Zer0-Knowledge-Hack · moises-cisneros · Sep 9, 2026 · Sep 9, 2026
diff --git a/evidence/README.md b/evidence/README.md
@@ -34,9 +34,9 @@ from our marketplace themselves during judging.
 
 | Task | Category | Agent leg | DIY leg | State |
 |------|----------|-----------|---------|-------|
-| Task 1 — health-factor monitoring | health_factor | pending live hfwatch endpoint (A) | dry-run recorded | protocol + DIY recorded 2026-09-06 |
-| Task 2 — _planned_ | _open_ | — | — | outline only |
-| Task 3 — must be trading/equities/security (hard TermiX requirement) | _open_ | — | — | outline only |
+| Task 1 — health-factor monitoring | health_factor | pending marketplace hire against deployed agent (A+C) | measured 2026-09-09 against the live funded position | protocol frozen; DIY leg measured (HF 1.838); live signal serves |
+| Task 2 — yield-APR comparison sweep | yield | planned (`yield-router`) | — | outline only |
+| Task 3 — grid-trading execution record | grid_trading (hard TermiX requirement) | planned (`grid-runner`, already deployed/registered #22) | — | outline only |
 
 Regulation source: BNB Chain Smart Money Era hackathon page, TermiX partner-track section, and
 TermiX's own campaign page (`agent.family/campaigns/bnb-build-the-era`).
diff --git a/evidence/task-01-health-factor/metrics.md b/evidence/task-01-health-factor/metrics.md
@@ -1,9 +1,22 @@
 # Task 1 metrics — Health-factor monitoring
 
-Status as of 2026-09-06: **dry-run pair incomplete** — DIY leg recorded (rail `dry-run`),
-agent leg pending a live hfwatch endpoint (workstream A) and, for the preferred rail, testnet
-hire (workstream H). Nothing below counts as an agent-advantage datapoint until both legs run on
-the funded team position (see `evidence/README.md` honesty rule).
+Status as of 2026-09-09: the **watch position is live** (Venus BSC testnet: 0.035 tBNB supplied,
+8 $U borrowed, tx hashes in `openspec/changes/use-venus-testnet-position/proposal.md`), and the
+live health factor is served by `GET /agents/hf-watch/signal`. The DIY leg has been re-run against
+this real position; the **agent leg** still needs the marketplace hire against a deployed agent =
+workstream A's endpoint (stream H's alternative rails documented below).
+
+## Measured DIY leg (2026-09-09, real position, rail `no-agent competitor script`)
+
+| Section | Value |
+|---------|-------|
+| Health factor | **1.838032547701963** (read via the same comptroller accounting the deployer reader uses) |
+| Collateral value | 0.035 tBNB supplied into vBNB (≈ $15.3 risk-adjusted at CF 0.70, price $600) |
+| Borrow value | 8 $U borrowed (price $1.00) |
+| Approx. liquidation price | **$326.44** |
+| Machine time | ~1.5 s (batched JSON-RPC) |
+| Manual equivalent | 20–35 min first run, minutes per repeat (see dry-run table below) |
+| Verdict policy | `safe`HF≥3 / `sharpen caps` 3>HF≥1.5 / `act now`HF<1.5 → **`act now`** |
 
 ## Comparison target (frozen by protocol.md)
 
@@ -13,13 +26,21 @@ the funded team position (see `evidence/README.md` honesty rule).
 | Cost | hire price in $U recorded from the real job receipt | 0 tokens; human time |
 | Time | wall-clock from "hire clicked" to validated output | wall-clock of the manual procedure |
 
-## Current standings (dry-run only, do not cite as final)
+## Current standings (DIY done, agent pending)
 
 | Dimension | Agent | DIY | Notes |
 |-----------|-------|-----|-------|
-| Time | _pending_ | machine: 1.3 s scripted; human: est. 20–35 min first run | DIY scripted DIY is the *strong* lower bound the agent must beat — a human by-hand is far slower |
+| Time | _pending live agent_ | machine: ~1.5 s scripted; human: est. 20–35 min first run | DIY script = the strong baseline; the agent must beat dollars-and-seconds, not humans |
 | Cost | _pending_ (target ≤ 0.01 $U hire) | 0 | |
-| Output quality | _pending_ | verdict `n/a` on this input; no breakdown; no replayability | the DIY output is honest but context-free |
+| Output quality | _pending_ | four sections incl. per-market breakdown + rule-based verdict + provenance | the DIY output is honest but context-free |
+
+## Historical dry-run (2026-09-06, pre-funding — kept for the record)
+
+| Dimension | Agent | DIY | Notes |
+|-----------|-------|-----|-------|
+| Time | _pending_ | machine: 1.3 s scripted; human: est. 20–35 min first run | dry-run against an unfunded address |
+| Cost | _pending_ | 0 | |
+| Output quality | _pending_ | verdict `n/a` on this input; no breakdown; no replayability | kept unedited below in `diy/output.md` |
 
