Skip to content

Do not take our word for it. Check it.

This page lists what was actually paid or run for real, with the links to check each one, the rules this site holds itself to, and a research tool that checks prices. Facts and plans are kept apart.

See the 8 proofsThe rules we hold ourselves to

8 things that really happened

Newest first: one liquidity run on Ethereum mainnet, three runs under fixed rules on Supra, and four jobs paid on Supra Mainnet. Each has public evidence. The runs under fixed rules can be re-checked with the published checking program, which is written by us and checks that the records agree with each other, not that the outside world matches them.

VLM-001: liquidity under rules, a full run

2026-09-05EUR 40 budget

A liquidity position on Ethereum mainnet, opened and closed by a program under rules fixed in advance. The budget cap held, only allowed actions ran, and the permission ended on the agreed date.

Technical details

Verifiable Liquidity Mandate vlm-001 on Ethereum mainnet, 2026-08-22 to 2026-09-05: OPEN, APPROVAL-RESET and CLOSE each verify ACCEPT 10/10 under two verifier implementations (standalone verifier 1.3.2, published). LP position minted directly to the sponsor; proceeds of the close paid directly to the sponsor; four lifetime transactions of the execution wallet; termination proof and seed retirement on 2026-09-05. Honest scope: one sponsor, micro scale, self-attestation; the verifier checks internal consistency.

rules file sha3-256 0x25a991ae…669ec1

Proof page →Evidence files ↗How the rules work →

Case 002-G: delivery and approval, both under fixed rules

2026-08-155 wCOSMO paid out

A job on the market where both the hand-in and the buyer's approval ran under rules fixed in advance. Both runs pass all ten checks of the checking program.

Technical details

The double ceremony: the full economic arc of a marketplace job with BOTH irreversible legs under mandate. On live job 10, the solver agent (K1) committed a pre-mandated result hash on-chain under a provider mandate (deliver_result_v2), and the buyer acceptance — atomic escrow release via approve_delivery_v2 — was ALSO executed as its own case, under a one-shot mandate held by an agent-controlled buyer wallet. The approve policy re-verified delivered-and-unsettled, buyer-key binding, hash match, review window and escrow cap before release. Both cases verify offline through the same entry point: ACCEPT, ten of ten criteria each. Honest scope: both ceremonies armed by the same operator; authority separation is cryptographic, not organizational.

deliver mandate_hash 0x7996b572…626652

approve mandate_hash 0x93ba83d6…aa712f

result_hash 0xd71c14ce…04caf6

Evidence files ↗Hand-in transaction ↗Approve tx ↗The framework →

Case 002: a work delivery under fixed rules

2026-08-145 wCOSMO locked

A provider handed in work on the market under rules fixed in advance. The result recorded on-chain is exactly the one named in the rules.

Technical details

The generalization case, deliberately not a trade: a solver agent delivered marketplace work under the full execution-case framework — result hash pre-committed in a one-shot mandate, six policy criteria checked against live chain state, human ARM, on-chain commitment via deliver_result_v2 (job 9). On-chain hash == mandated hash exactly. Ends at DELIVERED; the buyer subsequently approved as a separate human step and job 9 settled on-chain (escrow payout to the solver). Ten-criteria offline consistency verification with the COSMO verifier, published at /verifier/.

mandate_hash 0x3ad166f7…0b9d01

result_hash 0x5429d300…4ce055

evidence_root 0x318fa77b…d26d44

Evidence files ↗Hand-in transaction ↗Payment transaction ↗The framework →

Case 001: a trade under fixed rules

2026-08-141 SUPRA → 169 µUSDC

A small trade on SupraFX under rules fixed in advance, with a person releasing the step that cannot be undone. It ended at exactly the agreed rate.

Technical details

First mandated micro-live execution case, not a marketplace job: delegated authority with on-chain caps and an on-chain revoke, a one-shot signed mandate, a pinned policy, a human ARM ceremony — closed EXECUTED at exactly the mandated rate (normalized to 169 micro-USDC from the platform’s float delta). Ten-criteria offline consistency verification with the COSMO verifier, published at /verifier/ (SupraFX Mainnet).

mandate_hash 0xb0d3911a…6bab11

policy_hash 0x59f7c39f…7dd75d

evidence_root 0x5799bf59…1b50c2

Evidence files ↗The framework →

PILOT-001: first job paid on the market

2026-07-172 wCOSMO

The first job on the market that ran from posting to payment. Every step is its own public transaction.

Technical details

First marketplace trade settled end-to-end: escrow, quote, accept, deliver, settle — every step its own mainnet transaction.

spec_hash 0x335f1242…7328be

result_hash 0x96700509…d8040c

Evidence files ↗Job page →Payment transaction ↗

PATCH-001: a software fix, checked automatically

2026-07-10285 wCOSMO

A buyer paid for a software fix. Payment depended on an automatic ten-point check.

Technical details

A software patch fixing a real defect, paid only after a ten-criteria machine acceptance check returned ACCEPT.

input_hash 20eef0578a…6395e5

diff_hash 92b0c80ad1…cfaaa9

result_hash 7a5847bb18…7afdb3

Evidence files ↗Details →Approve tx ↗

ATTEST-001: a signed report, checked automatically

2026-07-08200 wCOSMO

A buyer paid for a signed report on four live facts about the protocol, checked automatically before payment.

