← Back to blog

The Miyake Event Problem

What dendrochronology’s solution to the floating-chronology problem teaches distributed systems and agent coordination about the difference between better clocks and better anchors.

Published April 2026 · 9 min read

In 2021, a team led by Margot Kuitems at the University of Groningen pinned the Viking settlement at L’Anse aux Meadows — the only confirmed pre-Columbian European site in the Americas — to a single year: 1021 CE. Before that work, the best anyone could say was “roughly 1000 CE,” with multi-decade uncertainty.1

They didn’t build a better mass spectrometer. They didn’t invent a new statistical method. They found a cosmic-ray spike from 993 CE recorded in three separate pieces of wood, counted 28 rings to the bark, and arrived at a date no one can argue with.

The precision improvement was 40–120x over standard radiocarbon dating — achieved not by better measurement but by identifying the right external signal to anchor to.2

Your distributed system has the same problem those archaeologists had before 2012. It has a floating chronology.

The floating chronology

Before Miyake events were discovered, dendrochronology had an elegant but limited solution to dating. You could count tree rings. You could measure relative widths, identify drought years, see that Year 47 was drier than Year 46. But without an external anchor, the sequence floated — it could be off by decades or centuries, and nothing internal would tell you how far.3

Crossdating — matching the distinctive pattern of wide and narrow rings between trees across sites — solved this regionally. Andrew Ellicott Douglass formalized the method at the University of Arizona in the 1920s. It works because climate is a shared signal: trees in the same region experience the same drought years, the same cold snaps.4

But crossdating only works within a region’s master chronology. A Japanese cedar can’t be crossdated against a California bristlecone pine based on ring width alone. For global anchoring, you need a signal that every tree on Earth records simultaneously.

In 2012, Fusa Miyake and colleagues at Nagoya University found one. A 1.2% single-year spike in atmospheric carbon-14 in Japanese cedar rings for 774–775 CE — roughly 20 times the normal year-to-year variation. A second event was confirmed for 993–994 CE. Nine confirmed events have since been identified, with the oldest at approximately 717 BCE.5

The physics: extreme solar particle events — coronal mass ejections orders of magnitude stronger than any instrumentally observed — that enhance cosmogenic-isotope production in the upper atmosphere. The spike is recorded in every growing tree, every ice core, everywhere on Earth, in the same calendar year. No coordination required. No shared protocol. No trust in any single observatory.

Five properties make a Miyake event uniquely powerful as a dating anchor:

PropertyWhat it means
GlobalEvery tree, every ice core, everywhere on Earth
SimultaneousAll recorded in the same calendar year
Unambiguous20x above background noise — impossible to mistake
IndeliblePhysically imprinted in wood and ice, cannot be altered retroactively
Independently verifiableAny lab can extract the spike from any sample

No other natural dating signal has all five.

The same problem, worse

The moment you distribute a system across multiple machines, “now” becomes ambiguous and “before” and “after” lose their absolute meaning. Leslie Lamport formalized this in 1978: in a distributed system, you cannot rely on physical clocks alone to establish event ordering because clocks drift, networks have variable latency, and there is no single reference frame.6

This is exactly the floating chronology problem. A single node’s log is internally consistent — Event A happened before Event B on this machine. But across nodes, without an external anchor, you can’t reliably say whether Node 1’s Event A happened before or after Node 2’s Event B. The logs float.

Lamport timestamps solved this by tracking causality rather than wall-clock time — the distributed-systems equivalent of crossdating. They establish relative ordering between events that share a causal chain, just as crossdating establishes relative dating between trees that share a climatic signal. But like regional crossdating, they break down at the boundaries: two causally disconnected event streams can’t be ordered by logical clocks alone.7

AI agent systems face this in a harder form:

Why better internal clocks can’t solve it

Google Spanner takes the brute-force approach: deploy GPS receivers and atomic clocks in every data center, synchronize them into TrueTime — a globally distributed clock service that represents time as an uncertainty interval rather than a single timestamp. Before reporting a transaction as committed, Spanner waits out the uncertainty interval — about 7 milliseconds — guaranteeing no subsequent transaction can commit at an earlier timestamp.11

CockroachDB, without atomic clocks, uses NTP, which produces 100–250ms offsets — 14–36x worse than Spanner’s 7ms. When clock drift exceeds the configured maximum, a CockroachDB node self-terminates rather than risk producing inconsistent timestamps.12

That self-termination is remarkable: the system would rather die than lie about time. It maps to dendrochronology’s practice of marking uncertain dates with a “?” suffix and refusing to report a date rather than reporting an unreliable one. Both disciplines understand that a timestamp you can’t trust is worse than no timestamp at all.

But neither Spanner nor CockroachDB is the Miyake-event approach. Both try to improve the internal clock rather than finding external universal anchors. Spanner works brilliantly — if you have Google-scale infrastructure. CockroachDB shows the cost of not having that: wider uncertainty windows, more retries, occasional self-termination. The L’Anse aux Meadows team didn’t build a better mass spectrometer. They found the right signal.

Digital Miyake events

Ethereum’s Beacon Chain divides time into slots of exactly 12 seconds, with 32 slots making an epoch. Finalization occurs via Casper FFG: once two-thirds of validators attest, a block becomes irreversible. Reverting a finalized block would require burning one-third of all staked ETH — over $30 billion at current prices.13

Score blockchain finalization against the five-property framework:

CandidateGlobalSimultaneousUnambiguousIndelibleVerifiable
Miyake eventYesYesYesYesYes
Ethereum finalized blockPartialYes (~12s)YesYes ($30B+)Yes
Bitcoin block hashPartialYes (~10 min)YesYesYes
NTP timestampPartialYesYesNo (spoofable)No

Blockchain finalization scores 4.5 out of 5 — the closest digital analog to a Miyake spike. An agent that records “I observed state X as of Ethereum block 22,456,789” has made a timestamp claim that can be independently verified by anyone, forever, without trusting the agent’s local clock.

