TASK_SUBMIT, TASK_RESPONSE, TASK_PROGRESS, TASK_VALIDATE, TASK_REWARD

The five-stage AI task pipeline.

USER                 MINER                  VALIDATOR              CHAIN
[TASK_SUBMIT] ─────► (poll) ──► run LLM ──► [TASK_RESPONSE] ─────► validate ──► [TASK_VALIDATE]
                                                                                       │
                                                                                       ▼
                                                                              [TASK_REWARD]

A miner may also emit TASK_PROGRESS mid-inference for long jobs (informational only).

TASK_SUBMIT

Field	Type	Notes
type	"TASK_SUBMIT"	—
from	string	Submitter
taskId	string	Caller-generated unique ID (typically 32 hex from randomBytes(16))
taskType	string	One of: chat, code_simple, code_agentic, image, audio, video, embedding, nft_svg, challenge
prompt	string	User input
fee	string (bigint)	Micro-OMBRA, must be >= MIN_TASK_FEE = 10_000
timeoutMs	number	Auto-computed from TASK_TIMEOUTS[taskType]
nonce, timestamp	—	—

Validation:

• taskType is a known value
• fee >= MIN_TASK_FEE
• state.balance(from) >= fee
• taskId not already used in state

Effect:

• state.balance(from) -= fee (escrow)
• state.tasks[taskId] = { status: "submitted", submitter: from, fee, prompt, taskType, timeoutMs, ... }

TASK_RESPONSE

Field	Type	Notes
type	"TASK_RESPONSE"	—
from	string	Miner wallet
taskId	string	Refers to existing TASK_SUBMIT
minerId	string	Miner NFT id (sha256 of MINER_REGISTER hash)
contentHash	string	sha256(content) — for verification
content	string	Inference output
thinking?	string	Optional chain-of-thought (for reasoning models)
model?	string	Model identifier (e.g., "llama-server:Qwen2.5-7B")
tokensIn?	number	Input tokens (informational)
tokensOut?	number	Output tokens (used for PoATU reward)
thinkingTokens?	number	Reasoning tokens (used for PoATU)
durationMs?	number	Inference duration

Validation:

• Referenced task exists and status == "submitted"
• Submitter is a registered miner (MINER_REGISTER tx exists for from)
• contentHash == sha256(content)
• Tx submitted within submittedAt + timeoutMs

Effect:

• Append to state.taskResponses[taskId]
• If first response, set state.tasks[taskId].firstResponseAt

TASK_PROGRESS

Field	Type	Notes
type	"TASK_PROGRESS"	—
from	string	Miner wallet
taskId	string	—
minerId	string	—
message	string	Free-form progress note

Effect: Append to state.taskProgress[taskId]. Pure informational, no balance changes.

TASK_VALIDATE

Field	Type	Notes
type	"TASK_VALIDATE"	—
from	string	Validator (registered miner with judge capability)
taskId	string	—
validatorAddress	string	Same as from
scores	array	[{ minerId, score (0-100), reasoning }]
bestMinerId	string	The winning response's minerId
validatorModel?	string	Judge model used
judgeRawOutput?	string	Raw LLM scoring response

Effect:

• state.tasks[taskId].status = "validated"
• Stores bestMinerId, scores for downstream TASK_REWARD

TASK_REWARD

Field	Type	Notes
type	"TASK_REWARD"	—
from	string	Validator (issuing reward)
taskId	string	—
rewards	array	[{ address, amount (string bigint), reason }] where reason ∈ {"miner", "validator", "burn"}

Validation:

• Task is in "validated" status
• Sum of rewards matches task.fee (modulo rounding)
• Reward distribution matches FEE_SPLIT_V2 (70/25/5)

Effect:

• For each reward: credit address, or burn (no recipient)
• state.tasks[taskId].status = "rewarded"

Builders

sdk.tasks.submitTask({ taskType, prompt, fee, taskId? })          // high-level: build+sign+broadcast
buildTaskSubmitTx(from, taskId, taskType, prompt, fee, nonce, privKey)
buildTaskResponseTx(from, taskId, minerId, content, nonce, privKey, opts)
buildTaskProgressTx(from, taskId, minerId, message, nonce, privKey)
buildTaskValidateTx(from, taskId, validatorAddress, scores, bestMinerId, nonce, privKey, opts)
buildTaskRewardTx(from, taskId, rewards, nonce, privKey)

Next: MINER_REGISTER, NFT_SETNAME, NFT_TRANSFER →