Strategy 1 — Jito-Shreds Copy Trading Feasibility (Conditional GO)

Executive feasibility verdict for slv1 Strategy #1 — copy trading profitable Solana wallets via Jito ShredStream. This note is the condensed, decision-relevant layer; the source of truth is the full report (now Rev 2) at ~/coding/slv1/docs/copy-trading-feasibility.md.

Provenance

Two deep-research rounds, 2026-07-03 (same day). Round 1 (Fable): 103 agents, 21 sources, 98 claims, top 25 adversarially verified by 3-vote refute panels → 12 confirmed / 7 refuted / 6 unverified (verifier infra errors, not evidence). Round 2 (gap-filler, slimmed, Opus): 56 agents, 24 sources, 99 claims, top 12 verified by 2-vote panels → 11 confirmed / 1 refuted / 0 errors. Note the weaker round-2 quorum — a 1-1 split survives as “confirmed”; treat single-vendor findings as existence proofs, not accuracy claims.

Verdict

CONDITIONAL GO

Build a 2–4 week measurement MVP on managed shred feeds (RPC Fast / Chainstack class, ~commodity RPC-plan cost) and decide on self-measured numbers — every vendor-published quantification of the shred latency edge was refuted or (round 2) only corroborated at order-of-magnitude. The edge is real in ordering, unproven in magnitude. Build feed-agnostic (Jito can revoke any feed/reseller access at will — see Round 2 — Gap Closure). No colocation or bare metal before the Rotor/Alpenglow picture resolves.

Round 2 sharpened this (same day): desk research is exhausted — the next step is building the Week-1 measurement rig. The two surviving kill-risks (incumbent bot speed, entry-decay/capacity) are only measurable by the MVP.

The Structural Ceiling (most important fact)

Latency cannot capture the leader's price

Shreds reveal the leader’s transaction only as it enters the block being produced. The leader’s fill is already in slot N when you see it; a mirror trade lands in slot N+1 at the earliest (~400ms+), regardless of feed. Latency tech only ranks you among other copiers reacting to the same fill — it never gets you the leader’s entry. Durable edge, if any, lives in wallet selection + adverse-selection filtering — for which round 2 found only an off-the-shelf classifier (Wallet Master), no base rates or accuracy benchmarks.

Confirmed (survived 3-0 refute panels)

• Latency ladder ordering is real: shreds < Yellowstone gRPC < RPC WebSockets (400ms–30s depending on commitment: processed ~400ms, confirmed ~2–3s, finalized ~12.8s typical). Magnitude of the shred edge is NOT confirmed; best medium-confidence reading is ~100–300ms over well-placed gRPC. (Round 2 corroborated the ~120ms-avg magnitude at order-of-magnitude, 2-0, but every source sells the feed — self-measurement is still the gate. See Round 2 — Gap Closure.)
• Jito’s “lowest latency” claim is unbenchmarked marketing, actively contested by BlockRazor (claims >60% first-arriving shreds per region vs Jito’s 2.7–38.6%), LocalShred (~4ms ahead), and Shreder.
• Access is commoditized: direct ShredStream is free via Solana-pubkey approval in Jito Discord (max 2 regions); resellers (RPC Fast, Everstake, Chainstack) bundle ShredStream gRPC at no add-on cost, pre-deshredded, no Jito approval needed.
• No validator required: shredstream-proxy exposes gRPC SubscribeEntries → protobuf Entry streams; client work is roughly one bincode deserialize. Reference deshred client exists.
• ShredStream is signal-only: entries are not queryable state — no execution results, includes failed txs, pre-consensus (skips/forks possible). Must run Geyser/indexed RPC alongside for pool reserves, balances, wallet PnL, backfill.
• Wallet discovery is commoditized, no moat: Dragon (open source, GMGN-data scraping for PnL/win-rate/top-trader/copy-wallet lists), 46+ similar repos, GMGN’s own APIs/leaderboards. Anyone finds the same wallets — selection quality and filtering are the game.
• Execution costs: tip floor 1,000 lamports minimum; landed tips in practice ~6,000 lamports (25th pct) to ~0.01 SOL (99th pct); recommended 70/30 priority-fee/Jito-tip split. Live percentiles: bundles.jito.wtf/api/v1/bundles/tip_floor.
• Rate limit: 1 request/sec/IP/region default on send, no auth key; scaling via Discord-ticket-approved x-jito-auth key. Fine for an MVP, a constraint for burst mirroring. (2-0 panel)

