Cowrie Research

threshold-ceremony-proof-completeness-and-fresh-signer-exposure-discipline

14 min read

Threshold-Ceremony Proof Completeness and Fresh-Signer Exposure Discipline

  • Name: threshold-ceremony proof completeness and fresh-signer exposure discipline

  • Category: custody-design reference / MPC-TSS ceremony hardening note / signer-admission blast-radius discipline / positive-design companion to threshold-share leakage

  • Summary: Healthy threshold-signing systems do not stop at the marketing statement M-of-N MPC. They maintain one current list of which participant-supplied cryptographic objects must be proven well formed, one explicit policy for how quickly newly admitted signers may accumulate transcript exposure to high-value vaults, and one bounded-loss design that still matters if the ceremony layer fails anyway. The design target behind this note is the positive counterpart to threshold-ceremony share leakage: a malicious signer should not be able to turn ordinary participation in keygen or signing rounds into long-term share extraction, and fresh admission into the quorum should not immediately buy enough high-value transcript access to make a latent ceremony bug catastrophic.

  • Why this note exists:

    • threshold-ceremony-share-leakage preserves the negative pattern where the threshold breaks from the inside because participant-supplied ceremony material was not constrained tightly enough.
    • The sharper design-side question is different: what exact proof set, signer-admission policy, transcript-exposure budget, and vault blast-radius control should have existed so one malicious participant could not graduate from routine ceremony participation to unilateral custody authority?
    • THORChain’s 2026 Asgard-vault drain is a good anchor because the public lesson was not merely GG20 had risk. It was that a newly churned signer appears to have been able to join a live high-value vault, participate in routine ceremonies for roughly two days, and extract enough long-term share material to sign raw withdrawals directly.
    • The Fireblocks GG18 / GG20 disclosure sharpened the protocol lesson that missing Paillier modulus and biprime-soundness checks are not cryptography trivia. They are custody-boundary conditions.

Three ceremony-hardening promotions healthy design should keep separate

  1. A participant may still enter enough ceremonies to matter
  • Healthy systems treat signer admission and transcript exposure as a real control surface rather than as a harmless byproduct of being added to the set.
  • The first design question is reachability: can one fresh or still-live participant accumulate enough observation and influence for a latent leakage bug to become worthwhile?
  1. Those ceremonies may still be promoted into canonical threshold truth
  • Production code still needs to prove that participant-supplied Paillier setup, modulus structure, proofs, nonce commitments, transcript rules, and abort semantics satisfy the malicious-participant assumptions the scheme depends on.
  • The second design question is truth promotion: what exactly lets the system keep treating these rounds as sound threshold truth rather than attacker-shaped leakage?
  1. That accepted key may still carry live reserve authority
  • Even after a ceremony family is judged sound, the resulting vault key should not automatically inherit ordinary-speed authority over the deepest reserves in the system.
  • The third design question is authority budget: if this ceremony family turns out to be weaker than believed, how much release power still survives behind the key it produced?

Healthy design is stronger when those three promotions are staged explicitly instead of collapsing into one marketing sentence like we use MPC.

The four things healthy ceremony discipline should keep explicit at all times

1. Exact proof-completeness inventory

  • The system should be able to enumerate every participant-supplied object that must be proven well formed for malicious-participant security: Paillier modulus structure, factor / biprime soundness, range proofs, nonce commitments, transcript consistency, identifiable-abort semantics, and any scheme-specific bounds.
  • We use GG20, we use threshold ECDSA, or the library handles that is not an inventory.
  • The useful design target is a table that maps each proof obligation to the code path that enforces it, the upstream advisory that motivated it, and the deployed artifact version that currently satisfies it.

2. Progressive signer exposure

  • Freshly admitted signers should not immediately gain unlimited access to the system’s richest ceremony surface.
  • A healthy system treats new participant can observe and influence many high-value ceremonies as its own control surface, distinct from economic bond or governance approval.
  • Signer probation can be implemented through staged vault assignment, smaller vault caps, delayed access to large reserve pools, or explicit transcript-exposure ceilings before a signer can participate in the most security-critical ceremonies.

3. Ceremony telemetry as security telemetry

  • Retries, aborts, churn timing, failed rounds, signer-specific anomalies, and unusual ceremony concentration are not mere reliability metrics.
  • If a protocol only notices these as ops noise, it can miss the extraction setup phase of a malicious-participant attack.
  • A healthy system preserves enough telemetry to answer which participant supplied which cryptographic object in which round, and how much high-value ceremony exposure did that participant accumulate before we intervened?

4. Bounded-loss custody even after cryptographic failure

  • A recovered full key should still face meaningful release limits.
  • Vault segmentation, per-vault caps, fast halts, staged release, route-specific throttles, and reserve compartmentalization are the design-side admission that cryptographic assumptions can fail operationally.
  • This note is not satisfied by perfect ceremony validation alone. It also asks whether the balance sheet remains survivable when one ceremony family turns out to be weaker than believed.

