Temporal-Arc Reference Arrays

Temporal-Arc Reference Arrays

I think this idea is real enough to publish if we keep the claim discipline tight.

The core problem is simple: multiscale or temporally layered systems are often described as if one layer can simply access another layer's state. In practice that almost never happens directly. What actually moves across the boundary is some compressed, partial, stale, or purpose-shaped representation.

That suggests a better object:

a Temporal-Arc Reference Array — a bounded structured record describing what one temporal arc can hand to another.

You can also read the R/D split as Reference / Dynamical:

• reference because the packet is meant to persist long enough to be consulted
• dynamical because the underlying source state keeps changing, so the packet is always at risk of drift

Defensible now

• Cross-temporal reasoning usually depends on representations, not raw full-state transfer.
• Those representations should state their confidence, timing, and loss profile explicitly.
• A small array contract is more honest than vague language about direct access across scales or times.

Plausible but unproven

• Temporal-Arc Reference Arrays could become a shared scaffold for continuity systems, simulations, and interfaces.
• They may help connect note-level theory to executable control or memory mechanisms.

Speculative

• A mature TARA framework could become a reusable design pattern for cognition, narrative systems, and multiscale control.

Minimal fields

A useful first array should include:

The structured sourceArc and targetArc carry temporalBand (fast/meso/slow), role (observer, chaser, branch, spine, etc.), and cadenceHz — making the coupling legible without requiring consumers to parse naming conventions.

driftEstimate can be a single number or a structured object with per-field rates and a decay model (exponential / linear / step), for cases where position drifts faster than identity.

A Mechanical Operationalization: Worldline Gears

There's a deeper, more operational framing:

Worldlines as computational gears operating at different cadences, with gear ratios enabling steganographic communication across temporal boundaries.

This is an analogy, not a claim about literal mechanics — but it is a productive one because it maps to measurable quantities:

• Gear ratio — the cadence ratio between two arcs (e.g. 3:1), the computational analog of a proper-time rate ratio between clocks
• Phase alignment — when in the source cycle a transfer is attempted; determines synchronization quality
• Slippage tolerance — maximum allowable phase error before the coupling is considered failed
• Information capacity — bits per gear cycle, derivable from ratio and tolerance

Mechanical failure modes (slippage, backlash, tooth-skipping, resonance) become observable diagnostics rather than vague warnings.

Extended fields for gear-coupled TARAs

• gearRatio — cadence relationship ("3:1", "7:3")
• phaseAlignment — normalized phase match at transfer time (0–1)
• slippageTolerance — max allowable phase error
• informationCapacity — theoretical bits per gear cycle
• cadenceHz — operating frequencies of source and target
• mechanicalFailureModes — observed: slippage, backlash, tooth-skipping, resonance, phase-drift

Physics grounding

TARAs are intended to be implementable on any architecture. The way to ensure that is to keep the primitives in contact with first-principles physics.

Causality. Information can only flow forward along the causal cone. A TARA can only be consumed after it was emitted.

Bounded information. Any physical representation has finite information content. Compression and loss are not optional — they are the cost of crossing a temporal boundary.

Coarse-graining and renormalization. Moving from a fast arc to a slower one is the computational analog of a renormalization group (RG) step: fast degrees of freedom are integrated out and an effective description at the coarser scale is what survives. This gives the framework a well-understood physical interpretation:

The lossNotes field is better than standard RG in one way: it keeps a record of what was integrated out. In physics RG that information is typically implicit. Here it is declared.

A testable prediction follows: if this analogy holds, confidence should decay predictably along a chain, and some loss note patterns should appear at every coarse-graining boundary — robust irrelevant operators.

Composition

If TARA(A→B) and TARA(B→C) exist, they compose into TARA(A→C). The composition rules:

• Confidence compounds multiplicatively: confidence(A→C) = confidence(A→B) × confidence(B→C)
• Drift accumulates additively: driftEstimate(A→C) = drift(A→B) + drift(B→C)
• Loss notes merge: union of both loss sets, plus a note that the intermediate representation was consumed
• Validity window takes the minimum: the chain expires when its weakest link expires
• Reconstruction target narrows: the composed target is constrained by both hops

Composition is associative for all these operations, which means chains of arbitrary length are well-defined.