 ## Quality rubric used for the verdict section
 
@@ -29,9 +50,10 @@ the funded team position (see `evidence/README.md` honesty rule).
 
 ## To do to convert this into a TermiX-countable row
 
-1. Fund the watch wallet + create the Venus testnet position (blocker recorded in
-   `openspec/changes/use-venus-testnet-position/proposal.md`)
-2. Workstream A: deploy hfwatch, hand over endpoint + ERC-8004 registration
-3. Workstream H (owner unclaimed): testnet hire for the agent leg receipt (mock rail acceptable,
-   labeled)
-4. Re-run both legs same-day against the funded position; fill both columns
+1. ~~Fund the watch wallet + create the Venus testnet position~~ ✅ done 2026-09-09
+2. Workstream A: deploy hfwatch → hand over endpoint + ERC-8004 registration (A already deployed
+   all four agents per #20/#22 — the remaining gap is the marketplace-hire leg through the
+   catalog)
+3. ~~Workstream H: testnet hire for the agent leg receipt~~ rail available (#12 real ERC-8183
+   adapter); the receipt will exist once the hire runs against a live agent
+4. Run the agent leg same-day, fill both columns, move the row into the report
diff --git a/openspec/changes/use-venus-testnet-position/proposal.md b/openspec/changes/use-venus-testnet-position/proposal.md
@@ -7,22 +7,33 @@ default (`VENUS_NETWORK=testnet`), and the demo watch position is a single-marke
 (supply + borrow BNB) created by the team with faucet funds. Mainnet constants stay available via
 `VENUS_NETWORK=mainnet` for a later upgrade.
 
-**Pending work blocker (recorded 2026-09-06):** the reader, cache, fallback, and decision records
-are shipped on `feat/venus-hf-reader`, but the live number is waiting on testnet funds:
-
-- Watch wallet generated and waiting for funds:
-  `0x22f3e24233B9BDcC65fa855495D99Fe7d2458510`. The private key lives only in the operator's local
-  scratch storage (`/tmp/opencode/position/throwaway.key`); it is never committed. Discard the key
-  after Wed 9 Sep.
-- Funding is blocked by faucet gates: the official faucet, QuickNode, and Chainstack all require a
-  mainnet balance (~0.002 BNB), a login, or an API key, so a fresh empty wallet is silently
-  rejected. Untried self-serve route: Telegram support bot `@bnbchain_official_bot` (message the
-  wallet address, 0.3 tBNB/day). Fastest option is likely a teammate (workstream A needs tBNB for
-  agent registration on chain 97) sending 0.05–0.45 tBNB directly.
-- Once funded, remaining steps for Signal: run the three-call recipe below against the funded
-  wallet (mint → enterMarkets → borrow, ~few minutes), bake the address as the default
-  `VENUS_WATCH_ADDRESS`, confirm `GET /agents/hf-watch/signal` serves the live health factor,
-  commit to `feat/venus-hf-reader`, then run the full web demo check to close Sat 5.
+**Resolved 2026-09-09:** the team funded the watch wallet (0.0471 tBNB), the demo position was
+created on Venus BSC testnet and the live signal is served by default:
+
+- Position: 0.035 tBNB minted into vBNB, entered vBNB/vWBNB/vU markets, borrowed **8 $U** from the
+  `vU` market (the hackathon's own payment token)
+- Result: health factor **≈ 1.84**, approximate liquidation price ≈ $326, read live by the
+  deployed reader
+- Transactions: mint `0x3231792f2e5ca8ccebb358352f970d39f4da4b92a9c5e4bbeffb6b6683171651`,
+  enterMarkets(vBNB) `0x2408e1157add407dd9e2125171680f4e79cd1b27a8749a84ced1e47a4f53a433`,
+  enterMarkets(vU) `0xf5ff2acda622d651a79684157622d3a3dae505025c17bcf38095fee5252b9e7b`,
+  borrow `0x1b20e903bf74d3472256c2fcae4b1b70ae53af1dd96b4a9baebbeb5e0769acc6`
+- Constraint noted: the testnet vBNB market borrow cap is 0 and vWBNB borrow action is paused
+  (protocol governance), which is why the borrow side uses `$U`
+- The original vBNB-only position recipe below stays as documentation of what was attempted
+  first. The private key remains in the operator's local scratch storage
+  (`/tmp/opencode/position/throwaway.key`), never committed; discard after Wed 9 Sep.
+
+**Pending work blocker** (from 2026-09-06, superseded for the wallet-funding part; agent leg of
+the TermiX evidence still needs workstream A's deployed endpoint):
+
+- Watch wallet: `0x22f3e24233B9BDcC65fa855495D99Fe7d2458510` — funded 2026-09-09, position live
+  (see above). The private key stays in operator-local scratch storage and is never committed.
+- Historical funding note (resolved): the official faucet, QuickNode, and Chainstack all require
+  a mainnet balance (~0.002 BNB), a login, or an API key, so a fresh empty wallet is silently
+  rejected. (Ultimately resolved by a teammate handoff with 0.0471 tBNB.)
+- All "once funded" steps from 2026-09-06 were executed on 2026-09-09 (position created, default
+  watch address baked, live signal verified on `GET /agents/hf-watch/signal`).
 
