COSMO · public evidence bundle

SupraFX maker window B2: four verified live fills

4 FILLS PROVEN 2026-10-17 · 12:00–13:00 UTC (14:00–15:00 Berlin) Supra chain 8 · SupraFX · iUSDC/SUPRA · maker side GO allowed 5 · supply ran out after fill 4

4 / 5fills proven (limit set by the GO)
5 min 44 sfirst to fourth fill
0errors in 57 ticks
46 / 46checks per fill bundle, offline re-verify
6live fills over three windows, 16.–17.10.

Scope of this proof: COSMO ran as a maker on SupraFX for one armed window under one written GO. The runner decided every open iUSDC/SUPRA request-for-quote (RFQ) against pinned policy and gates, placed a quote only after a FINAL AUTHORIZED decision, and proved each fill quant-exact against the venue's own balance snapshot. Four fills were placed and proven, one RFQ was declined with a named reason, and after minute six no new RFQ arrived until the quote cutoff. Capital was bounded by the venue delegate's caps, not by trust in the runner.

Every file below is a byte-identical copy of the window directory and the five case bundles. The record that was written from them is b2-record.md (German). Fingerprints are SHA-256; the list is also in index.txt.

← Media (the showcase video of the first two fills) · All evidence folders

The window

steptime (UTC)resultfile
Preflight G1–G1011:50:35READY_WITH_DEVIATIONS (G8 = two named dApp deviations, see limits)window/go-check.md
Policy check11:53:56PASS (P-ACTIVE … P-EXPIRY 10/10, F6 live: exactly the armed delegate, no open orders)window/policy-check.json
Arm before window start11:54:12refused by code (ARM_WINDOW)—
Arm12:00:09armed, max_fills 5, bound to commit 975241d, arm hash 0x4544f5c0…window/arm.json
Run12:00:28 – 13:00:1657 ticks, 5 cases, 4 attempts, 0 errors, ended NO_FILL at the quote cutoff (no further RFQ)window/run-state.json
Unwind13:10nothing to do: 0 open RFQs, 0 open quotes, 0 lockswindow/unwind.json
Offline re-verify13:1x5 bundles ACCEPT; tampered copy REJECTreverify.log · tamper.log

The fills (quant-exact)

#quote placed (UTC)RFQquote idrate SUPRA/iUSDCSUPRAiUSDCproofbundle
112:00:41R-261017-113847-169c647c59-643e-36f0-6d3e-32b50ba1c4486984.12−13,966.85220925+2.00000000PROVEN FP1–FP4kdm-…-4eee4645
212:02:14R-261017-113432-c8c047fcf9-bc02-6c96-40f8-165da82608056986.66−13,971.92413590+2.00000000PROVEN FP1–FP4kdm-…-14259336
312:03:45R-261017-112617-89aa2e5453-63bc-f0f7-7698-f3b82d191dde6986.26−13,971.12061833+2.00000000PROVEN FP1–FP4kdm-…-c7825c49
–12:05:13R-261017-112313-45(no quote)6986.2200NOT_AUTHORIZED V-3M: remaining batches 28 < 60kdm-…-7e75046d
412:06:26R-261017-120543-a3c5ff58dd-0504-ca06-f022-533a090c7be66983.98−13,966.57257221+2.00000000PROVEN FP1–FP4kdm-…-1fcac31c
sum−55,876.46953569+8.00000000

Venue balance (COSMO master account, SupraFX council snapshot): SUPRA available 131,252.38 before the window (G6), 75,375.91 after fill 4 (batch 12555491, window/fill.json); the difference equals the sum of the four deltas. Each quote = fair rate minus 50 bps (fair 7019–7022 from the SupraFX reference 7073.5 → 7021.2), 1.50× above the buyer floor 4664.83. Each fill locked ~13,970 SUPRA (lock ceiling) and released it minus a 1 bp maker rebate (~1.397 SUPRA). All four RFQs came from the same taker address on the venue (one counterparty).

What was checked, and by what

Tamper probe: one word changed in a copy of a FINAL decision → REJECT (2 of 46 failed). Log: window/b2-tamper-probe-20261017.log.

