PayAna — Track A Submission Package

Status checkpoint: 2026-09-07 19:35 BRT

Official Track A deadline: 2026-09-08 23:59 UTC / 20:59 BRT.

This document is the final handoff package for the human/account steps that cannot be truthfully automated from the PayAna runtime.

Project: PayAna — Falsifier Agent for Binance Agent OS

Track: A — Build an AI agent using Agent OS

One-line pitch: PayAna is an evidence-first market-analysis agent that does not merely state what the market appears to support; it exposes the evidence against its own conclusion, explicit falsifiers, missing context and the next test that could prove it wrong.

Live demo: https://payana-agent-os.val.run/

Machine-readable submission evidence: https://payana-agent-os.val.run/api/submission-evidence

Technical release: payana-market-sentinel-1.2.0-rc1

Validation receipt: 189 assertions + 35 rate-limit mutation steps = 224 executed checks PASS.

Safety boundary: MARKET_DATA_READ != TRADING_AUTH and every analysis ends with NO_TRADE_DECISION.

Verified against the Binance Mini Hackathon announcement and the Binance Developer MCP page, which was last modified on 2026-09-07:

• Track A closes at 2026-09-08 23:59 UTC / 20:59 BRT.
• Track A entry requires following @Binance, reposting the announcement, replying or quote-reposting with the submission, and completing the official survey. Binance says Track A means a video or demo plus GitHub if applicable.
• The event page explicitly excludes the US, UK, EEA, Hong Kong, Singapore and Binance-prohibited jurisdictions. Brazil is not explicitly named in that exclusion sentence, but participant/account eligibility is still UNVERIFIED until the logged-in Binance flow accepts the account.
• Official MCP docs classify Market Data as public, no auth at the capability level, but the supported-client MCP server connection still has an authentication step. Do not equate public data capability with an anonymously usable MCP catalog.
• Official supported clients include ChatGPT on the web, ChatGPT/Codex Desktop, Codex CLI, Claude Code/Desktop, VS Code and Grok Bot.
• Binance's official connection check is: ask the client to use the Binance MCP Server for current BTCUSDT price + 24-hour change, then verify that the response visibly shows the Binance MCP tool and live market data.
• The official docs say do not paste the MCP endpoint into a chat and ask the model to install it, and do not open the MCP endpoint directly in a browser; use the client's MCP/plugin setup flow.
• Market-data reads run without a trading confirmation; every trade/transfer requires confirmation, and withdrawal scope is never available. PayAna does not need Trade, Account or Transfer for the Track A read-only demo.

Official references:

• Hackathon: https://www.binance.com/en/blog/community/8802181509900814931
• Agent OS: https://www.binance.com/en/agent-os
• MCP docs: https://developers.binance.com/en/docs/agent-native/mcp-server/agentic

1. Falsification instead of answer-only confidence. Every market briefing contains a winning/current hypothesis, the opposite hypothesis, an alternative mechanism, supporting evidence, opposing evidence, observable falsifiers and a next discriminating test.
2. Missing evidence is not fabricated evidence. Failed or malformed sources become null and are excluded from aggregation; they never silently become zero.
3. Traceable evidence. Ticker, top-20 order book and 5-minute candle reads produce explicit source receipts with HTTP state and latency.
4. Currentness-aware. Live timestamps are checked and the UI automatically refreshes while visible.
5. No trading authority required. The reference cockpit uses read-only Binance market evidence and has no Trade, Account, Transfer, wallet or arbitrary upstream action endpoint.
6. Agent OS integration is fail-closed. The optional MCP/OAuth verification path uses CIMD + Authorization Code + PKCE S256, one-time state, no persistent Binance token and no automatic tools/call.
7. Negative controls are part of the product. Invalid symbols, malformed HTTP-200 responses, total/partial provider failure, replay, concurrent OAuth state consumption, MCP dangerous-tool classification and rate limiting are executable tests, not documentation-only claims.

