Since September 15, ten audit payments have settled at the gate. Every one was paid by a wallet I fund. Agents that have paid, not counting my own wallets: still zero.
What changed is everything around the verdict. The gate now runs a show: one narrative a day, audited before it airs and logged with a receipt on Base. Every row waits for its own date in a public ledger, and the first one has already been graded. A second way to pay, MPP, went live on November 1.
// SYS_ID : djzs-mainnet-01
// TOOLS : verify_pm_trade · verify_perp_trade · LIVE · PAID
// PAY : 2.00 USDC · x402 or MPP · Base
// FREE : djzs.ai/try · 3 a day, no wallet
// LEDGER : scan.djzs.ai · misses publish too
// STATUS : OPERATIONAL
The engine did not change. The doors to it did. One rule set judges every prediction-market thesis, DJZS-PM-v1.0, and it still answers in three outcomes, never two.
PASS: every scored fact is known and no blocking rule fired. Proceed.
WAIT: a decision-critical fact is unknown. The caller halts.
FAIL: a rule fired on facts the engine knows. Stop.
The four codes are unchanged: M01 narrative resolution gap (critical, 30), M02 falsification absent (critical, 30), M03 probability unsourced (high, 25), M04 consensus no edge (medium, 15). The FAIL threshold is 25. A critical code fails on its own; M04 alone rides a PASS with the flag on the certificate.
Door | For | Cost | What comes back |
|---|---|---|---|
People | Free, three a day, no wallet | PASS, WAIT or FAIL and the reasons in plain words | |
| Agents, prediction markets | 2.00 USDC on Base, x402 or MPP | Verdict, flags, verdict_hash, intent hashes, an EAS attestation |
| Agents, perps and spot | 2.00 USDC on Base, x402 or MPP | Same shape under DJZS-LF-v1.2, four of eleven codes wired live |
| Anyone | Free | Certificates on Irys; an agent's trust record from the subgraph |
The install line has not moved:
claude mcp add --transport http djzs-trust https://mcp.djzs.ai/mcp
Two changes since September matter for payers. The free route at /try went live on September 28: the same engine, no payment, no attestation. And since November 1 the paid routes also accept an MPP credential beside x402, same facilitator, same treasury, same receipt. No payment, no verdict. No verdict, no execution.
Five narratives went through the gate as show rows, in six paid audits, and all six came back FAIL. M02, falsification absent, fired on every one: none of the stories named what would prove it wrong.
Logged (UTC) | Row | The narrative, verbatim | Book | Verdict | Codes | Grades | Receipt |
|---|---|---|---|---|---|---|---|
Nov 5 02:28 | SCAN 007 | “…with traders pricing an end to the Iran war…” OilPrice, Nov 4 | Polymarket intl: US ends blockade by Nov 30 | FAIL 85 | M01 M02 M03 | Nov 30 | |
Nov 5 00:55 | SCAN 006 | “This is our election. We’ve won it.” Scott Colom, Nov 4 | Polymarket intl: MS Senate, Colom | FAIL 75 | M01 M02 M04 | Nov 13 | |
Oct 30 20:28 | SCAN 003 (re-run) | “Binance will cover all losses from the SAFU fund and the company’s funds.” Binance, Oct 30 | Polymarket intl: major CEX insolvent in 2026 | FAIL 75 | M01 M02 M04 | Dec 31 | |
Oct 30 19:20 | SCAN 003 (first run) | Same pledge, from a stale memo | Same book | FAIL 75 | M01 M02 M04 | Superseded | |
Oct 27 17:15 | SCAN 002 | “Anthropic S-1 filing, ticker ‘ANT,’ Nasdaq listing week of Nov 16” SoSoValue, Oct 27 | Polymarket intl: Anthropic IPO in November | FAIL 45 | M02 M04 | Nov 20 | |
Oct 26 04:15 | SCAN 001 | “50 basis points in total hikes expected” J.P. Morgan Wealth Management, Oct 14 | Kalshi KXFEDDECISION-26OCT-H25 | FAIL 70 | M02 M03 M04 | Graded Oct 28 |
One row is corrected, by name. The first SCAN 003 memo was frozen from an early read and missed a Binance post published 17 hours before the audit. That record stays on Base, unedited. The re-run on a corrected memo is a new record, and it returned the same verdict.
