The Time and Data Problem: Moving Information Orbitally

# The Time and Data Problem:Moving Information Orbitally

1 · The problem, in plain words

The moment information moves between two places, the hardest fact to agree on is when. Every moving piece of data carries a claim about time, and every receiver has to decide whether to trust it. Systems with many moving parts run into three enemies:

• Disagreement: two machines stamp the same moment differently, and nobody can say which record is right.
• Drift: clocks wander, and a sequence reconstructed from timestamps quietly stops matching what actually happened.
• Lost ordering: when "which came first" is inferred from clock readings instead of the structure of events, cause and effect get swapped.

Here is the thing: this problem is not new, and it has been beaten before. If you have worked in games, video, film, or VFX, you have lived inside its solution for years. Every edit suite, every sound stage, every render farm runs on a house clock: one timing reference that every device locks to. And every frame of media that has ever moved between systems carried SMPTE timecode, an industry standard that has kept frames honest across devices, vendors, and continents since the early 1980s.

The broadcast and film industries learned the hard way what data science is learning now: when streams must interlock, "roughly the same time" is a lie you pay for later. The failure mode is always the same: a record you can no longer trust. Not a crash. Just data whose relationship to reality has quietly rotted.

This is the Time and Data problem. It shows up wherever information moves "orbitally": between machines, between agents, between organizations, between industries, and eventually between planets.

We know because we live it. The same drift had to be solved on our own shop floor first: drift between Actors in an AI estate, between the Estate and its distributed compute, between the NOC and the various cloud providers. Many writers, no shared "now," messages stale on arrival at microsecond scale. The delay is smaller. The problem is identical.

And this is what plagues AI — and will plague ASI. A fleet of agents that cannot trust its own timestamps cannot trust its own memory: state diverges between actors, context rots in flight, and cause and effect quietly swap. The intelligence is only as good as its record of what happened. Scale that fleet toward ASI — machines across datacenters, across distances, someday across planets — and the problem does not shrink; it becomes the discipline that keeps such a mind honest.

2 · What the production world already knows

The film and video world's answer deserves respect, because it worked at planetary scale for decades.

• The house clock. One reference signal, every device genlocked to it. Cameras, switchers, audio desks, recorders: all phase-locked to the same timing spine. Nothing drifts because nothing is free-running.
• SMPTE timecode. Every frame carries an identity, hours to frames (HH:MM:SS:FF), in a format standardized in the early 1980s and still running the industry today. Timecode rides audio tracks (LTC), hides in the picture (VITC), speaks to musical instruments (MIDI timecode), and travels with every EDL, AAF, and conform that crosses a facility boundary.
• The frame is a unit of causality. Two events in the same frame have no order between them, and everyone in the industry knows it. That is the same honesty as "unknown stays unknown": the discipline tells you what it cannot resolve.
• Fractional frame rates taught provenance the hard way. NTSC's 29.97 fps required drop-frame timecode just to keep the numbers near the wall clock, and the 23.976 "23.98" variant still bites conforms today. When the rate is not what it says it is, the label is everything. Every old hand in post has a scar from a mislabeled rate.

The lesson the production world teaches: a named, labelled, locked time discipline is what lets high-speed media interlock without trust problems. Frame-accurate collaboration between hundreds of vendors is possible because no one is guessing about when.

3 · The answer that landed: LuxClock, the modern house clock

LuxClock is the master timing and state reference for our operation: an indelible record of what happened, when, and what was in flight at any point in time. It is the house-clock pattern, generalized from the studio to a fleet of AI agents. Two of its rules were amended and adopted on October 14, 2026, and they carry the synthesis.

Rule 1 (amended): UTC for operational records, local time for people

Every persisted operational timestamp is ISO 8601 UTC. No local zone ever touches the record. Human-facing local times are rendered views: local first, in a named zone, with the date and zone shown so a copied display stays interpretable. Scientific clock measurements keep their native scale and units and are never mislabeled UTC. The production world knows this as keeping your master rate honest and never letting a mislabeled conform leave the building.

Rule 8 (adopted): Three times, ordered by cause

Every event records three temporal roles, each with explicit provenance:

• Source event time: the reading of an identified source clock when the event occurred. Proper time only when that physical interpretation is supported.
• Coordinate event time: the event's time in an explicitly named time scale and reference frame. UTC for current Earth operations. Off-Earth applications may use the appropriate local coordinate scale, and separately labelled TCB or TDB, the relativistic time scales used in deep-space navigation.
• Receipt time: observed at a specified receiving or ingestion boundary. The moment the information actually arrived.

Epoch, units, conversion model and uncertainty are recorded where needed. Unknown stays unknown; derived values are labelled. Causal order comes from documented event relations, never inferred from clock order alone. The record is forward only: corrections are new events referencing the original, never edits of it.

The supporting discipline that makes the rules work is the SMPTE inheritance, formalized: append-only event log (lines are never edited, deleted, or reordered); SHA-256 checksum on every event line, with a verify command that walks the whole log and reports tampering; git-backed writes (cryptographic commit hashes, distributed cloning, offline readability, full audit trail); a daily checkpoint at a fixed time capturing all in-flight work as the recovery anchor; session discipline (every working session logs its start and clean end, and a session that started and never ended is classified as lost at the next checkpoint, so the record knows what it does not know); and a drift event in the taxonomy, so the clock watches itself.