Files (byte-identical copies)

Check integrity with: sha256sum <file> · each case folder carries its own SHA256SUMS (cd <case> && sha256sum -c SHA256SUMS) · full list in index.txt

filesha256
b2-record.md25f7c607aade682bee39e83ea6c6fe83b49ccb4b628942ac6e733c620fe81f87
window/go-check.json01cbeaadf5e3f52ae52756c8f9ea732361e9c0bfe9160a54d7b4cecbeba69ac3
window/go-check.mdaf65370580ef745ad689110cff952a912bb1bd8fd74297c497c952c68a1b662b
window/policy-check.json02862aff57e292a3e2ffdedbac120b941f2df1e765167c5fabe293367e00d85e
window/arm.json1bcba86649d950370d5bc3cf145a1fbd194485b6a04e85eafd6a1e43adfe1edc
window/run-state.json3b8893659d099f0a3577148d4d3a89b00eda48cd3e15cda6d86c42a7eb29b4a7
window/fill.json (fill 4, with the council snapshot)9425ede3211a4400d7afce0f9138ee69aaf7bee560fc69b8cbd58a717ad4adad
window/fills-ledger.json75fa86787d5a0ea37439bb1c7e0790bdfb6f63285b088cf76178450e6abd44f8
window/unwind.json289dc535c8af861920e764130f3fe212b6f812f0adf43311d1ebd0d55075e80c
window/live.ndjson (runner events)ad89564b22acc163c1cf19cfb755357a2764a17db5dcefb8a6d16830d4e57e52
window/log.ndjson (watch and case log)2bd18c55557dd48284363f925ef57f11c80d004bfa40f9cada3bff692d9eee45
window/decided.json42d6b8795310c928bdba9216bdcef49413fdd0deec342160d833d49c7ca11ca0
window/tests.json8202674c8463f7400c5b16a47727fbe61eb64bfb7bb578c5e20fd38441c161e6
window/expiry-calc.json39196d334ac040677987cd8f32bb4cdab186a353b04d82d737500c0b2a4a9283
window/b2-reverify-20261017.logc9b59607db4d2cdd3583d60843c95f6adf558527364d86b9250d4f92433a16a4
window/b2-tamper-probe-20261017.log1a434a924ae3a51f1b85ee858a2209bbea9420d3dc8e5335a84121542517f502
cases/…-4eee4645/SHA256SUMS (fill 1)4791229fedb8461e4f15314cd9b0ddff7456f6f85341c8dcac15ad40641c255d
cases/…-14259336/SHA256SUMS (fill 2)2d8146473aa5f9c024a15e43caa3272f1a354a7e804108de26d16e2877b03cea
cases/…-c7825c49/SHA256SUMS (fill 3)0a389eca3b2c06297e4b4e168e17dfd7a987214d489e654ca8d0be893401f3dd
cases/…-7e75046d/SHA256SUMS (declined)83b6d3b580552a3004e378a05b352885c7ad03aa6b021dea2e7093c176e064ce
cases/…-1fcac31c/SHA256SUMS (fill 4)1e8fdd9c4925f686e4b98ee158b138b8c54f3e50294be14fb9d95d8b2f7797a1

A case bundle holds the intent, both decisions (RESERVATION, FINAL), the stored venue reads the decisions were made on, the pins, the zero-capital check and, for a placed quote, live/ with the arm, the signed execution envelope, the ledger, the funds snapshot after the fill and the fill proof. The bundle hash in the fills table is the sha256 of that folder's SHA256SUMS.

Honest limits

· This is a self-attestation by the COSMO runner. The offline re-verification checks internal consistency against the venue reads the runner stored, not world truth, and the verifier is written by COSMO. The verifier program is published at /verifier/ (cosmo-verify-maker.mjs 1.0.0, sha3-256 0x06d97f90f03767ec4393535aec00ca4514ef86256d164faebe7c90d2c06e6e50); it runs the same MB1-MB12 checks as the repository verifier at commit 4a28c82 (46 on a bundle with a fill, 29 on the declined case; a tampered copy fails 2 of 46). Run it on any cases/<id>/ folder of this bundle.

· All four RFQs came from one taker address. This proves a repeatable maker chain, not market depth and not a second counterparty.

· Economic size is pilot size: 2 iUSDC per fill at fair minus 50 bps, roughly 0.01 USD margin per fill.

· Two named deviations from the written policy were accepted in the GO text, because the venue's delegate form has no pair field and no expiry field: the pair restriction is carried by the asset caps (only SUPRA and iUSDC are capped, every other asset 0), and the delegate's venue expiry (~84 days) is longer than the window. Since the operator's decision after window 2, the delegate stays active between windows; the capital bound is the venue cap (15,000 SUPRA / 3 iUSDC, maker only, no withdrawal), plus the runner's own policy.

· The runner decides one case per 60-second tick. Fill 5 did not happen because no new iUSDC/SUPRA RFQ appeared between 12:06 and the 13:00 cutoff, not because of a refusal or an error.

· Fingerprints here are SHA-256 (the bundles' own convention); older evidence folders on this site use SHA3-256.

· Windows 1 and 2 (16.10., one fill each, −13,542.71 and −13,560.53 SUPRA for +2 iUSDC each) are recorded in the repository and shown in the showcase video; their bundles are not published on this page.

Frozen 2026-10-17. Originals are never modified after publication. · All evidence folders · Proof overview