Quant Agent — multi-agent quant terminal on Hyperliquid

Live HL data → in-browser TA → 4-agent committee → LONG/SHORT/SKIP verdict.
Terminal grid command bar + interactive agent-workflow graph. Paper-trading
worker reuses the browser brain (dry-run + Telegram). Forgejo static deploy.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs/token-strategy.md

@@ -0,0 +1,95 @@
+# Quant Agent ($QUANT) — token utility & engagement strategy
+
+The honest, shippable answer to: **"what does a user get?"** and **"what keeps them clicking?"**
+Built only on what the product really is — a transparent, paper-trading multi-agent terminal on
+Hyperliquid (pre-TGE). Vetted by an adversarial honesty/legal pass.
+
+---
+
+## 1. What's the profit for the user? (the one-liner)
+
+> **Hold $QUANT to see the whole swarm in real time and steer what it watches — access and
+> influence, never a promised return.**
+
+Free users see a **delayed teaser** (one coin, lagged verdict). Holders unlock the **full live
+committee** — every coin, every timeframe, real-time verdict flips to their Telegram, and a vote
+on what the agents watch next. You're selling **access + influence + speed**, not a payout.
+
+This is the safe, defensible thesis. The moment the pitch becomes "hold and earn" or "hold for
+profitable signals", it becomes a securities + advice problem (see §4).
+
+---
+
+## 2. Token utility — the 5 mechanics to build (in order)
+
+1. **Hold-to-access tiers (Free vs Holder).** Connect a Solana wallet, read $QUANT balance
+   client-side, gate UI depth. Free = 1 coin + delayed verdict; Holder = all coins, live verdict,
+   full agent vote-matrix + reasons. *Two tiers only.* This is the spine — everything hangs off it.
+2. **Holder alerts (real-time vs delayed).** The worker already detects verdict flips and Telegram
+   is already wired. Map wallet→chat_id (one-time sign), DM holders instantly on "committee flipped"
+   / "high-conviction (>0.9)". Free = delayed digest. *Lowest effort, highest perceived value.*
+3. **Premium signal surface.** The brain is parameterized (`CONFIG.coins`/`interval`). Run more
+   coins + timeframes (15m/4h, top-20 perps); expose via the worker's `/api/state.json`. Breadth is
+   genuinely compute-scarce, so gating it is honest.
+4. **Free prediction game (no staking).** Call LONG/SHORT/SKIP before the next candle, scored vs the
+   real price move (we already track prices + equity). Cosmetic ranks only. *Retention engine for
+   non-holders → seeds future holders.*
+5. **Governance over the agents/markets.** Balance-weighted snapshot votes on real knobs the config
+   exposes — which coin/timeframe to add next, agent weights, the confidence floor. Real because the
+   worker loads the result. *Strongest non-financial reason to hold: influence over a live product.*
+
+(#6 once liability copy is locked: a **"Open this setup on Hyperliquid"** deep-link that prefills
+coin/side/size/TP/SL from the verdict — convenience, the user signs their own trade. Never
+auto-execute.)
+
+---
+
+## 3. On-site engagement — features that make people click & return
+
+Ranked by leverage. Most reuse data we already compute.
+
+| # | Feature | Why it drives clicks | Effort |
+|---|---|---|---|
+| 1 | **Public agent track record** (honest paper PnL: equity curve, win rate, last fills) | The #1 "come back" hook — a live scoreboard of how the committee is actually doing. We already persist `equity/closed/equityCurve` in the worker. | M |
+| 2 | **Predict-the-verdict game** + seasonal leaderboard | Daily reason to return: "can you read the agents better than the swarm?" | M–L |
+| 3 | **Shareable verdict cards** (X/Twitter image) | Free top-of-funnel — every share is an ad with the real signal. | M |
+| 4 | **Tune the committee** (weight the 4 agents, re-run live) | Hands-on play → ownership; explore every coin/timeframe. | M |
+| 5 | **Multi-coin signal board** (whole universe at a glance) | A reason to scan & click instead of staring at one coin. | M |
+| 6 | **Watchlist + price/verdict alerts** | Classic retention — pulls the tab back. | M |
+| 7 | **Backtest / replay slider** on the chart | Explorable toy — drag across coins/timeframes. | L |
+| 8 | **Live committee ticker / "debate feed"** | Makes the desk feel alive, never static. | S |
+| 9 | **Streaks & daily check-in** (cosmetic points) | Connective tissue that turns the above into a daily habit. | M |
+
+---
+
+## 4. Red lines — what NOT to do (honesty + legal)
+
+The trust of an honest, paper-trading, pre-TGE product is the moat. Avoid:
+
+- ❌ **Fee-split / revenue-share to holders** (the $AUTR "50% to holders" model) — dividend-like,
+  strongly implies a security. Differentiate with **access/utility/status, no holder cash flows**.
+- ❌ **Points → guaranteed airdrop** framing — "earn now, tokens later" = expectation of profit.
+  Keep points as **pure status** (rank/badges/cosmetics); never pre-promise token value.
+- ❌ **Stake-to-earn / stake-for-rewards** — yield/securities risk. Staking may unlock only
+  cosmetics/role, never payouts.
+- ❌ **"Hold for profitable signals" / any return promise.** Sell access to *data/views*, not outcomes.
+- ❌ **Auto-executing trades** on the user's behalf — custody + advice exposure. Deep-link only; the
+  user signs.
+- ❌ **Fake track record / paid-pump signals.** Always label paper/dry-run.
+
+Everything in §2–§3 is designed to stay on the right side of these lines.
+
+---
+
+## 5. Pre-TGE acquisition (you have no token yet — use it)
+
+- **Points/season program as pure status now** (no airdrop promise) → daily habit before the token.
+- **Public track record + shareable cards** → organic reach proving the agents are real.
+- **Telegram alpha bot** (already built) with a waitlist → capture users where crypto lives.
+- **Named agent personalities** (Trend / Breakout / Mean-Rev / Funding as a "cast") → memorable,
+  screenshot-able, turns dry TA into a story.
+
+---
+
+*Strategy synthesised from a multi-agent pass (utility models · engagement · reference meta ·
+adversarial honesty critic). Not legal advice — have token mechanics + copy reviewed before TGE.*