The other paid audits were not show rows. Two were perps theses under DJZS-LF-v1.2: a PASS 0 on November 2, and a FAIL 15 on November 5 that was the Scan Room's own audit door checking itself. A FAIL against a story is not a call on the market. The verdict is on the narrative, never the contract.
SCAN 001 graded on October 28. The binding I set lost, and FAIL 70 stands.
The story was J.P. Morgan Wealth Management's October 14 piece: “Will the Fed hike rates in October? 50 basis points in total hikes expected…” The gate logged it on October 26 as FAIL 70: no falsifier, no sourced probability, and the crowd's own side. The body put the hikes “in October, December or in 2027.” A story with three possible dates cannot be proven wrong by any one of them.
So the row carried a binding of my own: a 25 basis point hike at the October 28 meeting, settled on Kalshi KXFEDDECISION-26OCT-H25.
What printed: the FOMC “decided to maintain the target range for the federal funds rate at 3-3/4 to 4 percent,” by a 10–2 vote. Lisa D. Cook and Beth M. Hammack preferred a quarter-point hike. Federal Reserve, October 28
What the book did: KXFEDDECISION-26OCT-H25 finalized No; the event settled on “Fed maintains rate” (Kalshi API, read November 5).
The grade: the bound leg lost. The narrative itself is still open, because it never named a date it could lose on. That is the defect the gate flagged two days before the meeting.
This is the standard, stated the first time it applies. A FAIL is a grade on the reasoning, not a bet against the outcome. When a FAILed story comes true later, it is still graded, and the grade goes up in the same place the verdict did. Misses publish too.
The gate's verdicts now have a public scoreboard at scan.djzs.ai, also served at djzs.ai/scan. It holds four entries today: two calls and two agent keys. Nothing in it has been graded yet, because nothing in it has reached its date.
A call is one line: what, a level, a date, and the source that settles it. Here is the first one in the ledger, as logged:
Bitcoin above 80000 by 2026-12-28 source: Coinbase BTC-USD daily close
It is graded on the source's print on the deadline. Touching the level earlier does not count. If the source never prints, the call is void and counts for no one. No edits, no deletes, three calls a day, and every line is hash-chained to the one before it (method).
Door | How | Logged |
|---|---|---|
The page | Sign in with X, fill four fields | Automatic |
X | Reply to @Djzs_ai with | Automatic |
YouTube | Comment | Automatic |
Agents | JSON API with a key minted on the page | Automatic |
My own first call carries a flaw I left in place. It names Brent below 100 by December 13, and December 13 is a Sunday. ICE does not settle Brent on a Sunday, so under the room's own rule it will void. The room now warns before anyone posts a weekend date. The call stays, because nothing in the ledger gets edited.
The show's rows are not in the room yet; they live on Base and in each episode's description. SCAN 002, 003, 006 and 007 can still enter before their dates. SCAN 001 cannot: the room refuses anything logged after its own deadline.
The ledger now has a front. The Daily Narrative Scan takes the loudest story of the day that a market can settle, runs it through the gate, and airs the row. The order never changes: audit first, log the receipt, then record.
Four videos are up on YouTube @djzs_ai: the explainer, SCAN 001, SCAN 002 and SCAN 007. SCAN 006 went out on X @Djzs_ai. Each episode reads the narrative verbatim, names the defect in one sentence, shows the receipts, then the verdict and what settles it.
The cadence on paper is one episode a day and one graded review a week. The record so far is five rows in eleven days. When time is short the ledger updates and the episode skips, never the reverse.
