Why OC Pledge

The other five OrangeCheck primitives all describe present-state:

OC Attest — I have stake now.
OC Lock — I can receive now.
OC Vote — I chose now.
OC Stamp — I said this now.
OC Agent — I authorize now.

None of them project commitment forward. None lets a sovereign actor stake their word against a future-verifiable proposition with a bond that has teeth through public record, not custody.

That gap is what the modern web fills with captured intermediaries: notaries, escrow agents, SLA attorneys, auditors, credit agencies, rating firms, preregistration registries, delivery-tracking vendors, review platforms, contract-management SaaS, insurance underwriters. Each is solving a variant of "someone staked their word on a future claim, here's the mechanism to check later."

Pledge is the sovereign version. It does not replace every function those intermediaries serve — it does not enforce physical-world obligations through state power — but for any commitment whose outcome is computable from public state (chain state, Nostr events, HTTP responses, DNS records, OC Stamp envelopes, OC Vote outcomes, counterparty signatures), it replaces the intermediary with a protocol.

Why this is Bitcoin-load-bearing

Standard test: if this works identically on Ed25519, it doesn't belong here. Pledge survives the test on three grounds simultaneously:

Universal stake oracle. The bond must be real and universally observable. OrangeCheck's sats × days is the only credibly-neutral economic signal on the open internet. Ed25519 has no economic layer.
Non-malleable time. Resolution deadlines expressed as block heights use Bitcoin block headers as the neutral time primitive. No other chain is credibly neutral over long horizons; L2s and altcoins introduce operator trust or reorg risk.
Universally-observable identity namespace. The public ledger of kept-and-broken pledges attaches to Bitcoin addresses, which are a globally enumerable namespace with no registrar. Pseudonymous Ed25519 keys have no economic or temporal grounding to attach that ledger to.

Strip any of the three and the primitive degrades to a known, commoditized, already-failed design — PGP, x509, OAuth claims, EAS attestations. Keep all three and it's a thing no other stack can ship without rebuilding the first five primitives first.

What we explicitly refused

The full argument for each refusal lives in WHY.md; the short version:

No self_proof resolution mechanism. Pledges whose truth cannot be checked from public state are not pledges — they are claims. The protocol refuses them at envelope-construction time.
No slashing. There is no mechanism by which a protocol actor seizes bonded sats. Ever. The bond is an economic signal; its only function is to make the attached public record meaningful. A slashing mechanism would require custody, which the entire OrangeCheck family refuses.
No aggregated reputation score. Same discipline OrangeCheck follows with sats_bonded and days_unspent vs. score_v0. The protocol ships raw metrics (pledgesSworn, outcomesFor, bondInFlight) and refuses to ship a reference score at all. There is no natural scoring function that doesn't embed strong policy opinions; any shipped score will be misused as canonical.
No revocation. A pledge you can walk back is not a pledge. A swearer can publish an abandonment envelope, but it counts as broken. No "honorable exit" — preempts the race-to-abandon attack where swearers dodge a foreseeable break by abandoning early.
No pledge chains in v0.1. The pattern of "this pledge's resolution triggers that pledge's resolution" is interesting and we expect to ship it post-v1 once we observe what real-world patterns demand it.
No bonded voter sets for dispute resolution in v0.1. v0.1 ships with a named address list for vote_resolves. Bonded voters is a v0.2 enhancement once real dispute volume exists.

What we kept from each adjacent system

From DocuSign / contract-management SaaS: the user-facing artifact — a single document with two signatures and a public timestamp — but reframed as a self-contained envelope verifiable from any web page, not a vendor's PDF render.
From smart-contract escrow: the deterministic-resolution discipline. Pledge mechanisms must be pure functions of public state. We dropped the custody.
From prediction markets: the bilateral-counterparty pattern. Two swearers can make opposing pledges with counterparty_signs resolution and treat the outcome as a wager. The protocol does not route payouts — parties settle out-of-band.
From academic preregistration registries: the "publish-by-deadline" shape. Pledge handles this via stamp_published resolution. No registrar to capture the namespace.
From SLA contracts: the periodic-acknowledgement shape via counterparty_signs outcomes signed by the customer at period end.
From OrangeCheck itself: the raw-metrics-over-aggregated-scores discipline, the no-custody-ever posture, the canonical-message format, and the offline-verifiability goal.

Closing note

Pledge is the sixth and final primitive. The family is now closed: six verbs of sovereign sociality (am, whisper, decide, declare, delegate, swear), each a present- or future-tense statement a Bitcoin address can make to the open internet, each verifiable from public state alone, none custodial.

If you find a seventh verb that meets all three Bitcoin-load-bearing conditions and the no-custody discipline, open an issue. We expect that list to stay at six.