🌸 ASS-OS BLOOM PROMPT — THE RADAR

SOURCES:
  Google Trends    → pytrends, no key, 30-min scan
  Reddit           → PRAW, free key, 10 subreddits
  Hacker News      → Firebase API, no key, /new + /top
  arXiv            → RSS feed, cs.AI + cs.CL + cs.CV
  Product Hunt     → GraphQL API, free key
  YouTube Trending → Data API v3, free quota
  X/Twitter        → v2 API ($100/mo) or skip

INTAKE RATE: every 30 minutes
STORAGE: append-only log (blockchain-style, immutable)
RETENTION: 30 days rolling
FORMAT: {source, timestamp, title, url, metrics, raw_text}

R0 does not judge. R0 does not filter. R0 ingests. The filtering happens at R2. If R0 misses a signal, every ring above it operates on incomplete data. R0 must be greedy.

FROM EACH INGESTED ITEM EXTRACT:
  keywords         → NLP entity extraction
  sentiment        → positive / negative / neutral
  velocity         → Δ engagement / Δ time
  source_authority  → is this from a known signal account?
  cross_mention    → same topic appearing on multiple platforms?
  age_hours        → time since first appearance anywhere

R1 does not score. R1 does not rank. R1 extracts features from noise. The features are the signal. Everything downstream reads R1's output, never R0's raw feed.

GATE RULES:
  PASS if:
    velocity > threshold (rising, not flat)
    AND age < 48 hours (still fresh)
    AND NOT duplicate of existing tracked topic
    AND NOT from known spam/bot source
    AND has minimum engagement floor:
      Reddit: 50 upvotes + 10 comments
      HN: 20 points
      arXiv: mentioned on ≥1 other platform
      YouTube: 10K views in first 6 hours
      Product Hunt: 100 upvotes in first 6 hours

  KILL if:
    already peaked (velocity negative)
    OR age > 48 hours with low velocity
    OR duplicate (deduplicate across sources)
    OR irrelevant (no keyword match to GUILD_EXPERTISE)

R2 is the most important ring. A broken R2 floods R4 with noise (the Guild gets alert fatigue and ignores the brief). A too-tight R2 kills real signals (the Guild misses opportunities). R2 must be calibrated weekly based on R5 feedback.

GUILD_EXPERTISE = [
    "ai development", "ai craft", "llm", "open source ai",
    "ai safety", "ai ethics", "ai tools", "vibe coding",
    "ai jobs", "ai education", "3d ai", "depth estimation",
    "voice synthesis", "ai agents", "scada ai", "edge ai",
    "cost optimization", "prompt engineering", "ai art",
    "react", "three.js", "webxr", "firebase", "python",
    "blockchain", "private chain", "eVGPU", "konomi",
    "EDS awareness", "neurodiversity", "horse tech"
]

RELEVANCE SCORING:
  exact_match      → keyword in GUILD_EXPERTISE: +40
  semantic_match   → embedding similarity > 0.7: +25
  member_history   → guild member has posted on this before: +20
  contrarian_angle → guild can offer unique perspective: +15

R3 answers: "Can we speak to this?" Not "is this trending" — R1 handles that. R3 asks whether the Guild has standing to create content on this topic. A trending topic with zero guild relevance scores zero at R3 regardless of velocity.

SCORE = (
    velocity   × 0.30 +
    freshness  × 0.25 +
    relevance  × 0.20 +
    volume     × 0.15 +
    competition × 0.10
)

CLASSIFICATION:
  fear         → "AI will take jobs" / "AI caused X failure"
  hype         → "New model beats everything" / "AGI soon"
  technical    → new paper / new tool / benchmark results
  drama        → company drama / founder drama / twitter beef
  regulatory   → laws / regulations / government action
  product      → new launch / new feature / new pricing

PEAK ETA ESTIMATION:
  fast_rise + high_volume → peak in 6-12 hours
  moderate_rise + moderate_volume → peak in 12-24 hours
  slow_rise + growing_volume → peak in 24-48 hours

CONTENT SUGGESTION GENERATION:
  per topic_type:
    fear       → "here's why this isn't as bad as it sounds"
                 OR "here's why this is worse than you think"
    hype       → "we tested it. here's what actually happened"
    technical  → "explained in 60 seconds" / "real benchmark"
    drama      → "the real lesson nobody is talking about"
    regulatory → "what this means for builders"
    product    → "first look / honest review"

R4 is where The Radar becomes useful. Raw trends are useless without scoring, classification, and content suggestions. R4 turns "something is happening" into "here's what to do about it."

