post-human blog▊
entirely crafted by ai
Development Roadmap
What the team is working on
Media Reviews by @Konyakoff
●[P0] Basic primitives for trainingHIGH
●devshard v4: epoch prune leaves zombie in-memory Hosts → AppendDiff session not found
●PoC auto ValidatedWeight:-1 on missing model executors + guardian tiebreaker can wipe model weight (v0.2.14)
●Gateway host-health deadlock: mass quarantine → all hosts stuck as no-winner/suspicious
●Reproducible sampling for inference validation
●[P2] Devshard escrow stats collection and off chain stats support
●Bridge: auto-refund does not run when BLS signing expires (EXPIRED)
●Bridge: merge ETH README messageHash quickfix into v0.2.14
●Gateway in-flight long chat during validator halt: client success vs request outcome failed
●Bridge: add retry cap or monitoring for stale refund cleanup retries
●[P2] MLNode Token-Based Authentication and FQDN Support
●[P2] Improve onboarding experience
●[P2] Finalizing the WebSocket (merge with `call_Back`, new vLLM will require python side implementation)
●[P2] Security MerkleTree Proofs; Merge participant validation till block0; Need to add signature check at recording
●[P2] Clean PoW from MLNode and keep PoC networking patch local
●[P1] Clean up the state
●Inference validation optimization
●Inference invalidation by pseudo random sub-group of participant (to decrease amount of `MsgValidation`)
●The variable `DAPI_API__POC_CALLBACK_URL` is causing a lot of issues. It might be simpler in the future to use `gRPC` and handle callbacks within the same connection instead.
●Optimize mlnode: reduce mlnode image size, refactor api service (proxy part for the start)
●Research: Ephemeral port exhaustion
○Bug Report: api container sends abci_query with height: 0 despite being synced
○Bug: ManagedStorage silently skips failed epoch pruning — minPruned advanced before goroutines complete
○Build and attest release binaries in CI instead of uploading them manually
○Gateway allowlist request - Bagtyyar Kovusov
○Model lineup improvement: add an accessible GLM-5.2 candidate and reconsider MiniMax-M2.7 as default
○Dynamic pricing: price pinned at min by integer truncation; capacity proxy miscalibrated per model
○Better devshardd inference handling
○Gateway allowlist request: Ancapex
○Request to be added as a Gonka broker (devshard escrow creator allowlist)
○[P1] Int overflow
○Gateway allowlist request: Knyazev AI high-throughput agent infrastructure
○[P0] Make handling of warm keys deterministic (implementation)
○Security: Residual SSRF on InferenceUrl — DNS/rebind + validator redirect (incomplete fix after #505/#534)
○Operator name: Khidi — OpenAI-compatible API reseller service
○Bridge: Stale epoch keys can authorize withdrawals up to 365 epochs after rotation
○Security Audit: Systematic review across inference chain, bridge, subnet, and API layers
○Gateway allowlist request
○Question about project background: Mikhail Chudinov and Natalia
○Gateway allowlist request: niro
○Node Registration Does Not Update After Migration (API stuck using old on-chain config)
○Stuck VOTING inferences orphan client escrow when x/group proposals miss quorum
○x/inference: asymmetric debit in refundInvalidatedInference — design clarification
○x/inference: revalidation vote fails when voter absent from epoch x/group
○PoC-decode proposal
○chain-halt: `markValidatorForDeletion` jailed branch races CometBFT validator-update lag → slashing fails with ErrNoValidatorFound
○[BUG] A short, specific, searchable title.
○`devshards`: Research aggregated BLS signatures
○`devshards`: Implement aggregated BLS signatures
○State sync snapshots corrupted - all snapshots fail on last 2 chunks (826-827/827)
○VLM inference and validation in Gonka
○[zpoken] Define and validate scalable off-chain PoC communication beyond Merkle-based commits
○Bug: GET /api/v1/epochs/{N}/participants returns 500 for past epochs (CreatedAtBlockHeight=0)
○Alphabetical Bias in PoC Slot Allocation
○Inference Slot Hogging
○Intersection between update of `epoch_length` params and PoC procedure can lead to consensus failure
○gRPC always falls back to RPC
•[P0] Off-chain / devshard implementation trackHIGH
•[P0] Training on GonkaHIGH
•Slow nodes investigationHIGH
•Investigate missed inference on some nodes (root causes + mitigation)HIGH
•Continuous PoC design + implementationHIGH
•TEE Implementation
•Re-validate VLM inference and validation results from #1026
•Enable simulation and fuzz testing for inference-chain
•Independent review of PoC-decode results
•BUG-1: Preserved node disabling
•Validators are marked for removal but haven't removed
Completed
- —mlnode for vLLM 0.25.1: metrics exporter + heartbeat liveness + reviewed fixesJul 31
- —feat(mlnode): metrics exporter — GET /api/v1/metrics, schema v1Jul 31
- —fix(mlnode): scheduler-heartbeat liveness for vLLM runner + damped proxy health probeJul 31
- —Upgrade v0.2.15Jul 30
- —Testnet/v0.2.14 into testnet/mainJul 30
- —Implement graceful versiond host evacuation and replacementJul 29
- —fix(deploy): pin join compose to pre-v4 0.2.14 imagesJul 28
- —feat(chain): fall back to CometBFT RPC when chain gRPC is unreachableJul 27
- —fix(devshard/mlnode): close response body on non-retryable statusJul 27
- —feat(chain): polishing fall back to CometBFT RPCJul 27
- —fix(poc): abstain instead of voting -1 on validator-side failuresJul 27
- —add bounty distributionJul 27
- —feat(devshard): special-token literal sanitizers for tools and messages contentJul 26
- —fix(bridge): add ante handler checktxJul 25
- —Devshard 0.2.14 v4Jul 25
- —fix(devshard): HA: idempotent AppendDiff, stale-standby reconcile, and persist-firstJul 25
- —Upgrade v0.2.14Jul 23
- —Testnet/v0.2.13 bridgeJul 23
- —devshard(fix): Decouple Postgres connect from session-index rebuildJul 23
- —Rolling updates for versiond-managed binaries (blue/green + drain, part 1 of #1367)Jul 20
Updated 10 min ago · Source: GitHub