NEAR Watchtower

A network-health agent for NEAR Protocol — and an open lab around it. The agent polls live public data, scores the network from evidence, watches its own history for anomalies, and publishes a feed (data/status.json) that other agents act on (one does, below). On this page you can also re-weight the model yourself and watch the score change: the methodology is inspectable, not a black box.
Snapshot: 2026-11-09T12:55:15.501635+00:00 · snapshot #2

live-recomputed as you move the controls

Health components

Cadence

95.2

Decentralization

60.0

Staking

73.8

Throughput

100.0

Liquidity

96.9

Alerts from this snapshot

No alerts this snapshot — every component is within its normal band.

Score history

2 committed snapshots — the agent's own time series (data/history.jsonl). First: 2026-11-09T12:54Z · latest: 2026-11-09T12:55Z.

The lab — re-weight the model

Weights are policy. The defaults are the published ones (decentralization counts most, at 25%). Drag them, or try a preset, and the gauge, bars and score above recompute instantly from the same evidence. Reset restores the author's published model.

Another agent, acting on this feed

Demo treasury agent → 70% growth / 30% reserve

SIMULATION — no real funds. Decision by agent_consumer.py from this feed (label STRONG, 83.2/100).

health 83.2/100 -> STRONG, base growth share 70%
weakest component: decentralization (60.0/100)

The evidence

$3.57

NEAR price (USD)

$4,682,467,760

Market cap

-82.5%

vs all-time high $20.44

339

Validators in the live set

Nakamoto coefficient

44.31%

Of supply staked

5.7% / 36.2%

Largest pool / top-10 share of stake

5,478,061,250

Lifetime transactions

0.571s

Average block time

151,208

Blocks per day (est.)

11.63%

24h volume / market cap

223,769,779

Latest block height

How another agent uses this

GET data/status.json
{
  "health_score": 83.2, "health_label": "STRONG",
  "components": { "cadence": ..., "decentralization": ..., "staking": ..., "throughput": ..., "liquidity": ... },
  "alerts": [ { "component": "...", "kind": "z_score | low_component | label_change", "detail": "..." } ],
  "evidence": { "nakamoto_coefficient": 9, "staked_pct_of_supply": 44.31, ... },
  "generated_at": "2026-11-09T12:55:15.501635+00:00"
}

Method & sources

NearBlocks API — supply, price, market cap, volume, lifetime transactions, nodes online, block time.
NEAR mainnet RPC — chain status and the live validator set; stake shares and the Nakamoto coefficient are summed directly.
CoinGecko — market data cross-check (price, market cap, ATH).
Alerts: any component under 40 flags immediately; label changes flag; with ≥5 snapshots of history, components also flag at |z| ≥ 2 against their own history.
Full methodology, weights rationale and limitations: the repository README. Raw responses: data/snapshot_raw.json.

Re-run the whole system yourself: python3 watchtower.py && python3 agent_consumer.py && python3 build_dashboard.py — Python 3 standard library only, no API keys.
Built by team zarroww for the NEAR Hackathon (Agent Economy track). Sibling project: NEAR by the Numbers, a re-runnable investment-thesis model. Research and education only — not financial advice.