Polymarket Market-Expansion Scan — Negative Result, Moat Identified (2026-07-09)

A 10-agent workflow scanned 8 Polymarket market families for reprice-lag/structural edges beyond the live crypto 5m lane: 0 of 25 candidates survived deep-dive. The negative result is load-bearing: the 5m/15m Chainlink-settled crypto family is the ONLY automatic, dispute-free, minutes-to-redemption settlement on the venue. Everything else resolves via the UMA optimistic oracle (~2h+ lockup), which — combined with our BUY-only / hold-to-resolution pipeline — kills capital velocity even where a genuine reprice lag exists. The lane’s real moat is not just the reprice-lag model; it is capital velocity from automatic Chainlink settlement (~2 min to redeemable).

Source of truth: repo docs/superpowers/market-expansion-scan-2026-07-09.md. Evidence rule enforced: every number is a live Gamma/CLOB/data-api pull made 2026-07-09 ~16:45–16:58 UTC (BTC $62,760 at scan), not model memory.

For Agents — binding conclusions

1. The moat = settlement velocity. Any candidate settling via UMA pays proposal (12–16 min post-close) + ~2h liveness ≈ 2h+ redemption lock per fire vs ~2 min on 5m. With no sell path that is a 45–70× capital-velocity penalty that dominates any real reprice lag. Expansion candidates died on settlement/velocity or competition — never on missing truth feeds.
2. Do not build: touch barriers (refuted by natural experiment), sports/esports live (pipeline mismatch), econ prints (capacity/cadence), tweet counters (occupied), structural arb (null result, documented below — do not re-check).
3. Watchlist (passive, zero build): HYPE 5m/15m liquidity (7th underlying, literally our infra), 4h cadence volume, hourly series if UMA settlement ever moves to the Chainlink adapter.
4. Best growth axis is depth in the current lane, not breadth on the venue.

Ranked outcome

#	Candidate	Family	Scan → dive	Verdict	One-line kill
1	Hourly Up/Down (BTC/ETH/SOL/XRP)	crypto-periodic	7.5 → 3.5	weak	Real + liquid, but UMA lockup = 45–70× capital-velocity penalty; strictly dominated by 5m
2	BTC/ETH monthly touch barriers	onchain-oracle	6 → 2	dead	Natural experiment: no post-touch lag window; entries sit above our 0.88 cap
3	Noon-ET strike ladders	crypto-structure	5	not dived	Same Binance-leads-book edge on a slower surface; ladders efficiently arbitraged
4	CPI release reprice snap	econ-data	4	not dived	Real 7–20s lag but $4.5–22k books, ~monthly cadence, pure parse race, 2h+ UMA lock
5	Weather daily buckets	weather (peer lane)	4	map-only	Delivered as a map for the weather peer (below)
—	20 others	all	≤4	—	see family notes

Deep-dive 1 — Hourly Up/Down: real, liquid, still not worth it

What’s real (verified live): btc-up-or-down-hourly (id 10114, running since 2025-05-27) does **$1.0Mvol24h\ast \ast ($27–54k per hourly window); ETH $241k,SOL$105k, XRP $101k.Settlementtext=BinanceBTC/USDT1Hcandle,Upifclose\ge open—Binanceisboththetruthfeedandthesettlementsource,\ast \ast zerooraclebasis\ast \ast (cleanerthanour5mChainlinklane).Deterministicslugs‘asset-up-or-down-month-d-yyyy-ham/pm-et‘.Depthisanon-issueatourclipsizes($50–300 vanishes into 8,000-share levels on BTC).

Why it still loses:

• UMA settlement (resolvedBy 0x65070BE9…, bond 500): outcome proposed 12–16 min after candle close (measured across 6 consecutive windows), then ~2h liveness → ~2.25h locked per fire vs ~2 min on 5m — a 45–70× capital-velocity penalty for the same mechanism.
• The edge is a strict subset of the 5m lane’s. Live probe at T-18s of a decided window: P_fair(Up)=0.0000, book already 0.998/0.01 — nothing left to capture. Only ~29% of hours are still “live” (|ln(S/O)| < 0.145% vs measured σ₁ₕ = 0.398%) at T-2min; fewer clear conviction ≥ 0.30 + net edge → ~8–14 genuine fires/day, ~$20–150/day gross before the resident-bot haircut.
• Resident fast competition confirmed: wallet 0xee55214e… appears 53× in the live window’s last 500 trades; five more wallets 17–50× each; ~16 trades/min in one window.
• The apparent “favorite underpriced at 0.905 vs 0.97” pattern is mostly time-value discount for the 2.25h lock + dispute tail — we would pay the same carry, so it is not capturable alpha.

Build gotcha if ever revisited

BTC hourly tick is 0.001 (ETH/SOL/XRP are 0.01). A 1h arm needs its own calibration re-fit AND a fresh walk-forward gate for the 1h τ band — the frozen 5m calibration.json must NOT be reused.

An observe-only probe sketch with a kill criterion (< ~8 genuine fires/day, or losing the fill race) lives in the workflow journal; the recommendation is don’t bother while the 5m/15m lanes have headroom.