The ecosystem is already building this. Chainpoint aggregates data hashes into a Merkle tree, publishes the root in a Bitcoin transaction, and generates a self-contained proof linking original data to the blockchain anchor — verifiable without reliance on any trusted third party.14 OpenTimestamps and OriginStamp implement the same pattern. This is crossdating for digital timestamps. The vocabulary is different. The information-theoretic structure is identical.

Trees are better archival substrates than blockchains

A tree passively records the Miyake signal without intentional participation. It doesn’t need to be online, running software, or connected to a network. It just grows. Blockchain anchoring requires active participation: hash submission, transaction fees, network connectivity. The biological substrate is more robust than the digital one for long-term timestamping — which is why we can date 7,000-year-old bristlecone pines but can’t verify a 15-year-old digital timestamp without the original chain.

And here is the deeper irony: the 774–775 CE Miyake event was at least 10 times stronger than the 1859 Carrington Event, releasing energy equivalent to roughly 660 billion Hiroshima bombs.15 If it happened today, it would catastrophically disrupt power grids, satellites, GPS, and unprotected electronics. Nature’s best timestamp generator is also a civilization-ending threat to the very digital systems that most need universal time anchors.

Where the analogy breaks

Bitcoin and Ethereum blocks aren’t truly global in the way carbon-14 is — they require network participation, not passive absorption. But verification is permissionless and trustless. Any agent can check any block hash at any time without participating in mining or staking. Active-query universality is sufficient for digital systems, even if it lacks the elegance of a cosmic-ray spike deposited in every tree regardless of intent.

NTP is good enough for many coordination tasks. Its 1–50ms accuracy handles the majority of use cases. But it requires trusting the time server. In December 2025, a power failure at NIST’s Boulder campus caused the UTC(NIST) signal to drift approximately four microseconds while servers continued responding to requests — answering with time they no longer had authority to report.16 “Trust the clock server” is the same epistemological error as “trust the single tree.”

Crossdating beats trusting any single log

Dendrochronologists have been solving distributed time synchronization for over a century. Douglass’s crossdating predates Lamport’s logical clocks by more than 50 years. The tree-ring community developed what amounts to a Byzantine fault-tolerant consensus protocol — majority-rules pattern matching with explicit outlier detection — before computer science had the vocabulary to describe it.

The lesson is convergent evolution. Two completely unrelated fields, facing the same information-theoretic constraint — unreliable local clocks, no guaranteed message delivery, potential for missing or spurious data, the need to reconstruct absolute chronology from relative sequences — arrived at the same solution architecture. The fix is not better clocks. The fix is better anchors.

The 993 CE spike had been sitting in every tree alive that year for a thousand years before anyone thought to look for it. The on-chain anchors are already there. The question is whether your system is looking.


Sources

  1. Kuitems et al. 2021, Nature 601:388–391. Confirmed via ScienceAlert, Smithsonian Magazine, and ZME Science (October 2021).
  2. Standard radiocarbon dating with modern AMS instruments gives ±20–60 year precision. Miyake-event anchoring gives ±1 growing season.
  3. Dendrochronology fundamentals: floating chronologies require external calibration to become absolute chronologies.
  4. Andrew Ellicott Douglass, Laboratory of Tree-Ring Research (LTRR), University of Arizona, 1920s. Douglass 1941, Journal of Forestry 39:825–831.
  5. Miyake et al. 2012, Nature 486:240–242; Miyake et al. 2013, Nature Communications 4:1748. Nine confirmed events: Scientific Frontline 2025; Royal Society Proceedings A 2022/0497.
  6. Lamport 1978, “Time, Clocks, and the Ordering of Events in a Distributed System.”
  7. Lamport timestamps and vector clocks (Fidge 1988, Mattern 1989) establish causal ordering across distributed processes.
  8. Oracle Developers, “Agent Memory: Why Your AI Has Amnesia and How to Fix It,” 2026.
  9. Augment Code, “Agent Memory vs Context Engineering,” 2026.
  10. Augment Code, “Why Multi-Agent LLM Systems Fail and How to Fix Them,” 2026.
  11. Google Spanner TrueTime documentation; Corbett et al. 2012, OSDI. CockroachDB blog, “Living Without Atomic Clocks.”
  12. CockroachDB blog: if clock drift exceeds configured maximum, the node self-terminates rather than risk consistency violations.
  13. ethos.dev, “The Beacon Chain Explained”; Ethereum consensus specs.
  14. Chainpoint.org: “Chainpoint proofs can be verified without reliance on a trusted third party.”
  15. Spaceweatherarchive.com 2025; davidmaiolo.com 2024.
  16. TechRadar, 2025-12-24, NIST time service outage report.

The fix is not better clocks. It is better anchors.

That is the dendrochronologist’s lesson for every distributed system: trust external signals you can independently verify, not internal clocks you have to take on faith. Chain of Consciousness applies this to agent provenance — every action is anchored to a verifiable external record, building a chronology that doesn’t float. Not “trust this agent’s log.” Verify which events it can prove it witnessed, in what order, anchored to what.

See a verified provenance chain · Hosted CoC · pip install chain-of-consciousness · npm install chain-of-consciousness