Refuted 0-3 (do NOT build the business case on these)

All precise vendor latency magnitudes died in adversarial verification:

• ShredStream ~50–100ms earlier than Yellowstone on average (RPC Fast)
• 100–150ms ahead of competing feeds / “saves hundreds of ms” (Dysnix)
• Shreder beats Jito 97.06% of the time, p50 +3.4ms Frankfurt benchmark (Shreder)
• WebSockets lag 150–300ms; Yellowstone 50–100ms faster than RPC subs (Dysnix)
• Yellowstone “consistently” sub-second vs WebSockets “sometimes” (Chainstack)
• Dragon reliably scrapes top-100 traders per token off the shelf (Dragon repo)

Round 2 — Gap Closure

Slimmed gap-filler run, same day (2026-07-03; Opus, 4-angle / 10-source / 12×2-vote ≈ 56 agents). Net effect: verdict sharpened, not changed — desk research is now exhausted; the two surviving kill-risks (incumbent bot speed, entry-decay/capacity) are only measurable by the MVP.

Resolved favorably / neutrally:

• Alpenglow defused near-term (2-0, primary SIMD-0326 + Anza): initial Alpenglow is Votor consensus only (finality ~12.8s → ~150ms) and keeps Turbine — shred-based early visibility survives the initial rollout. Real threat re-pegged to the later, unscheduled Rotor SIMD. Detail in Alpenglow — Defused Near-Term, Threat Re-Pegged to Rotor below.
• Latency edge corroborated at order-of-magnitude (2-0, all feed-reseller sources): ~120ms avg (Chainstack), 200–500ms best-case peek ahead of Turbine (RPC Fast), raw shreds 10–50ms vs ~400ms RPC. Promotes the round-1 ~120ms figure from “unverified infra error” to corroborated — but it is table stakes, still self-measure, and does not close the ≥1-slot mirror gap.

Resolved adversely → architectural constraint:

• Jito ToS is structurally adverse (2-0, both primary legal instruments): license is personal / limited / revocable / non-exclusive / non-sublicensable / non-transferable, terminable at Jito’s sole discretion, any time, without notice; resale prohibited (that sub-claim split 1-1). Direct and reseller feed access are contractually non-durable → build feed-agnostic: put the shred source behind an interface so a revocation is a config swap, not a rebuild.

Partially resolved / still grey:

• Adversarial-wallet detection — partial: master-vs-copy classification is commercially available off the shelf (Wallet Master, from $250/mo, Geyser-based, covers all major DEXes) — existence proof only, single-vendor, accuracy unbenchmarked. Base rates of wash / fake-PnL / copy-bait wallets still unknown.
• EU regulatory — grey zone (2-0, ESMA/EBA Joint Report, Art. 142 MiCA, Jan 2025): front-running / sandwiching “typically unlawful” in TradFi, but the prohibition is not extended on-chain; MiCAR leaves “fully decentralised” undefined. No US enforcement history verified either direction. Not legal advice — get a jurisdiction-specific read before scaling real capital.

Written off / still open (cannot move the verdict — MVP-only):

• Round-1 stranded claims resolved: Chainstack $49/mo — NOT confirmed (get direct quotes); 50ms bundle-auction ticks — NOT verified from primary docs (measure landed-fill timing instead); Turbine-bypass mechanism — explanatory color only.
• Incumbent copy-bot latency / fees / shred-usage — still zero verified claims; only measurable by paper-trading their fills (MVP Wk 2).
• Entry-decay & per-wallet capacity curves — still empty; the round-2 synthesis flags this as “the single most important economic gap.”

Kill Criteria

1. Self-measured shred edge < ~50ms vs commodity gRPC → latency premise dead; pivot to wallet-selection-only on plain Geyser
2. Realized entry decay consumes modeled copy PnL on the tracked cohort → strategy dead regardless of feed speed
3. Rotor SIMD gets scheduled — that is what actually kills shred visibility (initial Alpenglow does not) → full re-evaluation. Alpenglow mainnet (late Q3/early Q4 2026) still triggers a ladder re-measurement when it lands; the no-colo/bare-metal rule stands until both are clear
4. Paper trading shows incumbent bots landing in the same slot window → no exploitable gap among copiers

MVP — 2–4 Week Measurement Rig (not production)