Screen: PayAna landing view.

Narration:

Most market agents optimize for producing an answer. PayAna optimizes for showing why its own answer could be wrong.

Screen: BTCUSDT, then scroll through market metrics and source receipts.

Narration:

PayAna reads live Binance market evidence, validates three independent source surfaces, computes reproducible microstructure and momentum metrics, and records a receipt for every source.

Visible proof to capture:

• symbol BTCUSDT;
• evidence state PASS;
• three source receipts;
• current timestamp;
• NO_TRADE_DECISION.

Screen: Hypotheses and Falsification sections.

Narration:

Instead of a single confident narrative, it creates a current hypothesis, the opposite hypothesis and an alternative explanation. Then it shows evidence for and against the reading, conditions that would falsify it, and the next observation that should discriminate between them.

Visible proof to capture:

• Current hypothesis;
• Opposite hypothesis;
• Alternative mechanism;
• Supporting evidence;
• Opposing evidence;
• Falsifiers;
• Next test.

Screen: briefly show README/test evidence or terminal/Val Town release result.

Narration:

PayAna is designed to fail visibly. Missing or malformed evidence never becomes zero, and a partially available market does not receive fabricated certainty. The current release passes 224 checks, including malformed-source mutation, OAuth replay and concurrent-state tests.

Screen: permission boundary / Verify Agent OS authorization button, without claiming an authenticated session unless one has actually been proven.

Narration before real MCP proof exists:

The Agent OS connection path is permission-minimized and fail-closed. The hosted reference remains read-only, stores no Binance access token, performs no automatic MCP tool call and has no trading endpoint.

If a real supported-client Binance MCP Market Data proof is completed before recording, replace the last sentence with:

The same falsifier workflow can consume Binance Agent OS Market Data through the MCP connection while keeping Account, Trade and Transfer outside the PayAna workflow.

Screen: top of cockpit / project name.

Narration:

PayAna's thesis is simple: an AI agent becomes more trustworthy when it is required to expose what would prove it wrong.

Record one continuous witness where possible:

1. open the production demo at https://payana-agent-os.val.run/;
2. show BTCUSDT and the live timestamp;
3. show the three source receipts;
4. show hypothesis / opposite / alternative;
5. show falsifiers and next test;
6. show NO_TRADE_DECISION and the permission boundary;
7. if Binance MCP is connected in a supported client, separately show the client using a Binance MCP Market Data tool for BTCUSDT and capture the tool attribution;
8. do not show passwords, cookies, OAuth codes, access tokens, refresh tokens, wallet addresses, API keys or private account information.

Use this only after the final video/demo URL is available and only after actually following @Binance and reposting the official announcement.

Track A — Binance Agent OS Mini Hackathon

Meet PayAna: an evidence-first Falsifier Agent that asks not only “what does the market support?” but “what would prove this reading wrong?”

Live demo: https://payana-agent-os.val.run/ Video: [VIDEO_URL] GitHub: [GITHUB_URL — include only if created]

224 release checks · explicit source receipts · competing hypotheses · observable falsifiers · NO_TRADE_DECISION.

@Binance #BinanceAgentOS

Before posting, verify the final X character count after replacing placeholders. If the resulting post is too long, use this compact version:

Track A — Binance Agent OS Mini Hackathon

PayAna is an evidence-first Falsifier Agent: live Binance evidence, competing hypotheses, source receipts, observable falsifiers and NO_TRADE_DECISION.

Demo: https://payana-agent-os.val.run/ Video: [VIDEO_URL] GitHub: [GITHUB_URL]

@Binance #BinanceAgentOS

PayAna — Falsifier Agent for Binance Agent OS

PayAna is an evidence-first market-analysis agent built around falsification rather than answer-only confidence. It collects current Binance market evidence, validates source state and timestamps, computes reproducible microstructure and momentum metrics, and then produces competing hypotheses instead of one narrative. Every briefing includes supporting and opposing evidence, alternative explanations, observable falsifiers, missing context and a concrete next test. The reference cockpit is intentionally read-only and ends with NO_TRADE_DECISION.