4 · Why this is a data science discipline

In data science, time is the join key of reality. Almost every analytical failure that is not a code bug is a failure of temporal semantics:

• Temporal joins lie when the compared timestamps mean different things (one says when it happened, the other says when it was recorded). The three-times rule names the difference and makes the join honest.
• Ordering lies when sequence is inferred from clocks. Causal order from documented relations is the only reliable ordering.
• Metrics lie outside their window. A rate, a cost per unit, a throughput figure only means something inside a declared time window. We run our own numbers this way: cost per million tokens over a 30-day window, one ledger, stated explicitly.
• Replica lag is a data anomaly, not just an infrastructure nuisance: the same query against two stores returns different truths, and receipt time at a named boundary is how you reconcile them.
• Astronomical time is the missing ingredient. UTC is atomic seconds nudged by leap seconds to track the Earth's rotation; TAI is pure atomic time; GPS time runs on its own epoch with no leap seconds at all. Systems that mix scales silently break (the leap-second outage class is famous). Rule 1's "native scale, never mislabeled" and rule 8's "named scale, epoch, uncertainty" are the fix. This is the synthesis: data science's provenance discipline applied to the astronomer's clock discipline.
• Drift detection as a first-class event means the monitoring of time itself is data, feeding the same analysis it protects.

5 · Why it runs what we are building

Our company runs as a fleet of AI agents across machines: agents that build, test, and ship software and games, coordinate over a message plane, and produce work products continuously. That is a distributed system with many clocks and many writers, and it is also, structurally, a studio: the agents are the tape decks, the message plane is the SDI spine, and LuxClock is the house sync. We lived the failure modes before the policy existed: records that appeared and vanished in transit, silent replication lag, sequences that could not be reconstructed after a crash. Every one is a Time and Data problem, answered by the same discipline the post house uses for a conform.

The product point: the discipline itself is the product pattern. A fleet of agents with a trustworthy clock is an estate; an estate is a company that remembers correctly. That is sellable to anyone who runs high-speed data: game backends, pipeline operators, financial systems, sensor networks.

6 · The Mars connection, and the disciplines that prove it works

Scale the same problem up. SpaceX, the UAE's Mars 2117 program, China's announced crewed Mars ambitions, and the wider multiplanetary movement all inherit the identical problem at interplanetary scale.

• No shared "now": one-way light time between Earth and Mars runs from about 3 minutes at closest approach to about 22 minutes at maximum. Every cross-planet message is stale on arrival. Receipt time at a named boundary stops being a nicety and becomes the only way to reason.
• Communication has holes: around solar conjunction, roughly every 26 months (the same synodic cycle that gates launch windows), the link degrades or blacks out for days to weeks. The Mars side must operate autonomously, keep its own append-only record, and reconcile when the window reopens. That is checkpoint and replay, literally.
• A sol is not a day: Mars rotates in 24 hours 37 minutes. Local solar time, mission time, and Earth UTC coexist, and each must be a named, labelled scale or the record rots.
• Relativity stops being optional: deep-space navigation already runs on the IAU's relativistic time scales, TCB (Barycentric Coordinate Time) and TDB (Barycentric Dynamical Time), and rule 8 names them explicitly. The rule was written for a machine fleet and already speaks the language of deep space.
• Order across planets is causal, not chronological: with minutes of delay both directions, timestamps alone can never establish what caused what. Causal order from documented relations is the only ordering that survives a light delay.

And here is the precedent that proves the pattern works at distance: VLBI, very-long-baseline interferometry. Radio telescopes on different continents (and eventually, in one demonstration, from space) point at the same source, each recording against its own hydrogen maser atomic clock. The data is not streamed live; it is stamped locally, shipped, and correlated after the fact. The Event Horizon Telescope used exactly this to image a black hole: eight sites, atomic-clock-stamped recordings, hard drives flown across the world, and one trusted image reconstructed afterward. Record now with a named local clock, reconcile later at a boundary, and trust the result. That is the three-times rule, already running at interplanetary scale, decades before the first settlement.

Blockchain arrived at the same place from the economics side: block ordering comes from parent-hash chains (documented event relations), not wall clocks, and the ledger is append-only with checksums by construction. The games world lives it as rollback netcode and deterministic replays: predict fast, reconcile against the authoritative record, and the replay verifies itself. Every domain that needs high-speed data to also be trusted data lands on the same discipline.

7 · The bottom line

• Time is the hardest fact in any distributed record, and the first fact to rot.
• The production world proved the fix decades ago: house clock, SMPTE timecode, named rates, frame-honest causality.
• The fix is a discipline, not a product purchase: UTC in the record, local rendered for people, three named times per event, append-only, checksummed, checkpointed, forward only.
• For data science, it is what makes temporal data mean what it claims, with astronomical time as the provenance layer.
• For an agent fleet, it is the house sync that makes the fleet's memory trustworthy at speed.
• For Mars, it is the same discipline at light delay, and VLBI has already run the pattern from here to a black hole.