Transit Fabric releases

Red-team pack

Everything you need to look for ways to earn more than you carry on Transit Fabric (Bittensor testnet, subnet 611), and to show them working. The pack was taken from the source repository at commit 91be12e.

Read the bounty page first: it has the scope, the out-of-scope list, the tiers and the rules. Register once on the claim page so credit goes to your address.

What is in the pack

What Why you want it
Scoring spec (.md) How emission shares are computed, with every parameter and its default. The place to look for something worth gaming.
Measurement spec (.md) What validators probe and test, and how results are pooled.
Receipts spec (.md) How bytes are credited, and what a relay can and can't claim.
Capacity bonds spec (.md) The bonds that back declared capacity.
Identity spec (.md) Node keys, hotkeys, bindings and consents.
Encoding spec (.md) Canonical encoding and domain-separated signatures.
Threat model (.md) Every attack we have written down, its mitigation and its status. Read it before you file.
Whitepaper (.md, PDF) The design in one document.
The simulator bwos-sim, static binaries for Linux x86_64 and aarch64, with checksums and sample output. A walk-through of reproducing a finding in it.
Finding template The format we want findings in. The "Finding" Issue form has the same fields.

The simulator binaries are assets of the GitHub release redteam-pack-91be12e, a pre-release. Download them and check the SHA256SUMS file before you run anything.

What is in scope

The scope, the out-of-scope items and the reward tiers are on the bounty page. In short: the released software, the incentive mechanism as it runs on subnet 611, and the design documents. Testnet only. Never use mainnet keys or real TAO.

Open and accepted threat-model items

The threat model lists attacks we already know about. Items marked open or accepted are known: they earn credit only if you show a new exploit path or a bigger payoff. Items marked mitigated are fair game: if you break a mitigation, that is a finding. There are 66 open and 5 accepted items in the threat model at this commit.

Item Status What
T38 open Nothing fills ProbeResultBatch.endpoints in the daemon yet, so in production no node has an underlay
T39 open Nothing fills ProbeResultBatch.endpoints in the daemon yet, so in production no node has an underlay
T41 open Rented burst capacity
T42 open Team-test manipulation
T46 open SRG griefing
T47 accepted Inherited carrier diversity
T58 accepted Signed-input capture
T60 open Equivocating quorum
T62 open Stale evidence re-signed
T67 open Sample favouring and blame by omission
T68 open Forged digests and spoofed sources
T69 open False reports
T70 open The reveal names no egress
T71 open Omission and withholding blame the hop behind
T72 open The egress knows every key
T73 open The two ends have a veto
T74 open Determinism
T78 open Another process of the same user reaches the agent's control socket or its keychain entry and acts a
T79 open The owner is tricked into importing a hotkey, or a coldkey, into the agent
T80 open The binding request is altered between export and signing, binding the node key to another hotkey
T82 open Underlay ambiguity grief
T87 open Anchoring jump
T89 accepted State tampering
T90 open Forged agreed underlay
T94 open A node gives its probe sessions priority over real traffic, and looks better to vantage
T105 open Test target not bound to the endpoint
T107 open Cohort ceilings and chunk steering
T113 open (a) A late-queue node started at its last chance with three on 31-62% of tests, so T42's one-collude
T116 open (a) Cohort credit is kept for 8 history entries, but a cohort falls due about once per 2.5 windows,
T123 open Restore crash loop
T125 open A co-located group's pool is the max over its chunk cohorts of each cohort's lower median of few sam
T126 open (a) Rider keys split across many UIDs get one front cohort turn per UID, charged to the fleet too
T130 open Latch split brain
T131 open Hop receipts from real counters and background from receipt runs
T135 open Same-block clone griefing
T140 open Perturbed production leech
T143 open Damping shadows a node from validators
T144 open Damage spreads past the flapping link
T145 open Public canary pairs
T146 open A validator's plan is secret from nodes, not from the validator
T147 open Splitting above the canary weight cap
T148 open Canary packet shape
T153 open Mechanism 0 credits transit on current_prod_table, not the table as updated
T157 open Validators holding more than a third of stake can withhold attestations and stop updates
T200 open The participation controller ratchets
T201 open Quality drag on the controller
T202 open A teammate's zero report
T203 accepted The own-fleet rule avoided with a second coldkey
T204 open k depends on the full history
T205 open Too few independent recorders
T165 open Peer-load collusion
T167 open Hidden slices identified
T169 open Peer pool and duty
T170 open usable over-credits
T171 open Long burst credits
T172 open The coordinator leaks
T173 open Flood framing
T217 open Hidden-slice tells and relay drops
T187 open Validator site collusion
T189 open Unusable capacity paid at a scarce price
T190 open Herding: every operator moves to the same scarce region at once, on the same price map, and prices o
T193 open Burst-rented capacity in a scarce region
T195 open Atlas anchor forgery
T197 open Carry abuse
T221 open Version split
T226 open ASN grinding and churn by recommit
T257 open Placement steering
T259 open Allocation starvation
T230 open Selective serving
T264 open A validator feeds a solver a forged snapshot
T300 accepted Unmatched probe sessions feed batches

How to file

  1. Reproduce it against a named version. Points need a reproduction.
  2. Check the threat model above and known limits. The first valid report wins, and duplicates are closed with no points.
  3. Use the finding template. A bug goes in through the "Finding" Issue form. A security finding or exploit goes in privately, through the advisory or its API. Large artifacts go in as a zip attachment.