Technical details

The first traded good: a signed attestation of live protocol invariants, delivered against a security deposit and machine-accepted before approval.

input_hash 7516be60f3…b2b6ee

result_hash 665404a859…aec205

Evidence files ↗Details →Approve tx ↗

JOB-001: the first compute job

2026-07-06200 wCOSMO

The first compute job: a fixed calculation, paid after the buyer approved the result.

Technical details

The foundation: the first real compute job — a deterministic 1,000,000-step SHA3 workload — settled through the full escrow lifecycle.

input_hash 68747dd41a…4d3a36

result_hash ceb6d41ee6…d47e61

Workload (GitHub) ↗Details →Approve tx ↗

Buyer and provider on JOB-001, ATTEST-001 and PATCH-001 are operating-team accounts, disclosed on their detail pages. PILOT-001 was paid through the public market flow. The runs under fixed rules, including VLM-001, were run by the operating team; their closing records are statements by us about ourselves, and the evidence says so.

Earlier milestones: the first automatic token trade and the full demo round-trip are preserved in the archive.

Honesty principles

These rules apply to every page. They are not centralized here — each page carries its own honesty box next to the claims it qualifies. This section only states the rules.

Facts and plans never mix.

Paid means paid on-chain; planned means planned. Anything that is not live carries a label that says so (tested, experimental, planned), in the same place, not in a footnote.

Applied: the labels Live, Pilot, Tested, Experimental, Planned and Archive across the site.

Every claim links to something you can check.

Numbers on this site lead to a public transaction, a fingerprint recorded on-chain, or a published file whose fingerprint you can recompute yourself.

Applied: the proofs above and the files under /evidence/.

Limits are disclosed next to claims.

Every page carries an honesty box listing what the shown result does not prove — including operating-team involvement where it exists.

Applied: honesty boxes on the market, /compute and /cosmo; the limitations box below.

Evidence is frozen, not curated.

Published files are exact copies of the originals. Their SHA3-256 fingerprints are recorded on-chain, so a changed file would be noticed. Check with openssl dgst -sha3-256 against the list in each folder’s index.txt.

Applied: /evidence/pilot-001/, /evidence/patch-001/, /evidence/attest-001/.

Assurance module 01 — Price Integrity Guard

Verify the data. Then verify the decision.

Research PrototypeRead-onlyNot Live Protection

A valid oracle update can prove authenticity under its verification model. It does not by itself guarantee that acting on the resulting value is economically safe. COSMO independently evaluates whether critical numerical inputs are plausible enough to use.

Case studies →Technical baseline →

What the Guard evaluates

Freshness

Was the value recent enough to use?

Deviation

How far does it differ from an independent reference?

Magnitude

Is the value within an economically plausible range?

Evidence coverage

Was the required input actually observed?

The principle

No evidence is not zero.
No evidence is UNKNOWN.

COSMO does not issue a clean result for a check that had no supporting input.

Results

SAFE

The evaluated checks stayed within the registered policy boundaries.

WARN

A relevant limitation, uncertainty or material deviation was found.

HALT_RECOMMENDED

The observed value was economically implausible enough that dependent actions should be stopped or independently reviewed.

These are read-only recommendations. COSMO does not currently pause or control the evaluated protocols.

Case studies

Bonzo SAUCE/wHBAR exploit replay

Retrospective detectionNot prevention

COSMO reconstructed the manipulated SAUCE/wHBAR oracle value from frozen public evidence.

The Guard produced HALT_RECOMMENDED at the first manipulated submission in the reconstructed timeline, before the first observed borrowing activity.

The manipulated value differed from the market reference by approximately 12.7 orders of magnitude.

The replay proves what the Guard would have recommended from the frozen evidence. It does not claim that COSMO was operating during the incident.

Solido collateral snapshot

Single read-only snapshotNot continuous monitoring

COSMO compared Solido's SUPRA collateral value with an independent market reference and separately checked the internal stSUPRA-to-SUPRA conversion relationship against the Flow vault share rate.

The observed price relationships remained within the registered limits.

Feed freshness and submission provenance could not be established from the frozen snapshot and were therefore reported as UNKNOWN, never silently treated as zero or passed.

Only the SUPRA market comparison used an independent external reference. The stSUPRA check was an internal consistency check, not an independent oracle validation.

Reusable logic. Explicit integrations.

The evaluation engine is chain-agnostic and oracle-agnostic. Each concrete integration remains explicitly registered, tested and reviewed.

Adding an integration is not a configuration change. It requires new code, new tests, a separately pinned policy and a review.

Technical baseline
Tag
price-guard-subject-registry-v2
Baseline commit
29f0044
Tests
176 offline tests
Test environment
Tests run without network access or a production signing key
V1
V1 remains byte-frozen at commit 4bbf3d6
V2
V2 separates evidence mode from registered subject identity
Gates
Security gates are covered by mutation tests
Failure mode
Policies, subjects and allowed mode/subject combinations fail closed

Provenance, honestly: these identifiers reference a private repository, so you cannot verify them from this page today. Verification materials are available on request; publishing the repository is under consideration.

A guard that no test holds is not a guard.

Comments claim coverage. Mutations demonstrate it.

Checking the data is one thing. Checking the decision is the next.

COSMO publishes what happened instead of forecasts. The research on this page asks how protocols and agents can check not only whether data is real, but whether acting on it is safe.

How earning works

Static research content. No wallet actions and no on-chain interaction on this page. The proofs link to public transactions and published files. The Price Integrity Guard is a research tool: it reads public data and published evidence and only makes recommendations.