@forgesworn/moneyer 0.1.1 → 0.2.0

package/CHANGELOG.md

@@ -1,5 +1,191 @@
 # Changelog
 
+## [0.2.0] - 2026-08-22
+
+While LUD-25 is a draft, a `0.x` minor bump may be breaking; this one is
+additive on the wire apart from the node-capacity rename below.
+
+- The test suite has a 30 second timeout rather than vitest's 5 second
+  default. Nearly every test here starts an HTTP server, opens a database
+  and does real curve and KDF work; 5 seconds is a unit test's budget, and
+  on a loaded machine the slower cases crossed it while passing perfectly
+  well. A gate that fails at random teaches people to bypass it.
+- **`moneyer --dev` is a mint a wallet can actually use.** The fake
+  funding source settled nothing on its own: settlement was reachable only
+  through an in-process handle a test holds, so a wallet pointed at a
+  standalone `MONEYER_BACKEND=fake` mint asked for an invoice, nothing ever
+  paid it, no note was ever minted, and every flow needing a note - split,
+  merge, melt, send - dead-ended behind an empty wallet while the mint
+  looked healthy throughout. `--dev` now treats every invoice it issues as
+  paid the moment it is issued, so minting, splitting and melting all work
+  locally against a wallet. It can only be set on the fake backend, which
+  moves no money, and the startup banner says plainly that the notes are
+  worth nothing. `MONEYER_FAKE_AUTOSETTLE=true` sets it explicitly for a
+  compose file that cannot pass a flag.
+- The mint can announce itself. With `MONEYER_ANNOUNCE=true` and the Nostr
+  identity zap-to-note already uses, it publishes its own discovery
+  document hourly as a parameterised replaceable event (kind 30078, `d`
+  tag `lnurlcash-mint`), alongside the liabilities snapshot. Until now a
+  wallet could only learn of a mint by being told its address; this is the
+  half that has to exist before finding one means anything. The content is
+  the document exactly as the discovery endpoint serves it, so there is one
+  description of a mint and not two, plus a `sig` over the canonicalised
+  body made with the note signing key, so a holder can check the
+  announcement against the same key their notes verify against.
+  `verifyAnnouncement` is exported for that. Off unless the operator turns
+  it on: a mint that does not want to be listed says nothing. No new event
+  kind, and no recommendations or reviews; both are protocol decisions that
+  belong with the NIPs.
+- A note melts to an invoice that states no amount. Such an invoice used
+  to be refused outright, which is a papercut for anyone paying one: for a
+  bearer note the amount was never in question, since the note's value is
+  the amount. The mint sends the whole-sat floor of the note's value, the
+  same figure it already accepts from a wallet that fills the amount in
+  itself, and keeps the sub-sat remainder as the same dust. A note worth
+  less than a single satoshi has nothing left once the floor applies and
+  is refused with `insufficient value`; it can still be melted by a wallet
+  that invoices its exact value. Everything else about the melt path is
+  untouched, including the pending and restore discipline.
+- A wallet can name the note it is buying. It chooses the note's spend
+  secret itself, keeps it, and sends `h`, the sha256 of that secret, on
+  the pay callback; the mint credits the note at `h` when the invoice
+  settles. The invoice's payment preimage then buys nothing and is only a
+  payment proof, which is what it always should have been: a preimage is
+  known to the funding source, to every node that forwarded the payment,
+  and to anyone who polls LUD-21 `verify` with the payment hash written
+  inside the invoice. Naming the note leaves the buyer as the only party
+  who ever held the secret, and replaces "claim and rotate faster than
+  anybody else" with nothing to race for. A malformed `h` is refused
+  before an invoice is issued, so a wallet never pays for a quote the
+  mint would reject; an `h` that already names a note or an invoice is
+  refused with `Invalid or already spent k1.`, the same oracle-free
+  sentence a colliding output gets on the withdraw callback. The reply
+  carries `mintToHash: true` when the binding was made, and the payRequest
+  and the discovery document advertise `mintToHash: true` so a wallet
+  knows before it asks. Claiming needs no `verify` poll: the wallet asks
+  `/w?k1=<its own secret>` directly. Entirely optional and additive: a
+  wallet that sends no `h` gets exactly today's behaviour, and no existing
+  wallet or service breaks. The draft itself does need one additive
+  amendment, because it says a `SERVICE` MUST accept `k1=P` and a bound
+  mint deliberately does not; that is proposed separately. Wallets are urged to persist the secret before
+  requesting the invoice, which is the one thing that could make this
+  worse than what it replaces.