What sound design looks like

  • Healthy systems keep three promotions separate on purpose: a participant can still accumulate meaningful ceremony exposure, those ceremonies still qualify as sound threshold truth, and the resulting key still deserves this much reserve authority. The fastest way to hide this failure class is to narrate all three as one vague the MPC layer is secure claim.

1. One maintained ceremony-assumption ledger

  • The protocol keeps an explicit ledger of what its threshold scheme assumes to remain secure in production.
  • That ledger covers scheme choice, upstream library lineage, local forks and patches, proof obligations, identifiable-abort behavior, accepted parameter ranges, and any known advisories that changed the mandatory validation set.
  • Operations, audits, and incident response should all point to the same ledger rather than to folklore about the scheme.

2. Fresh signers enter through lower-blast-radius lanes first

  • New signers, validators, or MPC participants do not immediately inherit system-wide high-value custody exposure.
  • They begin with bounded vaults, probationary assignment, reduced transaction ceilings, or lower-value domains until the system has earned confidence that both the operator and the ceremony implementation behave as expected.
  • The design goal is not only social trust-building. It is to shorten the maximum damage from an undiscovered malicious-participant bug.

3. Ceremony-critical validation fails closed

  • If a participant-supplied proof is absent, malformed, unverifiable, version-mismatched, or newly called into doubt by an upstream advisory, the system should stop using that ceremony path until it is requalified.
  • Healthy implementations do not continue high-value signing because the rest of the round looked fine or because operators believe the library is probably unaffected.
  • Ceremony validation should be treated like signature verification, not like optional hygiene.

4. Exposure budgets are explicit and enforceable

  • A protocol should know how many rounds, how much value, how many vaults, and how much time a fresh participant can influence before rotation or escalation is required.
  • Those budgets should be machine-enforced where possible rather than left to ad hoc operator vigilance.
  • The useful design question is how much transcript exposure can one newly admitted or recently reconfigured participant accumulate before we narrow or remove that lane?

5. Rotation and eviction are rehearsed operational controls

  • If a signer, vault, or library family becomes suspect, the protocol must be able to freeze, evict, rotate, and reassign without inventing the process mid-incident.
  • Cryptographic safety is partly an operations problem: the best ceremony assumptions still need rapid withdrawal-cap reduction, signer quarantine, and vault migration when confidence changes.

6. One recovered key is not equivalent to unlimited immediate release

  • Even if the worst-case cryptographic assumption fails, the system should still preserve asset segmentation and release friction.
  • Healthy systems pair ceremony hardening with bounded-release design, especially when one vault key covers many chains or asset domains.

Useful positive design shapes

1. Proof-obligation matrix owned like production infra

  • Maintain a living matrix of:
    • participant-supplied object
    • required security property
    • proof or check enforcing it
    • code path / artifact version
    • upstream advisory / paper supporting the requirement
    • monitoring that would reveal unexpected failures
  • This converts abstract cryptographic assumptions into something architecture review and operations can actually own.

2. Fresh-signer probation tiers

  • Introduce signer tiers such as new, probationary, and full with progressively broader ceremony and vault access.
  • Probationary tiers can be tied to smaller vaults, fewer chains, lower transaction ceilings, longer observation windows, or manual co-approval.
  • The goal is not to claim humans can manually detect every malicious signer. It is to deny a fresh participant immediate access to the system’s highest-value transcript surface.

3. Ceremony-family kill switches

  • A production system should be able to disable one signing family, library branch, vault cohort, or chain route when ceremony assumptions become suspect.
  • Pause everything is sometimes necessary, but healthy design also wants narrower containment lanes that can localize risk before the full system halts.

4. Transcript-exposure accounting

  • Count exposure in economically meaningful terms: number of ceremonies, cumulative value signed, number of vault families touched, and elapsed time since signer admission or software upgrade.
  • The useful design target is not merely who is online but who has had enough access to make a latent extraction bug dangerous?

5. Bound vault value to trust freshness

  • Newly formed quorums, newly rotated signers, or recently patched cryptographic stacks should often start with smaller bounded custody until their assumptions have survived some time in production.
  • This is the custody-side analogue of canary deployment.

Design questions worth asking first

  • Which participant-supplied cryptographic objects must be proven well formed for this threshold scheme to resist malicious participants, and where exactly are those proofs enforced in deployed code?
  • Is the production stack running a maintained upstream release, a lagging fork, or a bespoke adaptation, and who owns tracking new cryptographic advisories for that exact artifact family?
  • How much high-value ceremony exposure can a newly admitted signer accumulate before additional controls, rotation, or narrower assignment kick in?
  • If one signer or one scheme family becomes suspect today, what exact operational switch can freeze or narrow only the affected custody lanes?
  • If one vault key is reconstructed offline anyway, what release caps, segmentation, or rapid halts still matter before system-wide value leaves custody?

