Robinhood Chain · prospective research
Does Cielo add useful evidence?
Status: collecting. Live trading policy is unchanged. The pilot compares fixed research overlays using evidence available before each decision.
Started 2026-09-22 05:17 UTC · collection ends 2026-10-05 10:53 UTC
Outcome follow-up ends 2026-10-06 10:53 UTC · heartbeat 2026-09-22 17:30 UTC
68First token / arm observations
28449Unique decision events
14294Valid native input snapshots
0Decisions with eligible Cielo style
14Matched closed book outcomes
234 / 4,200Credits reserved / lifetime cap
Version 2 — collection repaired September 22. Version 1 stopped at its storage cap and had no eligible Cielo style decisions. It recorded 52 first token/arm observations and 12 closed matched trades, with identical policy selections and −0.2702 ETH recorded net. This partial result does not establish an edge. V2 starts a fresh prospective cohort; the old gap is not backfilled.
Original protocol.
Open the token research queue · Contract data
Four areas under observation
Buyer support
Measures eligible buyers’ remaining curve inventory proxy from inputs captured before each decision. Transfers and routed attribution remain unverified.
Holding style
Observed buy-to-first-sell intervals, a seven-day window, and explicit censoring. Five completed episodes are required; unresolved episodes prevent a style veto.
42 wallets queried · 1484 provider event legs
Broader discovery
Provider buy legs are checked against native launches and graduations. This covers selected wallets, not the whole chain.
16 outside native launches · 48 graduated tokens
Post-entry support
Tracks each new funded attempt’s original qualifying buyers, with curve support proxies and sampled Cielo balances. Records observations without issuing exit orders.
20 attempts monitored · 6 balance snapshots · 1 latest support-loss flags
Recorded policy comparisons
Every captured pre-veto event is retained, including rejected candidates. Earlier scanner rejections are outside this journal. Unknown inputs retain baseline behaviour and are counted separately.
Conditional funded-book diagnostic
Collecting evidence. These figures reuse exactly matched funded attempts, original stakes and recorded exits. They do not estimate returns for rejected or unfunded tokens, and do not prove an executable strategy improvement.
18 exact event-ID matches · 6 unmatched attempts · 4 nonclosed matches · 0 closed matches without eligible native-quote accounting. No open-position mark-to-market; no cross-currency aggregation.
Collection freshness
Last captured decision: 2026-09-22 17:28 UTC. Last source decision: 2026-09-22 17:28 UTC. Unread journal bytes: 0. The process heartbeat alone does not establish current decision coverage.
Operating limits and evidence quality
- One API request per 15 minutes; 144 / 300 listed credits reserved today and 234 / 4,200 overall. Pending and failed requests count against the cap.
- Minute collector, half-hour publication. A maximum of 24 wallets is polled from a recorded universe of 280. The budget cannot provide comprehensive real-time coverage.
- 0 journal coverage breaks; 0 ticks with source errors. Latest errors: none.
- Provider feed coverage, event directions and state finality are not certified. Empty responses do not prove absence. RH balance access was empirically observed but is absent from the published endpoint enum.
- Current native head parity is unproven. The native own-list is based on graduation results, not wallet profitability. Unresolved or late evidence never becomes a historical policy input.
- Review eligibility requires 100 first token/arm observations, 30 eligible closed matches, 95% valid inputs and 80% Cielo style coverage, with no journal breaks. These thresholds are not statistical significance; insufficient data ends as inconclusive.
Read the frozen protocol · Aggregate status JSON · Original coverage audit