+- A live note's informational GET now carries `payLink`, pointing at this
+  mint's payRequest, which is the counterpart of the `withdrawLink` a
+  payRequest already advertises. It is the route home for a holder who has
+  nothing but a note: from it a wallet reaches the discovery document, and
+  so the mint's terms and its retired signing keys. Without it a wallet
+  that only ever received notes cannot tell an announced key rotation from
+  a substituted key. Optional and additive; unknown and spent notes still
+  say nothing about the mint beyond the refusal.
+- Both database connections now wait up to five seconds for a lock instead
+  of giving up at once. `node:sqlite` opens with no busy timeout, so a
+  second writer met `database is locked` the moment the first one held it.
+  The operator CLI is that second writer while the mint is running, and the
+  times you reach for `moneyer admin reconcile` are the busy ones. Nothing
+  was ever at risk of corruption; the command simply failed when it was
+  least convenient.
+- The discovery endpoint carries the human layer: `name`, `description`,
+  `contact` (`nostr` as an npub, `email`, `url`), `tosUrl`, `motd`,
+  `fees`, `version` and `previousPubkeys`. Every one is optional and an
+  unset one is absent rather than empty. New environment variables:
+  `MONEYER_NAME`, `MONEYER_CONTACT_NOSTR`, `MONEYER_CONTACT_EMAIL`,
+  `MONEYER_CONTACT_URL`, `MONEYER_TOS_URL`, `MONEYER_MOTD`.
+- **Wire-name change**: node capacity now goes out as `nodeCapacity`, the
+  name the reference mint, the conformance mock and `lnurlcash-kit` all
+  use. `nodeCapacityMsat`, which was moneyer's alone, is emitted alongside
+  it for this one release and then removed. Both are milli-satoshis.
+- The mint's own site and the fallback landing page show the message of
+  the day, the operator's contact and the terms link, and read capacity
+  under either name.
+- `GET /stats` states what the mint owes and what its funding node holds:
+  outstanding liabilities and note count, melts in flight and the age of
+  the oldest, the node's local balance, and the coverage ratio between
+  the two to four decimal places. Cached 30 seconds, never per-note,
+  public by design. `MONEYER_STATS=false` switches it off and
+  `MONEYER_STATS_RATIO_ONLY=true` publishes the ratio alone. Both the
+  mint's site and the fallback landing page carry a coverage row.
+- `MONEYER_STATS_PUBLISH=true` publishes an hourly signed snapshot of
+  those figures to Nostr (kind 30078, `d` tag `lnurlcash-liabilities`),
+  signed with the note signing key so anyone can check the history
+  against the mint's advertised pubkey. `scripts/verify-stats.mjs` does
+  exactly that, and `verifyStatsSnapshot` is exported for wallets.
+- The lnd and cln backends report the node's local channel balance;
+  the fake one reports a configurable balance, one bitcoin by default.
+- **Self-service lightning addresses.** With `MONEYER_NAME_PRICE_MSAT`
+  set, anyone with an npub can claim `name@<host>` by posting to
+  `/names` with a NIP-98 Authorization and a note of this mint. The key
+  that signs the request owns the name; no other identity is accepted.
+  The note is burned and the name recorded in one transaction, so
+  nothing is ever paid for a name somebody else got first. Unset means
+  registration is closed; `0` means free, three names per pubkey.
+  Registered and operator-configured names live in one table and are
+  served by one lookup, and `GET /.well-known/nostr.json?name=` resolves
+  them over NIP-05 as well, one name at a time. Discovery advertises
+  `namePriceMsat` while registration is open.