Red flags that usually mean the ceremony boundary is being treated too casually

  • The protocol is secure because it is MPC with no proof-obligation inventory
  • fresh signers immediately assigned to the richest vaults or broadest chain set
  • cryptographic libraries forked long ago with no explicit advisory ownership
  • ceremony retries and aborts tracked only as reliability issues, not as possible malicious-participant signals
  • one recovered vault key still able to command multi-chain or system-wide reserves immediately
  • no rehearsed path to evict a signer or quarantine one ceremony family without improvising under live attack
  • operator docs focused on threshold counts while saying little about transcript exposure and signer-admission blast radius

Failure-prevention checklist

  • Review three separate promotions every time: participant can still accumulate enough ceremony exposure, those ceremonies still count as sound threshold truth, and the resulting key still deserves this much live reserve authority. Healthy ceremony design proves all three instead of assuming one implies the next.
  • Inventory every threshold-signing dependency, fork, local patch, and scheme-specific validation requirement used in production.
  • For each participant-supplied object, document the exact well-formedness property required for malicious-participant security and the code path that enforces it.
  • Track upstream advisories, CVEs, and research for the exact protocol and artifact family in production; assign an explicit owner for acting on them.
  • Fail closed when proof verification is missing, downgraded, or uncertain; do not keep high-value signing live under ambiguous ceremony assumptions.
  • Stage fresh signers into bounded vaults, bounded chain sets, or bounded transaction ceilings before they accumulate broad transcript exposure.
  • Preserve signer-level telemetry for keygen, signing rounds, retries, aborts, and unusual ceremony concentration; treat this as security data, not only ops data.
  • Define exposure budgets for newly admitted signers and newly upgraded ceremony stacks in terms of value, rounds, and time.
  • Rehearse signer eviction, vault rotation, ceremony-family quarantine, and lower-cap degraded modes before an incident demands them.
  • Segment reserves so one recovered key cannot immediately drain the full cross-chain balance sheet.
  • Tie solvency halts, route throttles, and emergency pauses to the possibility that authentic-looking signatures may already be maliciously generated.

Audit prompts

  • What exact proof families keep a malicious participant from learning honest long-term shares in this deployment?
  • Which of those checks are scheme-mandated, which are advisory-driven, and which exist only in some upstream branches but not in the deployed fork?
  • Can a newly admitted signer participate in enough high-value ceremonies within days to make an undiscovered extraction bug immediately worthwhile?
  • What telemetry would distinguish routine ceremony instability from deliberate leakage harvesting or attack setup?
  • How quickly can the system quarantine one signer, one vault cohort, or one scheme family without freezing all custody indefinitely?
  • If a recovered key bypasses threshold participation entirely, what downstream controls still slow release or localize loss?
  • Does the public trust model say anything meaningful about signer-admission blast radius and cryptographic artifact freshness, or only about nominal threshold counts?
  • Are operators prepared to explain which assumptions changed after the latest advisory, patch, or signer-rotation event?

Lessons that should already be standard

  • Participant can still join ceremonies, those ceremonies still produce sound threshold truth, and the resulting key still deserves deep release authority are different properties.
  • M-of-N is a partial sentence. The missing half is under which validated ceremony assumptions and with what fresh-signer blast radius.
  • Missing proof verification in threshold cryptography is not a minor implementation defect; it is direct custody risk.
  • Signer admission is a security boundary, not only a governance or bonding milestone.
  • Fresh participants should not immediately receive the same transcript access as long-established signers when the ceremony family guards concentrated value.
  • Ceremony retries, aborts, and churn patterns deserve security review because extraction attacks often begin by looking like ordinary protocol participation.
  • The system should assume that cryptographic families can be wrong in production and keep the balance sheet survivable when that happens.
  • A protocol that can rotate or quarantine quickly after suspicion is safer than one that merely advertises a high threshold during steady state.

Negative exemplar cluster: what missing ceremony discipline looks like

  • THORChain Asgard vault GG20 key-extraction drain is the sharpest current corpus anchor: a newly churned participant appears to have been able to join one live vault, accumulate routine ceremony exposure, reconstruct the full key, and then bypass the threshold flow entirely with direct withdrawals.
  • threshold-ceremony-share-leakage preserves the recurring exploit-side family behind that outcome.
  • The Fireblocks GG18 / GG20 disclosure and CVE-2023-33241 are the broader ecosystem reminder that malformed Paillier modulus validation was not a THORChain-specific curiosity; it was a family-wide custody break if implementations lagged or skipped the right checks.

How to use this note with the exploit corpus

  • Start here when the sharper design question is what should a healthy threshold-signing system do so one participant cannot turn ordinary ceremony participation into full-key custody authority?
  • Pair with threshold-ceremony-share-leakage when the exploit-side question is still about the recurring leakage bug class.
  • Pair with operator-topology-map-and-effective-threshold-discipline when the trust-model question expands from cryptographic ceremony assumptions into the broader operator graph above the signers.
  • Pair with bridge-bounded-release-design when the decisive follow-up is why one recovered vault key still implied too much immediately spendable custody.
  • Pair with signer-quorum-compromise when the protocol-level framing is still how did the threshold collapse in practice?

Internal linkages

Control surface

Graph View

Backlinks