Common issues when submitting and iterating on solver code.

Submission endpoint errors

Error: cannot reach /v1/submissions*

Typical causes:

• Wrong base URL (using validator :9100 when submissions are served by API :8080 in local testnet)
• API service not running

Checks:

curl http://localhost:8080/health
curl http://localhost:8080/v1/submissions

No open solver round

Error: Cannot submit: solver round unavailable / Cannot submit: no open solver round

submit discovers the active round via GET /v1/solver/round and requires one that is open and accepting submissions — there is no epoch fallback. Check the round:

curl http://localhost:8080/v1/solver/round

If there’s no round_id, or accepting_submissions is false, wait for the next open round (on local testnet, ensure the API has opened one). --epoch is no longer required — the epoch is read from the round.

Signature or hotkey issues (git submission)

Error: HTTP 400/401 on POST /v1/submissions

Checks:

• --hotkey points to a real Bittensor wallet name
• wallet exists under BT_WALLET_PATH (or default ~/.bittensor/wallets)
• commit hash and repo URL match the signed message payload

Screening failures

For git submissions, failures often come from:

• Missing required files in repo root (Dockerfile, solver.py, README.md)
• Invalid Dockerfile base image
• CMD/ENTRYPOINT present in Dockerfile
• SOLVER_CLASS import/init issues

Local checks:

docker build --network=none --memory=4g -t test-solver .
docker run --rm --network=none --read-only --tmpfs=/tmp:size=64m --memory=2g --cpus=1.0 --entrypoint python test-solver -c "from solver import SOLVER_CLASS; print(SOLVER_CLASS.__name__)"

Submission not adopted

The inline source path (/v1/submissions/source) was removed (PR #599). All submissions now go through the git PR path.

Check the outcome_code on your submission status first — it names the exact reason. Common non-adoption cases:

• Tied on output. You matched the champion on every order (within the ±0.1% band) and lost or tied every tie-break rung — the incumbent is kept. To win an all-matched tie you must be cheaper on total metered gas (≥200 bps), better factored (max_region_nodes smaller by ≥100), or carry less dead code (unproductive_nodes smaller by ≥2000). Your report names the target.
• Not net-better on output. Your regressions were not outnumbered by wins: adoption needs (wins + blind_spot_covers) − regressions ≥ 1.
• Hard veto. You cut at least one order by more than 1% (n_catastrophic), or you dropped an order the champion serves (n_dropped) — either vetoes adoption regardless of how many other orders you won.
• Screening reject (too_entangled / dynamic_code). Stage 1 now rejects a solver whose largest AST region exceeds 4,200 nodes (too_entangled) or that uses bare exec()/eval() (dynamic_code). The reject still reports your max_region_nodes so you can see the number to get under.
• waitlisted, not rejected. If your outcome_code is rotation_not_selected or window_elapsed, you were not benchmarked this round through no fault of your own — you keep a next-round priority. Resubmitting an identical solver does not help (see below).
• fingerprint_repeat. A comment-only / nonce-only resubmit of an identical code tree is rejected pre-build — “comments don’t make code new”. Change at least one semantic byte. (Cross-hotkey resubmit quota is operator-gated and off by default.)

Use status endpoint to inspect:

python -m minotaur_subnet.miner.main status \
  --submission-id <id> \
  --validator-url http://localhost:8080

Score is always zero

Common causes:

• generate_plan() raises exceptions
• malformed ExecutionPlan / calldata
• invalid addresses, deadlines, or empty interaction list

Quick check: submit through the git PR path and review the benchmark/status details on your submission (python -m minotaur_subnet.miner.main status --submission-id <id>), or reproduce locally against the testnet, which runs the same screening → benchmark → scoring pipeline.

Agent loop does not generate submissions

Checks:

• Claude CLI is installed and available in PATH
• API URL points to a reachable server exposing /v1/apps/manifests and /v1/submissions
• strategy directory is writable

Run with explicit options:

python -m minotaur_subnet.miner.main agent \
  --validator-url http://localhost:8080 \
  --strategy-dir ./strategies \
  --loop-interval 30

Useful commands

# API health
curl http://localhost:8080/health

# List submissions
curl http://localhost:8080/v1/submissions

# Poll one submission
python -m minotaur_subnet.miner.main status --submission-id <id> --validator-url http://localhost:8080

See also: Quickstart, Configuration, Custom Solver.