+- The kind 2525 rumor for a zap-funded note carries the zap request in a
+  `description` tag - the same content the kind 9735 receipt carries - so
+  a wallet can show who zapped and what they wrote without fetching the
+  receipt. Readers take tags by name, so an older one does not notice.
+- `MONEYER_ZAP_NAMES` is now optional: a mint that opens registration can
+  start with no names of its own. `MONEYER_NOSTR_KEY` and
+  `MONEYER_NOSTR_RELAYS` still go together.
+- `moneyer admin names` gains `add <name> <npub>` and `rm <name>`, and
+  `list` shows configured names the running mint has not loaded yet.
+- **A retried rotate, split or merge is answered instead of refused.**
+  These are GETs, and transports retry GETs on their own: the retry
+  arrives byte-identical after the inputs are burned, and the old answer
+  - `Invalid or already spent k1.` - told the wallet to drop the only
+  copy of a secret the mint really had minted a note against. The mint
+  now records which request minted which outputs, and replays the same
+  reply: same signatures, nothing burned, nothing minted, no balance
+  moved. A request counts as the same request when it names the same
+  input notes and asks for the same `h`, `h2` and `amount`; input order
+  does not matter. Anything else naming a burned note is still refused
+  with the same reason string, so no oracle appears, and the melt path is
+  untouched. moneyer's own behaviour: the draft says nothing about
+  retries yet.
+- `moneyer admin <command>`: the operator surface. `status`, `notes`,
+  `note`, `melts`, `reconcile`, `sweep`, `snapshot`, `names list`,
+  `keys rotate` and `verify-note`, all reading the same `MONEYER_*`
+  environment the mint does, and all opening the database read-only
+  unless the command mutates. `snapshot` uses `VACUUM INTO`, so it needs
+  no sqlite3 binary and is safe against a live mint; `keys rotate`
+  generates a key and prints the two environment lines without touching
+  anything.
+- `MONEYER_METRICS=true` serves `GET /metrics` in the OpenMetrics text
+  format. Off by default and never authenticated by the app: restrict the
+  path at the reverse proxy. The README's Operating section names the
+  three alerts worth having.
+- A "What the mint knows" section in the README, on the mint's own site
+  and on the fallback landing page, in the same words: what a LUD-25 mint
+  can see, what it cannot, why the wallet-side mitigations are weak, and
+  why the design was chosen anyway. Blindness is the one ecash property
+  this cannot offer, and a reader should not have to infer that.
+- `MONEYER_PREVIOUS_SIGNING_PUBKEYS` lists the compressed pubkeys this
+  mint signed notes under before, published as `previousPubkeys` on
+  discovery. Rotating the signing key no longer invalidates every
+  outstanding note's signature or trips a pinned wallet's mismatch
+  check. Pubkeys only, validated as points on the curve at startup: the
+  old private keys are not needed to verify and should not stay on the
+  server. The runbook is in the README.
+
+## [0.1.2] - 2026-08-21
+
+- `MONEYER_ROUND_FEE_TO_SAT` ceilings the mint fee to a whole sat, as
+  dni's lnurl-mint does on purpose so the mint is "never short a sat".
+  LUD-25 says nothing about rounding and `lnurlcash-kit`'s `mintFeeBand`
+  accepts both readings, so this is a posture choice rather than a
+  compliance one. **Off by default**: turning it on raises what the mint
+  withholds, which is not a change to make on a redeploy.
+- The mint path called `applyMintFee` directly rather than the fee helper,
+  so it would have ignored that setting entirely - advertising one fee and
+  withholding another. Every fee site now goes through one
+  `netAfterMintFee`.
+- `roundFeeToSat` is optional on `MoneyerConfig`: that type is public API,
+  and an embedder constructing a config should not have to name a field it
+  does not care about.
+
 ## [0.1.1] - 2026-08-21
 
 - `withdrawLink` is now the plain URL (`https://host/w`), the form