissions, payments and proof.
Autonomous software creates new markets, but also new failure modes. RESONANCE follows the infrastructure underneath the hype: compute, identity, wallets, policy, verification, coordination and recovery.
Human
Technology matters because people do.
We cover cognition, learning, creativity, health, flow and meaning without pretending humans are just another component in the stack. Progress should increase agency, not erase it.
Issue 001 · Founding Edition
THE AGE OF AGENTS
The founding issue now connects analysis → framework → benchmark → recovery → authority → execution binding → atomic transition → Transactional Trust Protocol v1.0 → real PostgreSQL → isolation conflict taxonomy → transactional outbox → real HTTP ambiguity → durable execution memory → idempotency retention → clock authority and uncertainty → Temporal ABA → stale-worker fencing → lease renewal → mid-flight commit authority → result handoff → artifact freshness → dependency-aware applicability → causal-model completeness → model currentness → model compatibility proof → compatibility proof authority → revocation propagation / authority-view currentness → authority-head authenticity → authority-head anti-rollback / monotonic currentness → checkpoint storage rollback resistance / trust-history reconstruction → witness-history consistency / equivocation detection → multi-witness quorum consistency / conflicting-majority detection → witness-set currentness / membership-rotation authority → membership-authority anti-replay / monotonic membership currentness → membership-authority equivocation / same-epoch fork detection → recovery-authority equivocation / conflicting-resolution detection → governance-resolution equivocation / conflicting-finality detection → constitutional-root currentness / root anti-replay. The trust path now verifies that even a top-level authenticated root record remains bounded by monotonic history: historical root validity does not silently become current authority after a newer root epoch has been observed.
Open the issue
Market OS
PUBLISH → ASK → LISTEN → BUILD
Selected articles now open a structured market dialogue: content reveals real workflows, real workflows become product signals, pilots create evidence, and only repeated verified demand justifies productization.
See Open Problems