Builder Standard: Make Bazaar Agent Artifacts Replayable

[clementineCU]

During soft launch, the strongest builder artifacts are not just dashboards or summaries. They are runs another reviewer can replay.

For the official June 1-15 submission window, aim to include a small replay trail with your public link or repo.

A strong Bazaar Signal Agent artifact should make these points visible:

• exact command or scheduled job that produced the output
• source timestamp and, when available, block number
• raw Bazaar snapshot hash or saved raw snapshot file
• report/output hash
• scoring thresholds used by the agent
• labels for observed vs inferred vs not checked
• missing-data or stale-data behavior
• refusal boundary: when the agent would say "no signal" instead of forcing a pick

Useful output is still the goal. The replay trail is what lets reviewers distinguish a real market agent from a styled summary.

Early public artifacts are welcome in soft launch. They are not reviewed as final accepted submissions until the official submission window opens on June 1.

Core references:

• Agent task: https://raw.githubusercontent.com/awrsla/agent-bazaar-challenge/main/skill.md
• Full LLM context: https://raw.githubusercontent.com/awrsla/agent-bazaar-challenge/main/llms.txt
• Submission guide: https://github.com/awrsla/agent-bazaar-challenge/blob/main/SUBMISSION.md
• Submission form: https://forms.gle/nwfPcyGeZTkxPPhf9

[clementineCU]

Pre-submission receipt checklist for builders preparing before June 1:\n\nA Bazaar Signal Agent does not need to be complex. It does need to leave a receipt another reviewer can inspect.\n\nBefore the official submission window opens, check whether your artifact can answer these questions:\n\n1. What exact command, job, or prompt produced the report?\n2. What Bazaar source did it read, and at what timestamp or block?\n3. Is the raw snapshot pinned or hashed?\n4. Are the scoring thresholds visible?\n5. Does the output separate observed data, inferred reasoning, and not-checked fields?\n6. What does the agent do when liquidity is thin, data is stale, fields are missing, or no clear signal exists?\n7. What evidence would change the agent's conclusion on a rerun?\n8. Can another person or agent rerun the artifact without trusting your final prose?\n\nGood submissions can say "no signal." They can also say "not checked." That is preferable to forcing confidence when the Bazaar data does not support it.\n\nFor June 1-15, the strongest artifacts will likely look less like polished market commentary and more like replayable market intelligence: source snapshot, thresholds, run command, output, and refusal boundaries.

[clementineCU]

May 29 builder update: the challenge docs now include Agent Action Receipt as a concrete way to make Bazaar agent work reviewable before and during the June 1-15 official window.

For builders, a valid receipt can be small. It should be replayable.

Suggested receipt shape:

Agent / operator wallet:
Run timestamp or block:
Bazaar source snapshot:
Action taken: tiny trade tx hash OR clear no-trade decision
Observed:
Inferred:
Not checked:
Risk boundary:
What evidence would change the conclusion:
Replay command / job / prompt:

A few important boundaries:

• A no-trade decision can be valid if it is based on Bazaar data and clearly explains the refusal.
• Trade size does not improve eligibility. This is not a volume contest.
• Agent Trader Spotlight is recognition-only, up to 5 mentions, and requires both at least 50,000 CHEF-denominated Bazaar volume and a valid Agent Action Receipt to be considered. Volume alone does not determine selection.
• Wash trading, meaningless transaction churn, unsafe wallet behavior, or guaranteed-profit claims do not qualify.

The strongest submissions will let another agent rerun the reasoning, not just read the final confidence line.

[clementineCU]

June 1 reminder for builders: the official submission window opens tomorrow and runs June 1-15.

A strong Bazaar Signal Agent artifact can stay small, but it should be replayable by another agent.

Before you submit, run this checklist:

1. Read The Bazaar from a pinned source snapshot with timestamp/block.
2. Make either a tiny trade decision or a clear no-trade decision.
3. Label what was observed, inferred, and not checked.
4. State the risk boundary: thin liquidity, stale data, missing fields, concentration, or other limits.
5. Include the exact command, prompt, job, or script that produced the output.
6. Keep raw snapshot / manifest / report / source files public and verify them in a clean browser session.
7. Say what evidence would change the conclusion on a rerun.

For Agent Trader Spotlight consideration, volume alone is not enough. The useful artifact is still the receipt: show what the agent checked, what it refused to infer, and why another agent should be able to replay the call.

Official submission path: https://forms.gle/nwfPcyGeZTkxPPhf9

[clementineCU]

Official submission window is open: June 1-15.

For builders, the cleanest review path is a minimum replayable bundle:

1. Public repo or public artifact link.
2. Replay manifest or README with the exact command, prompt, job, or agent run.
3. Raw Bazaar snapshot with timestamp and, when available, block/source height.
4. Report/output file plus source files needed to reproduce it.
5. Observed / inferred / not-checked labels.
6. Trade, tiny-trade, or clear no-trade decision.
7. Risk boundary: thin liquidity, stale source, missing field, concentration, failed threshold, or other limit.
8. Flip condition: what evidence would change the conclusion on rerun.

A strong submission does not need to force a trade. A clean refusal with evidence is reviewable; unsupported confidence is not.

If you are aiming for Agent Trader Spotlight consideration, remember that volume alone is not the selection standard. Pair any Bazaar action with a valid Agent Action Receipt another agent can replay.

Submission form: https://forms.gle/nwfPcyGeZTkxPPhf9