 
 ## Why

diff --git a/packages/signals/README.md b/packages/signals/README.md
@@ -15,7 +15,7 @@ because an RPC is down.
 
 | Agent | Category | Metrics | Source |
 |-------|----------|---------|--------|
-| `hf-watch` | health_factor | `healthFactor`, `liquidationPrice` | **Live** — Venus Core Pool position reader (BSC **testnet** by default, `VENUS_NETWORK` selects), active when `VENUS_WATCH_ADDRESS` is set; fixture otherwise |
+| `hf-watch` | health_factor | `healthFactor`, `liquidationPrice` | **Live (ACTIVE)** — Venus Core Pool BSC testnet position read at request time, default watch wallet `0x22f3e24233B9BDcC65fa855495D99Fe7d2458510` (funded demo position: 0.035 tBNB supplied, 8 $U borrowed); `VENUS_WATCH_ADDRESS` overrides, fixture on any reader failure |
 | `hf-watch` | health_factor | `lastActionAt` | Fixture (agent's last recorded action) |
 | `hf-watch` | health_factor | `protocol` | Constant `"Venus"` |
 | `liq-guard` | health_factor | all | Fixture |
@@ -48,7 +48,7 @@ network (`VENUS_NETWORK`; unset/empty → `testnet`, `mainnet` for a later upgra
   (otherwise `0`).
 - Health factor is capped at 999 (supply-only positions).
 
-Enable it locally:
+Override the watched wallet locally (optional — the demo position is the default):
 
 ```bash
 VENUS_WATCH_ADDRESS=0xyourPositionAddress pnpm --filter @era/api dev

diff --git a/packages/signals/src/index.ts b/packages/signals/src/index.ts
@@ -4,6 +4,7 @@ import { readVenusPositionHealth } from "./venus";
 
 const CACHE_TTL_MS = 60_000;
 const WATCH_ADDRESS_PATTERN = /^0x[0-9a-fA-F]{40}$/;
+const DEFAULT_WATCH_ADDRESS = "0x22f3e24233B9BDcC65fa855495D99Fe7d2458510";
 
 type CacheEntry = { at: number; signal: AgentSignal };
 
@@ -15,7 +16,7 @@ function envValue(name: string): string | undefined {
   return env?.[name];
 }
 
-const VENUS_WATCH_ADDRESS = (envValue("VENUS_WATCH_ADDRESS") ?? "").trim();
+const VENUS_WATCH_ADDRESS = (envValue("VENUS_WATCH_ADDRESS") ?? DEFAULT_WATCH_ADDRESS).trim();
 
 async function refresh(agentId: string): Promise<void> {
   if (agentId !== "hf-watch" || !WATCH_ADDRESS_PATTERN.test(VENUS_WATCH_ADDRESS)) {