Some episodes are read by an AI presenter: a licensed Tavus stock face and its voice, reading a written script. She issues no verdict. Every verdict she reads was logged by the gate before she read it, and the label sits on screen the whole time.
She also has a live door. djzs.ai/present went up on November 5 and is now linked from the landing page and every page menu as Ask. It is a five-minute video call: she walks through how djzs.ai works, then takes questions, by voice or typed. Ask her whether your trade is good and she declines and sends you to /try. That is the gate's job, not hers. One call a day per browser.
What is live is stated as live. What is weak is stated too.
No outside payer yet. Ten paid settlements reached the treasury after September 15, 20.00 USDC in all, from three wallets I fund. The plumbing works; the traffic has not come.
Stated is not true. M03 checks that a probability names its source, not that the source is right. A well-formed invented basis can pass. Provenance stays the operator's job.
The gate audits the memo it is given. SCAN 003 shows the failure mode: a stale memo got a correct audit of stale facts. The fix was a new record, not an edit.
A verdict_hash names a verdict, not a row. The first SCAN 003 run and SCAN 006 share one, 0xb1a2…9bb3, because they failed the same way on different stories. To cite a row, cite its attestation or its intent hash.
Perps are partial. DJZS-LF-v1.2 has four of eleven codes wired live, and the audit does not read a venue's live price, funding or liquidation data.
Poisoning follows payment. Six zero-value USDC transfers to look-alikes of my payer wallets appeared beside real settlements. Addresses come from the live 402 challenge, never from wallet history.
Nothing in the room is graded yet. The boards rank no one before three graded calls, and the first call dates are in December.
Six dates are on the record, and each one grades in public whichever way it goes.
Date | Row | What settles it | Where to check |
|---|---|---|---|
Nov 13 | SCAN 006 · Colom | AP, Fox News and NBC all call it for Colom, or certification does. Uncalled as of November 5 (ABC7) | Polymarket intl: MS Senate book |
Nov 20 | SCAN 002 · Anthropic | ANT opens for regular-way trading on Nasdaq by Friday, November 20 | Polymarket intl: Anthropic IPO in November |
Nov 30 | SCAN 007 · blockade | The US officially announces an end to its naval blockade of Iran by 11:59 PM ET | Polymarket intl: US ends blockade by Nov 30 |
Dec 13 | C0002 · my call | Brent below 100 on the ICE front-month settlement. A Sunday, so it should void | |
Dec 28 | C0001 | Bitcoin above 80000 on the Coinbase BTC-USD daily close | |
Dec 31 | SCAN 003 · Binance | Binance announces insolvency or a bankruptcy filing by 11:59 PM ET | Polymarket intl: major CEX insolvent in 2026 |
If neither a three-network call nor certification exists on November 13, SCAN 006 waits for the first of the two. The grade needs the print, not the projection.
{
"STATE_OF_THE_GATE": {
"as_of": "2026-11-05",
"sys_id": "djzs-mainnet-01",
"paid_tools": ["verify_pm_trade", "verify_perp_trade"],
"free_tools": ["query_pol_certificates", "query_agent_trust"],
"price": "2.00 USDC · x402 or MPP · Base",
"free_route": "djzs.ai/try · 3 a day",
"rulesets": { "DJZS-PM": "v1.0 · 4/4 live", "DJZS-LF": "v1.2 · 11 frozen / 4 live" },
"since_2026_09_15": { "paid_settlements": 10, "usdc": "20.00", "outside_payers": 0, "attestations": 8 },
"show": { "rows": 5, "audits": 6, "fail": 6, "graded": 1 },
"scan_room": { "entries": 4, "calls": 2, "graded": 0 },
"next_grade": "2026-11-13 · SCAN 006"
}
}If you want this gate in front of your agent's next position, it is one line away at djzs.ai. If you want it explained first, ask the presenter. No payment, no verdict. No verdict, no execution.
END_TRANSMISSION. //

