diff --git a/docs/launch-checklist.md b/docs/launch-checklist.md index 7cba257..cb0aeb3 100644 --- a/docs/launch-checklist.md +++ b/docs/launch-checklist.md @@ -4,21 +4,19 @@ Sequenced by gate, not by date. The wider plan, adoption included, is the [roadm ## Where this stands -As of 15 November 2026. The SVR token is on Arc mainnet; the registries and the oracle run on Arc testnet only. What runs where is in the +As of 16 November 2026. The SVR token is on Arc mainnet; the registries and the oracle run on Arc testnet only. What runs where is in the [runbook](operations-runbook.md). There has been no external audit, and nobody outside the team gates on a score yet. **In progress, with dates.** - The adversarial slash drill: day 7 done on 10 November, day 21 from 24 November 03:45 UTC ([slash-drill.md](slash-drill.md), [slash-drill-log.md](slash-drill-log.md)). - The slash keepers went live on 5 November. Both hold C, and neither has yet executed a real slash. -- The publisher has not yet posted through the policy signer's new rule, which requires the feedback hash - to match its file. Three earlier real posts all pass it. **Open before mainnet.** - An external audit, and fixing what it finds (Phase 2). -- The staking half of the contract upgrade batch, on testnet after the drill: a cap on the challenge period - below the dispute period, refusing the admin as a slash's victim and a zero-amount withdrawal. The - reputation and oracle bond halves went in on 5 November. +- The staking upgrade, on testnet after the drill: a cap on the challenge period below the dispute period, + refusing the admin as a slash's victim and a zero-amount withdrawal, and a bound on slash evidence. A new + Reader follows it. The other contracts are upgraded, and [arc.md](arc.md) names each build. - The handover: moving testnet roles to a Safe and a timelock, after the drill, since `executeSlash` needs the reporter still on the committee. - A review of retention by counsel, before public outcome retrieval and peer import on mainnet. diff --git a/docs/operations-runbook.md b/docs/operations-runbook.md index 0f5e35f..c0e55dd 100644 --- a/docs/operations-runbook.md +++ b/docs/operations-runbook.md @@ -16,16 +16,16 @@ key is [incident-operator-key.md](incident-operator-key.md); bringing up an oper | State volume | `sigvara-oracle_oracle_state` | `sigvara-checker_checker_state` | | Local port, for `/metrics` | `127.0.0.1:3030` | `127.0.0.1:3031` | -**What runs what, as of 15 November 2026.** Each public `/health` names its own commit, and that is the +**What runs what, as of 16 November 2026.** Each public `/health` names its own commit, and that is the one to trust when this table has gone stale. | Service | Host | Commit | Since (UTC) | |---|---|---|---| -| Primary oracle | `archipelago-standby` | `9a55028` | 15 November, 01:34 | -| Checker | `packrai-vps` | `9a55028` | 15 November, 01:22 | +| Primary oracle | `archipelago-standby` | `2ed3453` | 16 November, 01:34 | +| Checker | `packrai-vps` | `2ed3453` | 16 November, 01:33 | | Event index | `archipelago-standby` | `b95f0aa` | 5 November, 13:31 | -| Watcher | a third host | `9a55028` | 15 November, 01:35 | -| Policy signer | `packrai-vps` | `b95f0aa` | 5 November, 13:33 | +| Watcher | a third host | `489372c` | 15 November, 23:49 | +| Policy signer | `packrai-vps` | `489372c` | 15 November, 23:49 | | ERC-8004 publisher | `archipelago-standby` | `e39ff5b` | 5 November, 11:08 | | Slash keepers | `archipelago-standby`, `packrai-vps` | `50386ef` | 10 November, 16:39 | diff --git a/oracle/README.md b/oracle/README.md index 77b1ebf..507cf27 100644 --- a/oracle/README.md +++ b/oracle/README.md @@ -955,7 +955,7 @@ and then its silence means nothing. The slash keeper (`keeper.js`) has its own s | Variable | Description | |----------|-------------| -| `RPC_URL` | Ethereum JSON-RPC endpoint for the target chain | +| `RPC_URL` | Ethereum JSON-RPC endpoint for the target chain. A request gives up after 30 s, and ethers does not retry it. Reads are retried by `readWithBackoff`, most of them four tries with waits of 0.5, 1 and 2 s. A refusal for the node's rate limit gets seven, each wait between half and all of a step that doubles to 8 s, so about 24 s at most | | `ORACLE_PRIVATE_KEY` | Private key for the oracle wallet (must have ORACLE_ROLE). Required unless `ORACLE_PRIVATE_KEY_FILE` is set | | `ORACLE_PRIVATE_KEY_FILE` | Path to a file holding that key, in place of `ORACLE_PRIVATE_KEY`; setting both stops the process. Prefer it in production: an environment variable shows in `docker inspect`, in `/proc//environ` and in every child process, while a compose secret mounts the file read-only. A malformed key is refused with a fixed message that never contains the value | | `IDENTITY_ADDRESS` | SigvaraIdentity contract address | @@ -1069,7 +1069,7 @@ A probe without the recipient filter would not do: it proves the asset and chain right and then leaves the normal case, this asset moved but not to any of our agents, indistinguishable from a broken recipient filter. -A long catch-up is what the pace is for. A cold scan from `FROM_BLOCK` on a chain this fast is hundreds of consecutive calls, and without spacing them the run is a coin flip rather than a wait: `readWithBackoff` allows four attempts, one chunk exhausting them throws out of the scan, the checkpoint is left unadvanced, and the whole range is retried next epoch. At the default pace a 255-call run takes a little over two minutes if nothing is refused, and longer in practice: a read-only probe of exactly that range on 3 Oct 2026 still had 27% of chunks refused at 500ms, because the bucket drains over a run that long even though 72 consecutive calls do not. The scan survives it where that probe would not, since every call goes through `readWithBackoff`. Pacing makes a catch-up finish; it does not make it refusal-free. +A long catch-up is what the pace is for. A cold scan from `FROM_BLOCK` on a chain this fast is hundreds of consecutive calls, and without spacing them the run is a coin flip rather than a wait: `readWithBackoff` allows four attempts, or seven for a rate limit, one chunk exhausting them throws out of the scan, the checkpoint is left unadvanced, and the whole range is retried next epoch. At the default pace a 255-call run takes a little over two minutes if nothing is refused, and longer in practice: a read-only probe of exactly that range on 3 Oct 2026 still had 27% of chunks refused at 500ms, because the bucket drains over a run that long even though 72 consecutive calls do not. The scan survives it where that probe would not, since every call goes through `readWithBackoff`. Pacing makes a catch-up finish; it does not make it refusal-free. **Enable it on every operator or on none.** Two operators scanning and one not is the same delivery asymmetry this exists to remove, pointed the other way. diff --git a/oracle/RUNBOOK-second-operator.md b/oracle/RUNBOOK-second-operator.md index 7cabee7..b7b14dc 100644 --- a/oracle/RUNBOOK-second-operator.md +++ b/oracle/RUNBOOK-second-operator.md @@ -103,7 +103,7 @@ cast call 0x3c9c12F27DDCa7048840eE3fbF0CAa1C547D8171 "bondAmount()(uint256)" --r refused 19 of 30 with `-32005`, Circle's endpoint refused 20 of 30. Individual calls are cheap, about 120ms for a 10,000-block window, so the limit is on request *rate* and not on per-call cost. The scanner used to have no pacing between chunks, which made a long - catch-up fragile: `scanRange` wraps each call in `readWithBackoff` with four attempts, so + catch-up fragile: `scanRange` wrapped each call in `readWithBackoff` with four attempts, so a few hundred consecutive chunks would probably exhaust one of them, throw, and leave the checkpoint unadvanced for a full retry next epoch.