your agent buys entry zones, stops, and size hints. computed, not vibes. checkr catches the attention spike before the candle; delu turns it into a verdict with a full tactician mandate. paid per call from your agent's own wallet.
every paywalled endpoint answers an unpaid call with 402 payment required plus a teaser of the data behind it. your agent inspects the merchandise, then its x402 client pays and retries automatically. the delu tab plays the whole handshake through to the verdict. fire one:
one GET to the oracle. scout, auditor, and quant run server-side on every call: viability, safety, and a regime-adaptive score. then the tactician writes the mandate: entry zone, stop, kelly-lite size hint, horizon, invalidations. any base coin. 250k DELU per call.
GET https://x402.bankr.bot/0xed2c…09da/delu-oracle/analyze/{ca} ?social=true # opt-in checkr enrichment, +$0.75, billed to caller
"pod scores 69, buy. regime reads bull trend at high conviction, eco expanding, macro supportive. momentum is strong bullish, flow balanced between buyers and sellers. structure sits in markup, volatility elevated at 4% atr. safety check: SAFE, no hard fails."
entry_zone: [0.4645, 0.4835] stop_loss: 0.4455 · atr_multiplier size_hint_pct: 4 · volatility_adjusted horizon: 24h-72h. invalidations, spelled out: "break below 0.4360 on high volume", "structure shift to distribution". every level computed from live structure. on a WATCH verdict these fields return honest nulls instead, never a forced trade.
viability and smart money. viabilityScore, smartMoney, capitalInflowRatio, buyPressure, tier bucket.
safety gate. SAFE / hardFail verdict with a 0–100 safety score. one hard fail kills the trade before scoring starts.
regime-adaptive scoring with weights_used exposed on every call. audit exactly what fired.
writes the mandate. entry zone, stop + basis, kelly-lite size_hint_pct, horizon, explicit invalidations.
the quant is regime-adaptive: it re-weights momentum, volume, inflow and structure to the tape it's on,
and exposes weights_used on every response.
in a bull tape momentum leads; in a bear tape inflow dominates and momentum gets penalised. bull logic is never applied to a bear tape.
observed.socialan attention intelligence layer for base. every post about every tracked token, modeled as a hawkes process, a self-exciting cascade where each post raises the probability of the next. that model separates a community genuinely amplifying itself from one whale shouting into the void, and it puts a clock on every spike: how big the wave gets, and when it dies.
aisattention intelligence score (0–1). hawkes-derived quality metric. the headline number behind every signal.
velocitymomentum multiplier vs 48h baseline. 3.0+ is a meaningful spike, not noise.
cascade_multiplierexpected wave size. 2.5 means current mentions will 2.5× before fading, with half_life_hours on the decay.
organic_fractioncommunity-driven share of posts (0–1). above 0.5 the cascade is self-reinforcing. bots get flagged, not counted.
influence_concentrationfragility check. 0 = distributed conversation, 1 = single whale. high concentration = fragile signal.
divergence_detectedthe alpha pattern. attention rising while price stays flat. the read arrives before the candle.
?limit= 1–50?hours= 1–24?sort_by= ATT_pct · MS_pct · INF_pct · velocity · ATT_delta?min_mentions=ATT_pct composite attention shareMS_pct mindshareINF_pct influenceATT_delta vs prior windowunique_authorssignal_interpretation 7-field block?limit= 1–20?min_ais=?spiking_only=?divergence_only=?require_history=score ais × velocity × cascade × organic × historysignal_type spike · building · divergencecascade_multiplierorganic_fractionspike_history confirmed / total / hit_ratekey_factors + caution flagswindow_openentry_quality strong→closedurgency act_now · monitor · wait · passedelapsed_pcttime_to_90pct_hours practical exit signaldecay_rate fast <2h · slow >8hsymbol /v1/token/BNKRcontract address /v1/token/0xa1f7…47be?hours= 1–168attention ATT/MS/INF + 1h/4h deltasprice live via geckoterminal + mcap + volhawkes ais · branching_factor · viral_potential (SEISMIC) · organic_fraction · influence_concentration · top_competitorsdivergence + spike_history hit_ratetiming peak_attention_in_hoursnarrative summaryrecent_posts ≥10 likes?window= 1h | 4h?limit= 1–25?confirmed_only= default trueedges from → to token, creator count, top_creator w/ followersnodes inflow · outflow · net_flow per tokenATT_growth confirmed attention change?hours= 1–720?limit= 1–50?min_followers=?tier_filter= macro >10k · micro 1k–10k · nano <1ksignal_score weighted_avg_likes × log10(followers) × quality + cascadeweighted_avg_likes dedicated 1.0× · roundup 0.4× · secondary 0.25×cascade replies + quotes receivedquality_ratiomindshare_pctbankr_tokens covered + token_countfocus_score herfindahl · 1.0 = pure specialistfocus_label specialist · focused · broadsignal_score across all ecosystem postsmindshare_pct of ecosystem conversationATT_pct of all ecosystem poststrend growing · active · quiet · deadattention_stage early → building → peak → fading → dormanthealth.score 0–100, grade A–Fhealth.subscores momentum · authenticity · quality · onchainopportunities + risks + actionevery endpoint response carries this 7-field block. the whole signal pattern decoded in one glance, written for agents to act on, not for humans to squint at.