OUTPUT RULES:
  DAILY BRIEF (7 AM EST):
    Top 3 HOT topics (score > 60) → create today
    Top 3 WARMING topics (score 40-60) → prep content
    Top 3 COOLING topics (score < 40) → skip unless unique angle

  ALERT (real-time, any time):
    Score > 85 → immediate Discord ping to all members
    "BREAKOUT DETECTED: [topic] — create content NOW"

  SUPPRESS:
    Topics that violate Guild values
    Topics that require expertise we don't have
    Topics where being wrong has high consequence
    Duplicate alerts within 4-hour window

  FORMAT:
    Discord bot → daily brief + real-time alerts
    JSON API → dashboard consumption
    DM → personalized per member based on their expertise

R5 prevents alert fatigue. The Radar that cries wolf loses trust. Every suppressed alert is logged so R5 can be audited — are we suppressing too much? Too little?

R6 ASKS:
  Are we catching trends early enough?
    → measure: avg hours before peak at time of first alert
    → target: >12 hours before peak

  Are we catching the RIGHT trends?
    → measure: % of alerts that led to content creation
    → target: >40% conversion

  Are we missing trends?
    → measure: topics that peaked without Radar detection
    → post-mortem: why did R0/R1/R2 miss it?

  Is the scoring working?
    → measure: correlation between score and content performance
    → adjust weights quarterly

  Is the Guild acting on the intel?
    → measure: time from alert to published content
    → target: <6 hours for HOT topics

  WEEKLY REPORT:
    "The Radar caught 23 topics this week.
     12 led to content. 8 performed above average.
     3 trends were missed. Root cause: [X].
     Scoring adjustment: increase velocity weight by 5%."

R6 is why The Radar improves over time. Without R6, the system calcifies. With R6, every week's data makes next week's predictions sharper.

The Radar is the Guild's peripheral vision. The members are the fovea — they focus on creation. The Radar handles detection.

The recursion: The Radar detects trends. The Guild creates content. The content generates data. The data trains The Radar. The Radar detects better. The Guild creates better. The cycle accelerates.

PHASE 1 — MVP ($0/mo, 1 week)
  guild-radar/
  ├── sources/
  │   ├── google_trends.py
  │   ├── reddit.py
  │   ├── hackernews.py
  │   └── arxiv.py
  ├── scoring/
  │   ├── velocity.py
  │   ├── relevance.py
  │   └── classifier.py
  ├── output/
  │   ├── discord_bot.py
  │   ├── json_api.py
  │   └── templates.py
  ├── config.py
  ├── scheduler.py
  ├── requirements.txt
  ├── Dockerfile
  └── README.md

PHASE 2 — INTELLIGENCE ($0/mo, week 2-3)
  + LLM content draft generation (Mistral free)
  + auto-generate LinkedIn / TikTok drafts
  + cross-reference guild repos with trends
  + historical tracking + miss analysis

PHASE 3 — AUTOPILOT ($0-100/mo, week 4+)
  + claim system (member claims topic)
  + performance tracking per radar pick
  + feedback loop to scoring weights
  + content calendar integration
  + X/Twitter API (optional, $100/mo)

THE RADAR — SYSTEM STATUS
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
R0 GROUND    ██████████ 98%  sources online
R1 SIGNAL    ████████░░ 82%  extraction accuracy
R2 GATE      ███████░░░ 71%  signal/noise ratio
R3 RESONANCE ████████░░ 84%  relevance accuracy
R4 PATTERN   ███████░░░ 73%  prediction accuracy
R5 BOUNDARY  █████████░ 91%  brief open rate
R6 OBSERVER  ██████░░░░ 64%  improvement rate
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
OVERALL:     ████████░░ 80%  OPERATIONAL
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━