feat: add read-only liveness keeper for ADR-013 resolution monitoring

docs/prediction-market/dao-dao-verdict-runbook.md

@@ -48,6 +48,12 @@ message directly. There is no generic relay and no authority rotation.
 
 ## Pre-execution review
 
+`scripts/governance/liveness_keeper.py preflight` performs steps 1-5 below as
+read-only LCD queries and emits a `prepare_rehearsal.py`-compatible request
+JSON with the queried fields pre-filled; it cross-checks question-ID,
+deadline, and arbitrator agreement between market and oracle before emitting
+anything. It never submits, votes, or executes. Steps 6-7 remain manual.
+
 1. Query market `config` and require `verdict_authority` to equal the DAO core
    above exactly; record the immutable market and oracle addresses.
 2. Query market `question` and copy its 32-byte `question_id` exactly.

docs/prediction-market/governance-rehearsal/README.md

@@ -10,7 +10,7 @@ The market, not x/gov, is cw-reality's configured arbitrator. The proposal there
 
 ## Offline preparation
 
-Create a request matching `request.schema.json` from same-height direct queries. Do not copy stale fixture values. The request binds:
+Create a request matching `request.schema.json` from same-height direct queries. Do not copy stale fixture values. `scripts/governance/liveness_keeper.py preflight` performs these direct queries and emits a matching request JSON for one pending-arbitration market; it is read-only and still requires the independent revalidation below. The request binds:
 
 - `juno-1`, the live governance module account, and distinct exact market/oracle addresses;
 - canonical base64 encodings of an exact 32-byte question ID and answer;

docs/prediction-market/liveness-keeper.md

@@ -0,0 +1,85 @@
+# Liveness keeper (operational tooling, not a new decision)
+
+**Status:** proposed operational tooling; does not amend any accepted ADR, decision, protocol parameter, or generated schema.
+**Implements:** the alerts/keepers half of [ADR-013](adrs/ADR-013-resolution-liveness.md) and the automatable steps of [dao-dao-verdict-runbook.md](dao-dao-verdict-runbook.md) and [governance-rehearsal/README.md](governance-rehearsal/README.md).
+
+`scripts/governance/liveness_keeper.py` is a dependency-free, read-only Python
+script. It performs LCD smart-queries over plain HTTPS GET and never signs,
+broadcasts, or holds key material — the same posture as
+`scripts/oracle/verify-deployment.py` and `scripts/governance/prepare_rehearsal.py`.
+
+## `report`
+
+```sh
+python3 scripts/governance/liveness_keeper.py report \
+  --lcd https://lcd-juno.example.com \
+  --market <MARKET_ADDR> [--market <MARKET_ADDR> ...]
+```
+
+For each market, queries `Config`, `State`, `Question`, and `Challenge`, then
+(if a question is bound) the oracle's `Question`. Prints one JSON report with
+a per-market `alert` of `ok`, `attention`, or `urgent`:
+
+- **`attention`** — `awaiting_resolution` (question closed, unanswered oracle
+  answer still pending) or an oracle question past its `opening_ts` with no
+  answer yet.
+- **`urgent`** — `pending_arbitration` on either the market or the oracle,
+  with the challenge/arbitration deadline countdown included.
+
+This is intended to run on a schedule (cron, CI scheduled job, or any
+existing alerting pipeline) so operators learn about a stalled question or an
+open challenge without polling the chain by hand.
+
+## `preflight`
+
+```sh
+python3 scripts/governance/liveness_keeper.py preflight \
+  --lcd https://lcd-juno.example.com \
+  --market <MARKET_ADDR> \
+  --answer <BASE64_32_BYTES> \
+  --payee <JUNO_PAYEE> \
+  --title "..." \
+  --summary "..." \
+  --out request.json
+```
+
+For one market that is `pending_arbitration`, queries every field
+`scripts/governance/prepare_rehearsal.py` requires and cross-checks
+question-ID, deadline, and arbitrator agreement between the market and the
+oracle before emitting anything. `--governance-module` defaults to the
+ADR-017 v1 verdict authority (the Juno Agents DAO core;
+`juno18k65at7fkf8elhece0fnhsvuxggqg6cved6trp5fyk3lftfn93xsmpeaac`) and can be
+overridden for a rehearsal-only `x/gov` profile.
+
+The emitted `request.json` is consumed unmodified by the existing tool:
+
+```sh
+python3 scripts/governance/prepare_rehearsal.py build request.json packet.json
+python3 scripts/governance/prepare_rehearsal.py validate packet.json
+```
+
+`preflight` only narrows what `prepare_rehearsal.py` already validates — it
+adds pre-fetch consistency checks, it does not relax any existing check. It
+does not choose the answer or payee (those remain a human/DAO decision), and
+it does not build the proposal packet, submit a proposal, vote, or execute.
+
+## What this does not do
+
+- No transaction is ever signed or broadcast; no private key material is
+  read, generated, or required.
+- No default is silently trusted: `preflight` fails closed if the market and
+  oracle disagree on question ID, deadline, or arbitrator, or if either side
+  is not actually `pending_arbitration`.
+- No accepted ADR, protocol parameter, generated schema, or release artifact
+  is changed. This is additive operational tooling plus its own test file
+  (`tests/governance/test_liveness_keeper.py`, wired into `scripts/validate.sh`).
+
+## Tests
+
+`tests/governance/test_liveness_keeper.py` covers the pure classification and
+request-assembly functions with fixture data matching the golden wire schema
+in `contracts/binary-market/tests/wire_schema_golden.rs`, mocks LCD HTTP
+calls to verify request construction and error handling, and includes an
+end-to-end check that `build_preflight_request`'s output is accepted by the
+real `prepare_rehearsal.py build`/`validate_packet` and produces the exact
+expected `governance_verdict` execute message.