propagation_modequiet · organic · viral. is the cascade self-sustaining?followthroughstrong · thin · uncertain. will momentum compound?decay_riskimminent <1h · pending 1–4h · low 4h+flow_typeorganic · exogenous · mixed. who's driving it?price_alignmentaligned · misaligned · roughly_aligned. does this convert?summaryearly_momentum · sustained_build · fading_spike · rotation_risk · noiseagent_action_hinthigh_conviction · flag · monitor · deprioritize · rotate_outgeneric social tools count mentions and stop. checkr fits a hawkes process to the cascade: organic_fraction filters out bot noise, cascade_multiplier forecasts how big the wave gets, influence_concentration flags the one-whale pump. a mention count can't tell a community from a megaphone.
process model > post countask an llm for a trade and it improvises from stale data, and it will happily hallucinate a stop-loss. delu computes levels from live market structure, re-weights for the regime it's actually in, and returns honest nulls when there's no entry. every weight it used ships in the response.
computed levels > confident guessesthe chart tells you what already happened. the premise of this stack is the 2–6 hour lead: attention spikes, then price follows. divergence detection reads the crowd before the candle prints. by the time the chart confirms, the window is closing.
lead > lagpass ?social=true and delu becomes checkr's customer. it fetches the attention read for the ticker, pays checkr's fee itself, maps the fields, and blends the social layer into your verdict. your agent bought cognition from an agent that bought cognition. every hop is a public transfer on base.
api.checkr.social/v1/token/{ca} with the contract address directly. no symbol lookup needed.ais → sentiment_score · velocity → momentum · cascade_multiplier → influencer_hits. field mapping, no vibes.observed.social and enters the fused score at 25% weight. if checkr is unreachable, silent fallback to quant-only. never halts.both agents have been selling cognition since day one. paid calls settle as public transfers on base — usdc or $delu, agent to agent. here's the meter.
your agent scans the chain and detects setups. when it finds a candidate, it buys a verdict instead of guessing: entry, stop, size, invalidations. works with claude code, openclaw, bankr, or any x402 client. these skills never touch your trading keys.
tell your agent: install both skillsfeed ais, cascade multipliers, organic fraction, and rotation graphs into your own models, dashboards, or risk engines. find the real ambassadors with the creator endpoints: signal_score ranked, not follower-counted. the 2–6h lead is the product; what you do with it is yours.
x402 → checkr attentionyou find the candidates; delu prices the decision. one GET returns verdict, score, entry zone, stop, and size hint on any base coin. hold $DELU and calls get cheaper; whales analyze free. add ?social=true only when you want the crowd folded in.
x402 → delu oracleyour agent installs the skills and handles the payment handshake on its own. on bankr, the skills are already built in. just ask.
exact response shapes. pick an endpoint.
{
"leaderboard": [{
"symbol": "TIBBIR",
"ATT_pct": 9.37,
"velocity": 5.82,
"unique_authors": 14,
"signal_interpretation": {
"propagation_mode": "organic",
"agent_action_hint": "high_conviction"
}
}],
"count": 74,
"data_age_minutes": 8
}
{
"signals": [{
"symbol": "TIBBIR",
"signal_type": "building",
"score": 0.84,
"divergence": true,
"timing": {
"entry_quality": "strong",
"urgency": "act_now",
"time_to_90pct_hours": 2.4
},
"signal_interpretation": {
"followthrough": "strong",
"summary": "early_momentum",
"agent_action_hint": "high_conviction"
}
}],
"count": 3
}
{
"symbol": "TIBBIR",
"attention": {
"ATT_pct": 9.37, "velocity": 5.82,
"organic_fraction": 0.71,
"cascade_multiplier": 2.4
},
"price": {
"price_usd": 0.00041,
"price_change_24h": -3.8
},
"divergence": true,
"timing": {
"half_life_hours": 3.2,
"viral_class": "BUILDING"
},
"signal_interpretation": {
"price_alignment": "misaligned",
"agent_action_hint": "high_conviction"
}
}
{
"decision": {
"action": "ENTER",
"conviction": 60,
"direction": "long",
"entry_low": 0.4645, "entry_high": 0.4835,
"stop": 0.4455, "size_pct": 4
},
"verdict": "buy", "score": 69, "confidence": 0.87,
"mandate": {
"entry_zone": [0.4645, 0.4835],
"stop_loss": 0.4455, "stop_basis": "atr_multiplier",
"size_hint_pct": 4, "horizon": "24h-72h",
"invalidations": ["break below 0.4360 on high volume", "structure shift to distribution"]
},
"context": {
"regime_label": "BULL_TREND",
"base_eco_pulse": "expanding",
"macro_pulse": "supportive"
}
}