private proof-layer workspace · handshake × hive · not indexed · complement, not competitor
Handshake measures.
Hive proves.

Your experts build the evals. Your simulator tests every change. Your loop feeds corrections back. Hive re-measures none of it. We turn each eval, correction, and verifier decision into an independently signed receipt. Your strategic IP stays provable long after the run, and we never see the words.

~$1B/yrAI training run-rate · source-reported
Evals are the new PRDstrategic IP needs an independent record
Flight simulatorevery change tested · every test receiptable
Cleanlab + Uplimitwider surface · more places proof earns its keep
watch it · thirty seconds
A score in a dashboard is a memory.
A score in a receipt is evidence.

Six months from now, a lab reviewer asks: which rubric version scored this, against which dataset? Watch what happens when the dashboard answer gets quietly changed, and the receipt does not.

The dashboard
lives where anyone with access can reach it

The Hive receipt
signed at run time · kept out of reach

Your evals are strategic IP. A dashboard cannot defend IP. A receipt can. Signed with ML-DSA-65.
carnac live ink™ · watch the proof form as it is written
Your expert writes the eval case.
The proof forms in the text, live.

Your experts save enterprises thousands of hours taking the first pass on the eval set. As each case is written, Carnac Live Ink™ marks it in the moment. The record is built while the work happens, not rebuilt after. Each span gets the route it earns.

routine · light record review · more evidence bound in hold · needs a human first
The eval set arrives already carrying its own proof of who wrote what, and when.
carnac™ · tap an eval, see how much proof it earns
Not every change needs the same proof.
Carnac™ sizes it before Hive signs.

Carnac™ reads the change as it is written and decides how much proof it deserves. A small tweak is a light record. A change that touches a regulated decision gets more. The weighting behind that choice is private. The route it picked is recorded so anyone can re-check it.

Routinelight record
Reviewmore evidence
Holdhuman first
Abstainout of scope
Carnac™ determines required proof and records provenance. It does not judge. Handshake owns the score. Carnac™ never re-scores.
Two planes, kept apart on purpose. That separation is what makes the proof worth anything to a buyer.
originproof · flip the conditions yourself
Is this really human work?
Now there is a receipt for that.

The risk in expert data is undisclosed model substitution, work sold as human where the AI use was never disclosed. OriginProof signs the conditions of production. Flip a condition off and watch the receipt refuse to overclaim. That honesty is exactly why buyers believe it.

Conditions, never a route. It never claims more than it can prove.
the receipt journey · auto-running
From eval run to audit export.
Six stops. Every one signed.
Eval runexperts + rubric
Gandalfverifier decides
Attestationrubric ver + dataset hash
Ship itsimulator says go
Runtimeevery live call receipted
Auditone bundle · checks offline
the measurement is Handshake. the signature, portability, and export are Hive.
Primitives on the rail: AFiR · AFiR-Stream · SiGR · R3Pv™ · Receipt Relay · Hive Ledger
r3pv™ · honest by design · tap the weak one
A group is only as proven
as its weakest receipt.

Five receipts, grouped into one proof with R3Pv™. Four are independently verified. One is only self-attested, another system word, signed as stated. Watch what the group badge honestly says, then tap the weak one to upgrade it.

eval ✓ chain-verified session ✓ relay-observed score ⚠ self-attested · tap to verify output ✓ relay-observed export ✓ chain-verified
GROUP BADGE · SELF-ATTESTED · the weakest link, stated out loud
a mostly-verified group is never sold as fully verified · that honesty is what makes the badge worth exporting
what could start monday · pick one
Four pilots. Each one artifact
you can hold and re-check.

These are ready to build, not live customers. Pick the one that maps to your nearest deadline and we scope it.

A · Gandalf receiptOne verifier decision, signed with SiGR: identity, criteria, evidence, route. Checks offline.
B · BankerToolBench bundleOne task deliverable plus score, sealed audit-grade for a regulated reviewer.
C · Runtime call receiptOne live production call, independently receipted with AFiR and tied to its pre-launch eval.
D · Voice segment receiptA conversational agent with every turn signed at the segment level with AFiR-Stream.
# try the live rail right now · no key · no install curl -sS -X POST https://hive-ask.onrender.com/v1/ask \ -H 'Content-Type: application/json' \ -d '{"prompt":"score eval #4471","options":{"receipt":true}}'
act ii · where the same layer goes next · options, not promises
Independence cannot be self-supplied.
That is what makes these hard to copy.
Handshake measures. Hive proves.
"Trust our experts" becomes "verify our experts."
Private by design. Hive does not store your prompts or your eval content. Every request is receipted by a one-way SHA-256 fingerprint, never the words. Proof, not surveillance.
Private proof-layer workspace for Handshake · noindex / nofollow / noarchive · not in nav, sitemap, or footer · scale and quote figures are source-reported by Handshake and the press, not asserted here · OriginProof attests conditions, never a route: a human-conditioned process with disclosed model-use boundaries, not "provably human output" · evidence tiers are stated honestly: self-attested, relay-observed, chain-verified · Hive never sees eval content, one-way fingerprints only · demonstrations illustrate the live rail at hive-ask.onrender.com · all Hive primitives are patent pending · sources: joinhandshake.com/ai · the right model for the right work · Carnac™, CarnacPrompt™, Carnac Gateway™, and Carnac Live Ink™ are Hive marks · Hive Civilization.