☿ Kairos War Room
← Back to master state
Gated source note · private

infra-assessment

Cold, unsentimental valuation of Mysterium's infrastructure: is it the moat, or dead weight? Companion to wave-map, market-map, Kairos — Research Dossier. Confidence marks: [confirmed] primary-sourced · [claimed] self-report unverified · [inference] reasoning. Frame set by Lee: the only hard bound is MN's core values (provisional: privacy, open access / censorship-resistance, decentralization / user-sovereignty, open-source, no user-surveillance — pending Ro's precise wording). Everything else, including the infra, is instrumental — leverage it where it's a real advantage, shed it without sentiment where it isn't.

Live network stats (corrects the dossier)

32,570 nodes · 182 countries · 3.21 PB transferred in the last 30 days [confirmed — MystNodes World Network Health dashboard, tooltip confirms 30-day window]. MYST ~$0.09. ⚠ CONFLICT (2026-07-18): the mystnodes.com landing page, fetched same week, shows 31,547 IPs / 135 countries — likely different counting bases (registered vs active; countries-ever vs active). Do NOT cite "182 countries" as fact until the basis is resolved; it is load-bearing for the Bucket-1 breadth claim (scanner W1). Details: public-surface-sweep-2026-07-18. The network is not idle — it actively moves petabytes monthly.

Bucket 1 — Where the infra genuinely IS an advantage

Test: does the use-case reward scale (MN loses, ~77× below Grass's 2.5M) or some other property MN has?

None of these is "raw residential bandwidth." That's exactly where MN has no edge.

Bucket 2 — Where the infra could be REPURPOSED

Scored on fit to MN's real properties (breadth ✓, churn ✗, light-compute ✗, rail ✓):

Bucket 3 — Where clinging weighs MN down (anti-sentiment)

The split verdict (the physics answer)

Not keep-vs-drop. A split:

Keep three properties — 182-country breadth, the permissionless settlement rail, the on-chain provenance capability. Shed two things — the raw-bandwidth volume ambition and the "decentralized VPN" identity.

Winning moves live where Bucket 1 ∩ Bucket 2 overlap: measurement/verification (breadth) and agent-payable access (rail), provenance as the high-ceiling stretch. All three kept-properties express the provisional core values; the two shed-things don't. Passes constraint, not just objective.

Value-capture math (the 3.21 PB reframe)

3.21 PB/30d = ~3,210,000 GB/month of real flowing traffic [confirmed].

Payout-vs-traffic cross-check (important): $568k paid to runners cumulatively against 3.21 PB/month current traffic. Timeframes don't divide cleanly (lifetime vs monthly), but illustratively: one year at current volume ≈ 38 PB ≈ 38.5M GB; $568k across even that single year ≈ ~1.5¢/GB, and across the multi-year life, fractions of a cent — vs. $1.40+/GB proxy retail and ~$0.20/GB even for passive bandwidth-sharing (Honeygain-style). That gap implies the traffic is overwhelmingly low-value (people using cheap/free dVPN, not premium proxy egress) — if a meaningful slice were premium, payouts would be orders of magnitude higher.

Conditional reframe (now qualified): IF burn ≈ €2.4M/yr (~$217k/mo — my 40-people estimate, unconfirmed), break-even needs ~$0.07/GB net on existing volume (~5% of retail). BUT the cross-check says most of that volume isn't premium-priceable — so it leans more a demand problem (need premium buyers) than a pure value-capture problem. Downgrade the easy-break-even read.

⚠ UPDATE (2026-08-13, from the 08-12 burn-canon repair): the conditional is now partly resolved. Šaras verbatim ("revenue is 2M, loss is 2M") fixes the cost base at ~€4M/yr (~$360k/mo) — this note's €2.4M estimate was closer than the then-canon €2M but still under by ~40%; the per-head read lands near €100k/head for ~40 people, which the 07-18 Fresh-Eyes audit had already flagged as the plausible level. The break-even gap is the net loss: ~€2M/yr ≈ ~$180k/mo, inside the "$150–250k/mo" bracket the encouraging-math section below already uses — so that section's conclusions (capacity dwarfs break-even need; Phase-1 is a proxy-demand/sales problem) survive unchanged. (Canon: CONTEXT/CURRENT.md push 3f550e9, Kairos — MASTER STATE.md.)