AI market analysis can sound coherent even when it is based on stale, incomplete or malformed data. PayAna makes these failure modes visible. Missing evidence is never silently converted to zero, partial evidence is explicitly labeled, source receipts are exposed and each conclusion is paired with conditions that could invalidate it.

PayAna is designed around Binance Agent OS Market Data as the market-evidence capability. The hosted reference implementation uses Binance's public Spot market-data surface for reproducible live analysis. An optional Agent OS connection-verification path implements current OAuth/CIMD/PKCE boundaries without persisting Binance tokens or invoking tools automatically. If the final supported-client MCP proof is completed, add the exact client/tool evidence here rather than making a generic claim.

PayAna does not need Account, Trade or Transfer authority for its Track A workflow. The hosted reference has no trade endpoint, no wallet action and no persistent Binance access token. Its hard boundary is MARKET_DATA_READ != TRADING_AUTH.

https://payana-agent-os.val.run/

[GITHUB_URL if created; otherwise use the public Val Town source link if the form accepts it]

[VIDEO_URL]

Use the official Binance MCP setup flow in a desktop supported client. Keep the PayAna proof read-only and permission-minimized.

1. In the client's MCP/plugin settings, add the official Binance MCP server using the documented client flow. Do not paste the endpoint into the normal chat as an instruction.
2. Authenticate the server through the Binance browser flow.
3. Grant only the minimum capability required for the witness. For PayAna, the target is Market Data; do not enable Trade or Transfer merely for the demo.
4. Return to the client and ask exactly: Use the Binance MCP Server to show the current BTCUSDT price and 24-hour change.
5. Capture one screenshot or short clip where the Binance MCP tool attribution is visible and live BTCUSDT market data is returned.
6. Record client name, UTC timestamp and the visible tool/result in the PayAna receipt ledger. Never record authorization codes, cookies, access/refresh tokens or account secrets.
7. If the client forces creation/selection of an Agentic sub-account during authentication, complete only the required account onboarding; do not fund it for Track A read-only evidence unless Binance explicitly requires funding for eligibility. Official docs only require funding before trading.

Expected classification after a clean witness:

SUPPORTED_CLIENT_MCP_CONNECTION = VERIFIED
MARKET_DATA_TOOL_USE = VERIFIED
ACCOUNT_SCOPE = NOT_REQUIRED_FOR_PAYANA
TRADE_SCOPE = NOT_GRANTED
TRANSFER_SCOPE = NOT_GRANTED
NO_TRADE_DECISION = TRUE

• Confirm participant/account eligibility for the hackathon.
• Connect Binance Agent OS in a supported MCP client and capture a real Market Data tool proof if feasible.
• Record the final video/demo.
• Create/publish GitHub mirror if desired/applicable and verify its URL.
• Follow @Binance on X.
• Repost the official hackathon announcement.
• Reply or quote-repost with the Track A submission.
• Complete the official Binance survey.
• Preserve the final X post URL, survey completion receipt, video URL and GitHub URL in PayAna's durable receipt ledger.

Before submission, the language must satisfy all of these:

PUBLIC_API_PASS != MCP_CONNECTION_PROOF
MCP_ENDPOINT_KNOWN != MCP_SESSION_AUTHORIZED
AUTHORIZATION_SURFACE_REACHABLE != CIMD_ACCEPTED
MODEL_SCORE != CALIBRATED_PROBABILITY
CURRENT_PRESSURE != FUTURE_PRICE
MISSING_EVIDENCE != ZERO
SCREENSHOT != BEHAVIOR_PROOF
NO_TRADE_DECISION remains true

The submission is ready when the technical package remains green and each required external account step has an actual receipt, not merely a planned action.