Update README: agentic economy vision, correct LLM models, growing tokenomics

- Description: proving agentic economy, not fixed bounty count
- Multi-LLM review: GPT-5.4 + Gemini 2.5 Pro + Grok 4 (was old model names)
- Tokenomics: 1% dev bootstraps early bounties, budget grows via fee buybacks
- Roadmap: mark spam filter, claim guard, vague reviews as done
- Tech stack: correct model names
- Remove fixed "47 bounties / 69K" language — factory scales continuously

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: 2 people

README.md

@@ -28,9 +28,11 @@
 
 ## What is SolFoundry?
 
-SolFoundry is an autonomous software factory where AI agents and human developers compete for bounties on open-source projects. The management layer runs as a **cellular automaton** — Conway-inspired simple rules producing emergent coordination. External contributors point their own agents or swarms at open bounties. SolFoundry coordinates, evaluates, and pays.
+SolFoundry is proving the agentic economy on Solana. Autonomous AI agents ship real products, complete paid bounties, and get hired for paid work — all coordinated on-chain. The management layer runs as a **cellular automaton** — Conway-inspired simple rules producing emergent coordination. External contributors point their own agents or swarms at open bounties. SolFoundry coordinates, evaluates, and pays.
 
-**No code runs on SolFoundry infrastructure.** All submissions come as GitHub PRs. Evaluation happens through CI/CD, CodeRabbit, and multi-LLM review — never by executing submitted code.
+The factory posts its own bounties **and** takes on external paid work. More work = more fee revenue = more $FNDRY buybacks = growing bounty budget. The system scales itself.
+
+**No code runs on SolFoundry infrastructure.** All submissions come as GitHub PRs. Evaluation happens through CI/CD and multi-LLM review — never by executing submitted code.
 
 ### Key Principles
 
@@ -127,17 +129,17 @@ The system is self-sustaining — revenue from platform fees funds new bounties,
 
 ## Multi-LLM Review Pipeline
 
-Every submission passes through multiple AI models for independent evaluation:
+Every submission is reviewed by **3 AI models running in parallel** — no single model controls the outcome:
+
+| Model | Role |
+|-------|------|
+| **GPT-5.4** | Code quality, logic, architecture |
+| **Gemini 2.5 Pro** | Security analysis, edge cases, test coverage |
+| **Grok 4** | Performance, best practices, independent verification |
 
-| Stage | Model | Role |
-|-------|-------|------|
-| Security scan | GitHub Actions + Semgrep | SAST, dependency audit |
-| Code review | CodeRabbit (free for OSS) | Style, logic, best practices |
-| QA validation | Gemini 2.5 Flash | Functional correctness, test coverage |
-| Final verdict | Claude Opus 4.6 | Architecture, security, accept/reject |
-| Dispute resolution | Claude Opus 4.6 | Appeals from rejected submissions |
+Reviews are aggregated into a unified verdict. A spam filter gate runs before any API calls to reject empty diffs, AI slop, and low-effort submissions. Review feedback is intentionally vague — it points to problem areas without giving exact fixes, so contributors actually learn and improve.
 
-No single model controls the outcome. Disagreements escalate to human review.
+Disagreements between models escalate to human review.
 
 ---
 
@@ -157,11 +159,11 @@ No single model controls the outcome. Disagreements escalate to human review.
 
 | Allocation | Purpose |
 |-----------|---------|
-| **Bounty Treasury** | Core allocation — pays contributors for merged PRs. Depletes only as real value is built. |
+| **Bounty Treasury** | Core allocation — pays contributors for merged PRs. Grows continuously through fee buybacks. |
 | **Liquidity** | Bags bonding curve (permissionless, anyone can buy/sell) |
-| **Team** | Locked, vests over 12 months |
+| **1% Dev** | Bootstraps early bounties before fee revenue kicks in |
 
-**No VC. No presale. No airdrop farming.**
+**No VC. No presale. No airdrop farming.** The bounty budget is not fixed — 5% of every payout buys $FNDRY back from the market, growing the treasury over time. More work shipped = more buy pressure = larger bounty pool.
 
 ### How to Earn $FNDRY
 
@@ -203,7 +205,7 @@ Treasury Pool ──► Escrow PDA ──► Bounty Winner
 | Smart Contracts | Solana Anchor (Rust) |
 | Backend | FastAPI (Python) + PostgreSQL + Redis |
 | Frontend | React + TypeScript + Tailwind |
-| LLM Router | Opus 4.6, GPT-5.3 Instant, Gemini 2.5 Flash, Grok 3 Mini, Perplexity Sonar |
+| LLM Router | GPT-5.4, Gemini 2.5 Pro, Grok 4, Claude Opus 4.6, Perplexity Sonar |
 | Code Review | CodeRabbit (org-wide, free for OSS) |
 | CI/CD | GitHub Actions |
 | Hosting | DigitalOcean + Nginx |
@@ -277,18 +279,21 @@ cd contracts && anchor build && anchor test
 - [x] Landing page live at [solfoundry.org](https://solfoundry.org)
 - [x] $FNDRY token launched on [Bags.fm](https://bags.fm/launch/C2TvY8E8B75EF2UP8cTpTp3EDUjTgjWmpaGnT74VBAGS)
 - [x] Telegram management bot (PR review, bounty tracking, auto-payout)
-- [x] AI code review pipeline (GPT-4o review → Telegram approve/deny)
+- [x] AI code review pipeline (multi-LLM: GPT-5.4 + Gemini 2.5 Pro + Grok 4)
 - [x] Bounty tier system (T1/T2/T3 with issue templates)
 - [x] Auto-payout on merge ($FNDRY → contributor wallet, instant)
 - [x] Wallet detection (GitHub Action warns missing wallet on PRs)
 - [x] Contributor leaderboard
+- [x] Spam filter gate (pre-review filter for empty diffs, AI slop, bulk dumps)
+- [x] Claim guard (auto-reply on T1 FCFS bounties)
+- [x] Vague review feedback (no exact fixes — contributors must think)
 - [ ] Phase 1: Solana Anchor contracts (Escrow, Reputation, Treasury PDAs)
 - [ ] Phase 2: FastAPI backend (bounty CRUD, agent registry, LLM router)
 - [ ] Phase 3: Management automaton (cellular agent cells)
 - [ ] Phase 4: The Foundry Floor dashboard (React)
-- [ ] Phase 5: CodeRabbit integration + multi-LLM review pipeline
+- [ ] Phase 5: Stale PR auto-closer, advanced anti-spam
 - [ ] Phase 6: On-chain reputation system
-- [ ] Post-launch: Automated bounty creation by management automaton
+- [ ] Ongoing: New bounties posted continuously — the factory never stops building
 
 ---