The decisive swing variable — and it's GoProxies-gated: what fraction of 3.21 PB is premium-monetizable egress vs. low-value dVPN, and MN's capture rate on it. The payout cross-check leans small — UNLESS GoProxies (Lee's 98%: in-house) books its B2B proxy volume + payouts separately, in which case $568k is only the consumer/dVPN side and the premium revenue hides off-dashboard in GoProxies' books. So interpreting both the 3.21 PB and the $568k depends on whether GoProxies is inside or outside these figures — the same master question.

Capacity is not the constraint — demand is (2026-07-17)

The network already carries 3.21 PB/month (proven), with likely idle headroom beyond it (home connections sit mostly idle), and nodes can serve proxy traffic alongside VPN — no cannibalization. So supply/capacity is abundant and NOT the bottleneck. The 3.21 PB is a measure of current VPN demand, not a "redirectable" reservoir — you can't point VPN customers' video streams at proxy buyers; you fill spare node capacity with new proxy demand.

The encouraging math: break-even (~$150–250k/mo) at ~$1/GB proxy = ~150–250k GB/month = only ~5–8% of current throughput-equivalent — capacity dwarfs the break-even need ~12–20×. MN's network can physically support a $30M+/yr proxy business it just can't sell. So break-even means growing proxy demand ~$8k → ~$200k/mo (~25×, hard but not moonshot-scale). Phase-1 survival is entirely a proxy-demand / sales problem, not a capacity one. MN does have a proxy sales motion — GoProxies is MN's own proxy brand + sales team — but it's small (~$8k/mo, growing); Phase 1 = scale it, not build from zero. The huge proven capacity is a sales story ("we scale to your volume instantly"), not itself revenue.

Two caveats that can flip conclusions

GoProxies-in-house caveat → RESOLVED (2026-07-17): confirmed IN-HOUSE (Lee, definitive). GoProxies is MN's own proxy product/brand (own sales team), not an arm's-length partner. Consequences now hardened: Bucket 3b's "feeding someone else's margin" collapses — there is no external partner; MN captures 100% of proxy revenue. All ~$75.6k/mo revenue is in-house. The market-map Workstream-1 "arm's-length partner" verdict is superseded (corrected there). Phase-1 proxy lever = scale the existing GoProxies sales motion, not build from zero.

Agent-payments reality check (the Hermes rail, honestly). The machine-payment market is real and exploding but owned by giants — Coinbase x402 (launched May 2025; ~165M txns / 69k agents / ~$50M cumulative by Apr 2026, ~half testing), Google AP2, Lightning Labs L402, Mastercard/Visa. Two hard problems for Hermes-as-product: (1) it's domain-specific plumbing for paying MN's own nodes, not a general agent-payment protocol — competing with x402 = rebuild, against Coinbase; (2) it settles in MYST, but the market standardized on stablecoins — wrong denomination. So the rail is not a hidden payments goldmine. The real move is adopt x402/L402 + stablecoin settlement on top of MN's access, so agents can pay the network directly for egress — ride the rails, don't build/sell them. Transferable asset = 4.5yr operational know-how (a head start, not a market position). [market facts confirmed; MN's ability to bolt this on = unverified engineering question]

Validation gates (before anyone bets on this)

  1. Node persistence / churn quality — kills or confirms every breadth-based (measurement) use-case.
  2. Does the rail have any paying demand today — and does MN even know x402/AP2/L402 exist / has anyone considered agent-payable access.

Open questions → canonical file

All insider-unknowns and investigate-topics now live in questions-to-excavate (single source). The two that gate everything: "is GoProxies us?" (master key) and the premium-traffic ratio (must-know).

Browser rendering of the War Room source library. The vault remains canonical.