CHRONONAV V2
REPRODUCIBLESIMULATED1,847.21
10,000-trial frozen-policy blind validation
Mean true utility vs. 1,315.77 EDF
Simulated blind trials under a frozen policy. Not a production cluster measurement.
EDEN Refinery
EDEN is a value-aware control layer for computation.
Turn raw compute, data and bandwidth into verified useful value.
Problem
Traditional infrastructure asks: can this workload run?
Every unnecessary token, inference, memory movement, network transfer, recomputation or low-value operation has a cost in compute, energy, bandwidth and time. Most systems measure how much they performed. They do not measure how much of it was useful.
Thesis
EDEN asks: what portion creates the most value per unit of resource?
EDEN is a value-aware control layer for computation. It determines what work matters, what can be reused or removed, when remaining work should execute, where resources should be allocated, and how the resulting decisions can be independently verified.
Live refinery
Workload in. Verified value out.
AI INFRASTRUCTURE · Cached prefixes, repeated system prompts, mixed unique completions. Envelope 48.0 MB · deadline 2400 ms · compute 1800 ms
COMPUTE WINDOW
1800 ms
envelope
BANDWIDTH
24 MB
envelope
PAYLOAD OUT
—
awaiting classify
DEADLINE
2400 ms
window open
INPUT
Raw workload envelope: bytes, deadline, compute, bandwidth, energy.
DECISION
Bind telemetry sources and freeze the observation window.
EVIDENCE
IMPLEMENTEDClient demonstration senses a synthetic envelope. Physical edge telemetry is separately labelled PARTIALLY MEASURED.
MARBLE — PENDING VERIFY
A Marble is minted when the run reaches VERIFY: run ID, SHA-256, timestamp, commitments, policy version, verification status.
Investor-grade proof
What is proven, and under which label.
Every metric carries an evidence class. Simulated or modelled results are never presented as physical measurements.
REUSE BENCHMARK
REPRODUCIBLESYNTHETIC79.9%
High-reuse synthetic workload
Reuse at 50k workload size
Synthetic reference test. Real-world reuse is workload-specific and must be measured per deployment.
MERKLE VERIFICATION
REPRODUCIBLE100 / 100
1 GiB commitment experiment
1,048,576 packets · sampled proofs valid
Offline reproducible experiment. Not an externally anchored public timestamp.
PHYSICAL EDGE NODE
IMPLEMENTEDPARTIALLY MEASUREDIMPLEMENTED
Android / Termux execution environment
Local compute, network and energy telemetry tested
Components ran on-device. This is not a claim of fleet-scale edge production.
Evidence lab
Methodology, sample size, and hashes.
Don't trust the claim. Verify the evidence class. Anything simulated says so. Anything demonstrated only locally says so. Anything measured on x86-64 rather than Termux says so.
| SURFACE | CLAIM | N | RESULT | CLASS | ID |
|---|---|---|---|---|---|
| ChronoNav V2 | Frozen-policy scheduler exceeds EDF on mean true utility Blind validation, frozen policy, synthetic task arrivals. No online learning during the trial. | 10,000 trials | 1,847.21 vs 1,315.77 EDF | REPRODUCIBLESIMULATED | a3f1c9e2b7d04c18 |
| Refinery reuse | High-reuse synthetic retains most bytes as KEEP/STRUCTURE Fixed corpus, repeated prefixes, deterministic classifier. Not a live user trace. | 50,000 units | ~79.9% reuse | REPRODUCIBLESYNTHETIC | b81e0c44d2aa9176 |
| Marble / Merkle | Packet commitments verify against a 1 GiB tree SHA-256 Merkle tree over 1 KiB packets. Sampled inclusion proofs. | 1,048,576 packets · 100 sampled proofs | 100 / 100 valid | REPRODUCIBLE | c0d9f77a1e5b3320 |
| Marble suite | Identity, provenance and tamper fixtures hold Unit + property tests including signature swap, payload rewrite, policy downgrade. | 52 tests · 12 destructive tamper fixtures | 52 passing, both signing backends | REPRODUCIBLE | d4aa19b0ce78f201 |
| Primary mint rate | Local mint throughput on x86-64 container Measured on x86-64 container. Not Termux. Not a networked service SLA. | timed run, x86-64 container | 1,031.9 marbles / s | MEASURED | e7c2b1d98a0045fe |
| Physical edge node | EDEN components execute on Android / Termux On-device run. Energy figures are device-reported, not calorimeter-grade. | single device, lab | compute + network + energy telemetry exercised | IMPLEMENTEDPARTIALLY MEASURED | — |
| Public anchor | Externally timestamped immutability Local marbles are tamper-evident against payload edits. They are not immutable against mint-key rewrite without an external anchor. | — | Not configured | PROPOSED | — |
| AOK clearance kernel | Evidence class is bound to admissible action before execute or public claim Rule table over labelled capabilities. This console is not a production admission controller. | policy aok-v1.0.2 · 12 capabilities · 6 actions | Deterministic CLEARED / HOLD / DENIED | IMPLEMENTEDDEMONSTRATION | f1a0c3e84b229176 |
Economics
A labelled counterfactual, not an asserted saving.