MVP measurement rig

graph TD
    SS["Managed shred feed<br/>(RPC Fast / Chainstack class)"] --> H["Dual-timestamp harness"]
    YG["Yellowstone gRPC<br/>(same region)"] --> H
    H --> STATS["Arrival-delta stats<br/><i>self-measured edge</i>"]
    GEY["Geyser / indexed RPC"] --> STATE["State tracker<br/>(pool reserves, balances)"]
    WD["Dragon / GMGN candidates"] --> COHORT["Wallet cohort (~20-50)<br/>own PnL indexer + wash/bait filters"]
    SS --> EXEC["Paper-mirror executor"]
    COHORT --> EXEC
    STATE --> EXEC
    EXEC --> DECAY["Realized entry-decay log<br/>tips via tip_floor percentiles"]
    style SS fill:#264653,stroke:#2a9d8f,color:#fff
    style YG fill:#264653,stroke:#2a9d8f,color:#fff
    style EXEC fill:#3d2020,stroke:#e76f51,color:#fff
    style STATS fill:#2d2d2d,stroke:#888,color:#fff
    style DECAY fill:#2d2d2d,stroke:#888,color:#fff

• Wk 1: dual-consume ShredStream entries + Yellowstone from one region; measure own shred-vs-gRPC arrival distribution per venue (pump.fun, Raydium, PumpSwap)
• Wk 2: wallet cohort (~20–50) via commodity tooling + own PnL indexing; paper-mirror; log leader-fill → earliest-landable-slot price decay per trade
• Wk 3: economics model on measured data (decay + tip percentiles + priority fees vs modeled copy PnL); per-wallet capacity estimate
• Wk 4: kill-criteria review; only if passed, live micro-size test within the 1 rps default limit

Open Gaps — MVP-only or time-gated (desk research exhausted after round 2)

1. Incumbent copy-bot speed (GMGN, Photon, BullX, Axiom, Banana Gun, Maestro, Trojan, Nova, Vector) and MEV firms — is shred-fed copying already table stakes? Zero verified claims; only paper-trading their fills answers it (MVP Wk 2)
2. Capacity & decay: entry-slippage decay curves 1–3 slots behind leader fills; per-wallet SOL capacity in illiquid meme pools — the single most important economic gap (MVP Wk 1–2)
3. Rug/honeypot base rates; colo/bare-metal economics and block-engine regions — deferred until the colo embargo lifts

Closed / re-classified in round 2 (see Round 2 — Gap Closure): Legal/ToS → resolved, structurally adverse (build feed-agnostic); adversarial-wallet detection → partial (Wallet Master exists, base rates still unknown); EU regulatory → grey zone, no enforcement.

Alpenglow — Defused Near-Term, Threat Re-Pegged to Rotor

Round 2 update (2-0, primary SIMD-0326 + Anza)

Initial Alpenglow replaces consensus only (TowerBFT/PoH → Votor; finality ~12.8s → ~150ms) and does not touch shred production — shred-based early visibility survives the initial rollout. Verbatim from the SIMD: “Initially we stay with Turbine as the data dissemination protocol. Rotor will be introduced later and will get its own SIMD.”

• Status: governance passed ~98.3% (Sept 2025); on Anza’s community test cluster since 2026-05-11; mainnet targeted late Q3 / early Q4 2026 via Agave 4.1 (already slipped from Q1).
• Real threat is the later, currently unscheduled Rotor SIMD — that is the kill trigger (kill criterion #3), not initial Alpenglow.
• Secondary effect to re-measure at activation: ~150ms finality compresses the shred edge over confirmed/finalized-commitment feeds (the slow end of the ladder), not over processed-level gRPC.
• No long-term / colo infra commitments before both Rotor scheduling and Alpenglow mainnet are clear.

Related

• slv1 — project overview (Strategy #2, market-intel sniper, is queued next)
• Full report (Rev 2, source of truth): ~/coding/slv1/docs/copy-trading-feasibility.md
• Primary sources (round 1): Jito ShredStream docs · Jito send/tips docs · shredstream-proxy · RPC Fast ShredStream gRPC docs · Dragon
• Primary sources (round 2): SIMD-0326 (Alpenglow) · Anza Alpenglow announcement · Jito Terms of Use · Jito legal · ESMA/EBA Joint Report (Art. 142 MiCA, Jan 2025) · Wallet Master (copy-trading classification API)