Knowledge Verification System

The Loom
Protocol

A knowledge reward and quality system where every piece of memory earns its place through verification, forming a woven fabric of trusted intelligence.

0.95
Max Reward
6
Trust Tiers
Knowledge Threads
Scroll to Explore
Trust Hierarchy

The Reward Ladder

Knowledge climbs the ladder through verification. Higher trust means higher reward scores—and priority in every context window.

🏆
Facts

Verified, confirmed, cross-checked knowledge. The bedrock of the knowledge graph. These threads never break.

Qdrant runs on port 6333
Climb: Multiple independent sources confirm
0.95
Never deleted. Facts accumulate over time, creating a permanent, verifiable foundation. Each fact strengthens the threads connected to it.
Approved choices with documented rationale. Decisions can be revisited if circumstances change.
0.90
Decisions
⚖️

Approved choices with documented rationale. The executive layer of knowledge—actionable, accountable.

Use MiniMax for subagents
Climb: Documented outcomes prove effective
Fall: New evidence contradicts rationale
📊
Reports

Status documents, summaries, and analysis. Timely snapshots of understanding at a point in time.

Weekly system status digest
Climb: Key findings validated by subsequent events
0.85
Time-sensitive. Reports reflect the state of knowledge at creation. Their value degrades gracefully as time passes.
Subject to Darwinism. Predictions evolve—accuracy is tracked, scores adjust up or down based on real-world outcomes.
0.50
Predictions
🔮

Future forecasts tracked for accuracy. The proving ground for foresight—where predictions are validated by time.

AI costs drop 50% by 2027
Rise: Outcome matches prediction
Drop: Outcome contradicts forecast
👁️
Observations

Patterns noticed but not yet verified. The frontier of discovery—seeds that may grow into facts.

Users prefer evening queries
Climb: Pattern repeats, survives testing
0.40
Promising threads. Weak but valuable—every fact started as an observation. Nurture them through verification.
All knowledge starts here. Wild guesses and speculation—the raw material of discovery. Every verified truth began as a hypothesis.
0.30
Hypotheses
💭

Speculation, wild guesses, untested ideas. The creative layer—where new threads first emerge.

Maybe GPT-5 arrives Q3 2026
Climb: Initial observations support theory
Knowledge Evolution

The Verification Path

How a single hypothesis climbs the ladder to become irrefutable fact. Each step adds weight, each verification strengthens the thread.

💭 0.30
Hypothesis
Initial speculation or guess
👁️ 0.35
Observed
Pattern noticed multiple times
🔬 0.40
Tested
Actively validated under conditions
📚 0.60
Confirmed
2+ independent sources agree
🌍 0.80
Accepted
Widely recognized as true
🏆 0.95
Verified Fact
Cross-checked, permanent knowledge
Adaptive Intelligence

Darwinism in Action

Predictions are living knowledge—they evolve. Success breeds trust. Failure prompts adaptation. The system learns from reality.

Prediction Confirmed
Verified
"ALGO predicted: MiniMax will handle 10K+ daily requests by Q1 2026"
Outcome: MiniMax reached 12,400 daily requests in January 2026
0.50 0.72 +0.22
Prediction Failed
Incorrect
"ALGO predicted: OpenAI releases GPT-5 in November 2025"
Outcome: No major release from OpenAI in that window
0.50 0.31 −0.19
🏅 Prediction Accuracy Leaderboard
#1 ALGO / Infrastructure +0.72 87% accuracy
#2 ALGO / Market Trends +0.45 72% accuracy
#3 ALGO / User Behavior +0.31 68% accuracy
#4 ALGO / Tech Releases −0.19 45% accuracy
Vector Database

Integration with Qdrant

Every vector in Qdrant carries a loom_score payload. The database becomes a living knowledge graph where trust is searchable.

1
Store Knowledge with Reward Score
Every piece of knowledge is embedded and stored alongside its verification tier
// Qdrant payload structure { "content": "Qdrant runs on port 6333", "loom_score": 0.95, "type": "fact", "verified_at": "2026-01-15" }
2
Query with Trust Filter
"Find only verified facts about this topic" — filter by minimum loom_score
search( collection: "knowledge", query: "find facts about vector databases", "filter": { "loom_score": { "$gte": 0.90 } }, limit: 10 )
3
Context Window Prioritization
When building context, higher loom_score = higher priority for limited token budget
// Priority scoring for context assembly "context_priority": (loom_score × 0.7) + (recency × 0.3) // Facts (0.95) rank above decisions (0.90) above reports (0.85)...
Vector + Payload = Trusted Knowledge
Vector Embedding
[0.123, -0.456, 0.789, ...] 1536 dimensions
+
Payload (loom_score)
loom_score: 0.95
type: "fact"
verified_by: 3 sources
created: 2026-01-15
The Metaphor

The Weaving Analogy

Knowledge threads connect and interlock. Strong threads (facts) create a tight weave. Weak threads (hypotheses) form loose threads that can be tightened over time.

🧵
Knowledge Threads
Each piece of knowledge is a thread. Relationships between knowledge items are connections woven into the fabric. The stronger the knowledge, the more weight it carries in the weave.
🔗
Relationship Links
[[A]] VERB [[B]] — knowledge connects through typed relationships. Facts link to facts strongly. A hypothesis links to facts loosely until verification strengthens the thread.
🕸️
The Knowledge Fabric
The entire knowledge graph is a woven textile. Pull on a fact, and you see all the threads it connects to. A tight weave (many verified facts) is strong and reliable.
📈
Tightening the Weave
As observations become facts and hypotheses get verified, threads thicken and brighten. The fabric becomes denser, more interconnected, more reliable.
Fact-to-Fact
Decision/Report
Hypothesis
Real Knowledge

Items from our Exocortex

Actual pieces of knowledge stored in our knowledge graph with their loom scores. Every item has a home.

🏆 Fact 0.95
"Qdrant vector database runs on port 6333 by default."
3 sources confirmed
📅 2026-01-10
⚖️ Decision 0.90
"Use MiniMax for subagent reasoning workloads—better cost/performance than alternatives."
Rationale documented
📅 2026-02-03
📊 Report 0.85
"Weekly status: System uptime 99.97%, 1,247 queries processed, 3 incidents resolved."
Auto-generated
📅 2026-03-15
🔮 Prediction 0.50
"AI infrastructure costs will drop 50% by 2027 due to competition and efficiency gains."
Pending outcome
📅 2026-01-20
👁️ Observation 0.40
"Users seem to prefer evening sessions (8-11 PM) for complex analytical queries."
Pattern detected
📅 2026-03-10
💭 Hypothesis 0.30
"Maybe GPT-5 arrives Q3 2026—speculative, based on release patterns."
Wild guess
📅 2026-02-28
Knowledge Metrics

System Statistics

A snapshot of our collective knowledge—how much we've verified, what we're still testing, and where the gaps are.

🏆
847
Facts
100% verified
⚖️
234
Decisions
94% active
📊
1,892
Reports
Archived
🔮
156
Predictions
71% accuracy
👁️
423
Observations
38% promoted
💭
2,341
Hypotheses
12% confirmed
Verification Progress by Topic
Infrastructure 92% facts
Decisions & Strategy 78% decisions
User Behavior 45% observations
Market Predictions 62% hypotheses
System Performance 83% facts