The first concrete test case is the Sandy Chaos diagonal coupling decomposition. Sandy Chaos's Nested Temporal Domains architecture disallows direct slow→fast transfer and requires decomposition through intermediate layers:

D_{observer,slow} → D_{observer,meso} → D_{observed,meso} → D_{observed,fast}

Composing those three TARAs yields: confidence 0.569 (from 0.85 × 0.76 × 0.88), drift 0.38 (accumulated), validity 50 ticks (the tightest window in the chain). The compound loss is substantial — which is exactly why the diagonal shortcut is disallowed. Pretending it's a single clean transfer hides this cost.

Frame-covariance

What survives a change of description frame?

Frame-invariant — these don't change:

• Causal ordering (source emitted before target consumed)
• Loss profile (what was dropped is frame-independent)
• Information capacity bound

Frame-dependent — these change with the observer:

• confidence (depends on the consumer's reconstruction policy)
• driftEstimate (depends on which frame measures time)
• stateSummary contents (source-frame representation)

The frame-invariant quantities define equivalence classes of transfers — what RG would call universality classes.

Retrodiction

TARAs flow forward. Running one backward — inferring source state from a target representation — degrades faster than forward prediction, consistent with entropy increase.

Two reasons: (1) lossNotes declare what was dropped, and that information is genuinely gone — the inverse map is not unique. (2) The source has continued evolving since emission, so the target's model of the source is strictly older than the TARA itself.

directionality is a consumer-side concern, not a schema field. A TARA is always emitted forward. Retrodiction is an inference operation on a consumed TARA, not a property of the record itself.

The discipline rule

No raw cross-arc state access; only bounded reference arrays with declared reconstruction limits.

In gear terms:

No direct state transfer; only gear-coupled representations with explicit ratio, phase, and slip limits.

This blocks: magic omniscience, metaphor drift, hidden uncertainty, post-hoc reinterpretation without declared cost.

Cross-project grounding

TARAs formalize patterns that adjacent projects already use informally.

Sandy Chaos (nested temporal domains) defines a TransferBundle { payload, source_domain, target_domain, latency, distortion, confidence, provenance, validity_window }. The mapping to TARA is direct: source_domain → sourceArc, payload → stateSummary, distortion → lossNotes, validity_window → validityWindow. The neighbor-layer codec (embed → extract → translate → reconstruct) maps to the TARA lifecycle: emission, reception, reconstruction attempt, and bounded model formation.

Yggdrasil (branching continuity architecture) uses branch→spine promotion — a fast→slow transfer with compression, loss judgment, and confidence decision. A promote-durable outcome is a high-confidence TARA accepted by the spine. A no-promote outcome is also a TARA — it exists, records what happened, but its reconstruction target was rejected by policy. A durable trace in Yggdrasil's sense is a consumed TARA whose reconstruction was accepted: the provenance field carries the acceptance record.

Failure conditions

This idea fails if:

• the arrays become decorative jargon,
• drift and loss are declared but ignored downstream,
• reconstruction claims become stronger than the data supports,
• or the concept expands faster than examples and experiments justify.

Mechanical failure is more specific and observable: slippage beyond tolerance, backlash in bidirectional transfer, tooth-skipping at specific cadences, resonance at harmonic ratios.

What exists

A standalone repo with:

• JSON schemas for base TARA and gear-coupled TARA, with $ref inheritance so extensions don't copy-paste the base
• Examples: fast→meso base, fast→meso gear-coupled, Sandy Chaos transfer, Yggdrasil promotion (accepted and declined), Yggdrasil gear-coupled weekly summary, a three-TARA diagonal decomposition chain
• tools/validate.py — validates any TARA JSON against the appropriate schema
• tools/compose.py — given N TARAs in a chain, computes the composed TARA with accumulated drift, compounded confidence, merged loss notes
• tools/drift.py — given a TARA and elapsed ticks, estimates effective confidence under exponential, linear, or step decay
• Mapping docs connecting TARA fields to Sandy Chaos's TransferBundle and Yggdrasil's promotion mechanics
• Theory docs on composition rules, retrodiction semantics, frame-invariants, and the RG connection

That is enough for the idea to exist in public without pretending it is complete.

Closing

This is worth publishing now, but only in its honest form: not as temporal magic, but as a strict representation discipline for systems that need to move information across time.

The mechanical operationalization adds an engineering layer that makes the discipline concrete and testable. The physics grounding — coarse-graining, RG flow, frame-invariance — gives it a principled foundation rather than just a useful metaphor. The composition rules make it composable, not just declarative.

The open questions that remain are gear-level: what are the composition rules for gear fields themselves (phase, ratio, slippage across a chain)? What does a beta function for TARA confidence flow look like under repeated coarse-graining? These are tractable and worth answering once real usage data exists.