Karma

Karma is the accountability layer for delegated governance. In a large token-weighted DAO most holders never vote directly; they delegate their voting power to a smaller set of active delegates, and the obvious question – are those delegates actually showing up? – has no answer in the raw on-chain record. Karma builds per-DAO dashboards that answer it: it reads each delegate's on-chain voting, Snapshot (off-chain) voting and forum participation, folds them into a single reputation score, and publishes a public delegate feed so token holders can see who is engaged before handing over their votes.

What it measures

For each delegate Karma tracks the metrics that reveal whether delegated power is being used: on-chain voting participation (share of executable proposals voted on), off-chain voting participation on Snapshot, and forum activity on Discourse – proposals authored, comments, and the health of that participation. These roll up into a per-DAO reputation score and an auto-generated delegate profile that reads like a governance resume, spanning every DAO the same wallet is active in. The weightings are set with each DAO rather than fixed: in the ENS configuration, for example, authoring a forum proposal carries far more weight than casting an off-chain vote, and the weights and scoring logic are open source so the number is auditable rather than a black box.

Why delegate accountability matters

Delegation is what makes token-weighted voting workable at scale – but it also creates a principal-agent gap. A holder who delegates and stops paying attention has no cheap way to notice that their delegate went quiet, switched interests, or started rubber-stamping. Low, concentrated participation is one of the most common ways a DAO decays (see how DAOs fail), and it is invisible until someone measures it. Karma makes the measurement standing infrastructure: DAOs surface it to re-delegate away from absent delegates, and platforms consume it directly – Tally integrates Karma's delegate score alongside a delegate's own voting percentages so the accountability signal sits next to the voting interface itself.

Limits of a reputation score

A participation score measures activity, not judgement. A delegate can vote on everything and vote badly, or author many forum posts that add little; turning “showed up” into a headline number risks rewarding volume over quality, and any single weighting encodes a contestable view of what good governance looks like. Karma’s answer is transparency rather than a claim of objectivity – the weights are published and set per DAO – but the score is a prompt to look, not a verdict. It tells a holder which delegates to scrutinise; it cannot tell them which delegate is right. Karma has also extended the same accountability pattern beyond delegates into grant reporting via its Grantee Accountability Protocol, tracking whether funded programs deliver against their milestones – the line of work that has since become the company’s main product.

Where the company went

Karma the tool and Karma the company have come apart, and the split is recent enough that most write-ups of the delegate dashboards have not caught it. Read on 4 September 2026, www.karmahq.xyz no longer serves a site: it answers with a redirect to www.karmahq.org, whose front door describes the product as “agentic funding software for foundations and donor advisors” – run, in its own framing, from ChatGPT or Claude.

Neither “DAO” nor “delegate” appears anywhere in that home page; the customers it names are UNICEF, the San Francisco Foundation, Regen Coordination and the Trinity Center alongside the crypto grant programs. The word survives one level down, on a page index whose Communities entry offers to “explore DAOs, protocols, and organizations running funding programs on Karma” – DAOs as grant-makers, not as electorates. The governance product is reachable from the company site only as a Governance link in the footer.

The dashboards themselves are still up and still maintained, which is the part a DAO relying on them needs to know. gov.karmahq.xyz serves the same delegate and contributor toolkit it always did, and on GitHub the Discourse plugin that renders governance stats inside ENS’s forum was pushed on 4 September 2026. But the cadence has tilted: the whitelabel delegate app was last pushed on 7 August 2026, against near-daily commits on the grant-accountability repositories. The Grantee Accountability Protocol’s own domain, gap.karmahq.xyz, redirects into karmahq.org too; only the documentation still uses the name.

What the grants product does

The company's own blog traces the shift. Karma GAP began as milestone reporting: a grantee posts updates against the milestones a programme funded, and the record is public. In March 2025 Karma added impact measurement built on the Common Approach, a reporting standard from the traditional grants world, with a chain of funding, activities, outputs, outcomes and indicators; for software projects it pulls the outputs automatically from GitHub, Dune or Open Source Observer (Karma).

In August 2025 it added the rest of the lifecycle, with Optimism's Grants Council as the first customer for direct funding, and in November 2025 it dropped GAP from the name and relaunched as a single platform: application intake, AI scoring of applications, the registry, impact measurement, and payouts released when milestones are met (Karma; Karma). Programmes can run it under their own domain: Celo Public Goods, Regen Coordination and Optimism were the first on the shared whitelabel (Karma). Polygon, Giveth, Livepeer, GoodDollar and Protocol Labs' Filecoin public-goods programme at filpgf.io followed (Karma blog).

Two 2026 moves take it further from DAO governance. In March Karma published agent skills so an agent can find funding opportunities, create and update a project and post milestones without opening the site (Karma). In June it indexed ten years of US Form 990 filings, the annual returns every American nonprofit and foundation files, into a plain-language search that also answers over MCP inside ChatGPT or Claude, and in July it launched a research agent for donor advisors that checks a charity's 501(c)(3) status against the IRS list, reads its recent 990s and state registrations, and ranks candidates against weights the advisor can change (Karma; Karma). Its July development log counts 55 of the month's updates as Filecoin work and three as platform initiatives, one of them on-chain milestone attestations (Karma).

Nothing here is a failure, and the delegate scores have not gone anywhere. It is worth recording precisely because of what it says about the layer: measuring whether delegates show up is a service somebody else owns and can reprice, rebrand or retire, and the accountability signal a DAO leans on is only as durable as its vendor’s interest in the market. Karma found a larger one.

How Caper approaches this

Karma exists because influence and participation are separated: voting power is delegated to agents, and a whole tooling layer is needed to check whether those agents stay active. A caper narrows that gap by construction. Voting weight is not a static balance you hand to a delegate – it is your holdings multiplied by the votes you have actually accumulated over the pool, so sustained participation is priced into the weight rather than measured after the fact by an external scorecard.

A holder who stops engaging simply carries less weight; there is no idle-delegate class to police. And a holder who is persistently outvoted is not stuck monitoring representatives to protect their stake – the vote and the exit right are the same right, so disengagement resolves by leaving with a fair share rather than by a reputation dashboard flagging a no-show. Karma is the right tool for the delegated model; a caper aims not to need one.

References

• Karma, How reputation scoring works and governance dashboards.
• Optimism Collective governance forum, “Karma – Dashboard showing delegate reputation” (June 2022, introducing the first Karma dashboard).
• Arbitrum DAO forum, “Tally integrates Karma’s delegate score and contributor metrics”.
• Karma, Analytics Suite for Optimism and for ENS (per-DAO weightings).
• Karma, company home and llms.txt (the 2026 positioning) · Grantee Accountability Protocol documentation · github.com/show-karma (repository push dates).
• Karma, blog: impact measurement (March 2025), Optimism Season 8 (August 2025), Karma V2 (November 2025), Karma Skills (March 2026) and philanthropic funding (June 2026).