AI die voorstelt. Mensen die beslissen.

De meeste AI-systemen handelen. OpenClashd niet. Het observeert de wereld, detecteert wat ontbreekt, en legt dat voor aan een mens. Pas na expliciete goedkeuring gebeurt er iets. Elke beslissing is traceerbaar. Niets is impliciet.

AI-systemen die te veel doen

Bestaande AI-frameworks laten agents autonoom beslissen, uitvoeren en leren — zonder dat een mens dat goedkeurt. OpenClashd inverteert dit model. Het systeem detecteert, clustert en stelt voor. De mens keurt goed of verwerpt. Alleen dan wordt iets kennis.

Architecture

OpenClashd web mirrors CLI governance: easy entry, constrained execution, observable outcomes.

System layers

• Mission Control web pages are additive operator views over existing gateway APIs.
• Kernel/runtime semantics stay unchanged; no route bypass into kernel internals.
• Radar, fabric, and knowledge panels rely on token-gated endpoints with explicit fallback.
• CLI and web share the same canonical flow: receive -> inquire -> attend -> attest.

Collision Computer

Traditional computing follows input - algorithm - output. OpenClashd computes differently: it computes collisions between signals. Agents move through CLASHD27 space; when they converge in the same cell, collisions occur. Repeated collisions create residue, residue accumulation leads to emergence, and emergence signals potential discoveries.

De twee kubussen

Het systeem werkt met twee complementaire kubussen die samen de kern van de kennisarchitectuur vormen.

Skill-architectuur

Elke cel in de Zoeker heeft een expliciete, versioneerbare skill. Een skill is geen prompt — het is een gestructureerde eenheid met naam, doel, inputs, outputs en evaluatiecriteria.

Skills leven niet in een bibliotheek — ze leven in een ecologie. Gebruik, context, kwaliteit en effect worden bijgehouden. SafeClash certificeert. Residue herinnert.

Instrument alignment

• Semantic Radar detects cross-domain signal pressure.
• CLASHD27 Field shows where agent signal collisions occur.
• Fabric Radar tracks structural pressure and active cells.
• Discovery Map highlights discovery hotspots in cube space.
• Discovery Log records meaningful collider outcomes over time.
• Collision Events combines radar collisions with fabric occupancy to explain who collided, where, and with which signal.

OpenClashd Observatory Protocol (OOP)

OOP is a lightweight event protocol for sharing discovery signals between observatories. It supports event types: collision, emergence, hypothesis, research_candidate, discovery.

{
  "type": "collision",
  "cell": 14,
  "confidence": "high",
  "source": "observatory-alpha",
  "signals": ["ai-safety", "cryptographic-governance"],
  "timestamp": 1710000000
}

Open Research Network

Multiple observatories can cooperate on the same discovery space. Shared protocol events turn local detections into network knowledge: signals -> collisions -> discoveries -> shared research directions.

AI Observatory

The Observatory monitors collision field activity across all 27 cells. It detects anomalies, emergence pressure, and untrusted agent activity, then surfaces them as scored alerts with severity (high, medium, low).

• Alert engine scores emergence clusters, cross-domain anomalies, and untrusted agent proposals.
• Activity log records challenge creation, escalation, and snapshot events in a bounded ring buffer.
• Cube heatmap maps alerts to 27-cell space for spatial awareness.
• Escalated alerts are routed to Jeeves for governed operator review.

Research Gravity Model

The gravity model assigns mass to each of the 27 semantic cells based on signal accumulation. Cells with concentrated cross-domain signal pressure become gravity hotspots. Gravity scoring uses four bands: blue (cold), green (warm), yellow (active), red (hot).

• Center of mass tracks where research interest converges in cube space.
• Field summary provides distribution across all four bands.
• Hotspots with rank #1-#3 are primary research directions.
• Cross-domain candidates (spanning multiple cells) indicate emerging interdisciplinary topics.

Discovery Engine

Discovery candidates emerge when gravity hotspots, emergence clusters, and signal residue converge. The engine identifies research directions that are not yet established but show convergence pressure.

• Candidates are scored by convergence strength, cross-domain reach, and signal source diversity.
• Each candidate maps to specific cells and axes in CLASHD27 space.
• Cross-domain candidates (multiple axes) carry higher discovery potential.
• Discovery events feed back into the Observatory activity log and Jeeves companion.

Global Idea Telescope

Multiple OpenClashd observatories cooperate to detect emerging research directions. Each observatory contributes Semantic Radar signals, CLASHD27 collision events, Discovery Map hotspots, and Knowledge Seeds into one shared discovery space.

Signals from multiple observatories converge into a global discovery map, allowing the network to observe idea emergence before discoveries are fully established.