COUNTERFACTUALESTIMATE
Enter a workload cost envelope. EDEN applies the published synthetic reuse rate and the ChronoNav simulated utility ratio only as scenario factors. Nothing below is a measured production saving on your fleet.
Baseline envelope
$24,800
Token cost (derived)
$3,200
Reuse reduction
COUNTERFACTUAL
−$3,283
VOID / payload reduction
ESTIMATE
−$1,325
ChronoNav schedule factor
SIMULATED
−$7,374
EDEN scenario cost
$12,818
Delta vs baseline
$11,982 (48.3%)
Counterfactual / estimate. The 79.9% reuse figure is a synthetic reference. The 1.40 utility ratio is from 10,000 simulated ChronoNav trials versus EDF. Do not treat this panel as a quote, SLA or measured ROI.
Platform
Six modules. One control architecture.
REFINERY
Decomposition and value classification
Turns a workload into KEEP / STRUCTURE / DETAIL / RESIDUAL / VOID and records why.
CHRONONAV
Adaptive scheduling under changing resource constraints
Decides when remaining work should execute given compute, bandwidth, energy and deadline.
PROGRESSIVE RETURN
Maximise useful output before a window closes
If the budget runs out, the most valuable subset should already have landed.
MARBLE
Cryptographic provenance and verification
Every persistent decision carries identity, a hash, a policy version and an evidence class that cannot silently increase.
AOK
Attested operational kernel
Reads the evidence class before a workload is observed, modelled, executed or claimed. Unsupported claims cannot silently become operational.
AURA
Distributed / edge / orbital infrastructure research
An R&D extension of the same control architecture. Modelled geometry is labelled as modelled. Physical RF is not claimed.
AURA — DISTRIBUTED INFRASTRUCTURE R&D
Shown last, and labelled. Ambition without a truth panel is a pitch deck. This is the truth panel.
- Orbital geometryMODELLED
- SGP4 positioningMODELLED
- DopplerMODELLED
- FSPLMODELLED
- Physical satellite RF receptionNOT MEASURED
- Operational satellite networkNOT CLAIMED
Markets
One control architecture. Multiple constrained-compute markets.
These are not seven unrelated businesses. They are the same question asked wherever compute, energy, bandwidth or time is scarce.
01
AI INFRASTRUCTURE
Token, KV-cache and model-call waste under latency.
02
DATA CENTRES
What to run, defer, or drop when power and cooling bind.
03
EDGE COMPUTE
Useful work on devices that cannot pretend they are the cloud.
04
TELECOMMUNICATIONS
Prioritised payload under bandwidth and energy.
05
RESILIENT / DEFENCE SYSTEMS
Degraded networks still need a value order.
06
FINANCIAL COMPUTE
Deterministic replay, provenance, deadline-true scheduling.
07
SPACE INFRASTRUCTURE
Research path (AURA). Modelled, not operational.
Pilot
Prove the economics before changing production.
The commercial entry point is shadow mode: EDEN observes, classifies and calculates alternative strategies. Execution control is a later decision.