Deep-dive 2 — Monthly touch barriers: refuted by natural experiment

Hypothesis was “buy YES the moment Binance 1m High touches the strike, before the book reprices.” The one BTC barrier that resolved YES this month (reach-62.5k, touched 2026-07-03 20:19:16 UTC) refutes it:

• The “grabbable” YES depth 0.90–0.99 ($19,498) filled over ~31 hours as BTC ground toward the strike — fair-value repricing with real reversal risk, not a lag window.
• At the actual touch second only **~$1,860traded,allat0.999\ast \ast (0.1cgross).Thefastestsweepsecondoffered$1,300 below 0.98 and repriced within the same second.
• Hard infra blocker: every economic entry is ≥ 0.90 — above our const-asserted EMIT_MAX_LIMIT 0.88. Un-fireable on our stack by design.
• A single algorithmic MM parks near-identical walls (~$35k@0.99+$14.5k @ 0.96–0.98) on every barrier and will out-cancel a NATS→executor path.
• dip-60k’s touch produced exactly **1 trade / $10\ast \ast .Touch-racecapacity\approx$0/day.

Structural arb — null result (do not re-check)

No static free money anywhere in crypto structure, live-verified 2026-07-09:

• All three BTC strike ladders monotonic — no P(>X) inversion.
• Up+Down complements all ≥ 1.00 (1.01 / 1.02 / 1.07) — no BUY-both underround.
• Exact-price buckets consistent with the above-ladder to <1c.
• Depth real only at BTC/ETH ATM strikes ($7–9.5k within 5c, 1c spreads); SOL/XRP ladders thin and gappy.

Family notes (all live-verified)

• Crypto periodic — the settlement split that defines the moat: timestamp-slug {asset}-updown-{5m,15m,4h} = Chainlink, automatic, no dispute (our family). Named-hour series = Binance candle + UMA. The 4h cadence is new, exists for all 7 assets, but does $0.5–2.2k/day — dead for now.
• Econ data: bimodal. FOMC deep ($2.38Mvol24h,$3.3M liq) but converges <5s under six-figure orders — saturated. CPI/PPI/core prints show a real measured lag — uncertain buckets snap 0.43–0.51 → ~0.95 in 7–20s with fills observed mid-flight at 0.70–0.95 — but books are thin ($4.5–22k), there is no model edge (the winner is a lookup), ~half of prints are pre-priced to 0.99, it needs new BLS/BEA parser infra, cadence is ~monthly, and measured release-to-redemption is 2h11m–2h27m. Rejected. (Next prints: CPI Jul 14, FOMC Jul 29.)
• Sports/esports live: the venue’s volume king (France–Morocco semi: $6.0Mmoneyline+$11.3M exact-score + $10.2Mprops;WTAlivematch$4.6M) and completely mispaired with our pipeline: pro-tight books (France 0.6025/0.605 with **$2.7Maskdepthwithin2c\ast \ast ),fast-resolvingsub-marketsthin(Set-2winner$1k within 5c; live Dota game book $2.3k), everything UMA (~2h+ to cash after match end). Zero candidates by design, not omission.
• Mentions/social: only liquid product is elon-tweets weekly (series $1.86M vol24h; three overlapping 7-day events). Resolution source is Polymarket’s own public xtracker API; resident algos already trade it every 10–90s; the 26 bucket asks sum to ~1.03 → no BUY-only set arb. All sibling counters (trump-truths, potus-tweets, powell-mentions) dead since early 2025.
• Fast news: only FOMC has liquidity; all UMA vs simultaneous public releases (no leading feed); 8–12 sub-second sprint moments per year. Pass.
• Commodities/TradFi (new since our last look): WTI monthly touch settles on Pyth ($392kvol24h);gold/silver/single-stock/10Ydailyproductsall\le$5k. Watch, don’t touch.

Watchlist

Item	What it is	Trigger to act
HYPE (Hyperliquid)	Newly listed 7th up/down underlying at 5m/15m/4h/daily, ~$0 liquidity today	It inherits our exact infra the day it gets flow — watch 5m/15m volume. Updates the “5m universe is exactly 6 assets” read (true as of 2026-07-05; HYPE listed since)
4h cadence	All 7 assets, Chainlink-settled like ours	$0.5–2.2k/day today — dead until volume appears
Hourly series	Binance-candle truth, UMA-settled	Only if settlement ever moves to the Chainlink adapter
WTI monthly touch	Pyth-settled, $392k vol24h	Watch, don’t touch

Weather map (deliverable for the weather peer)

184 open weather events, ~$3.11Mvol24h/$11.19M liq. 90% of volume = “Highest temperature in [CITY] on [DATE]” (125 daily events, 51 cities, ~11 negRisk buckets each; US cities 2°F buckets, metric cities 1°C).

• Discovery gotcha: only NYC ($158k/d)andLondon($181k/d) exist as series objects — the other ~49 cities are only discoverable via tag_slug=weather.
• No intraday-settling products exist; everything resolves next-day (“cannot resolve until the first data point for the following date has been published”) — the Gamma endDate (noon UTC) is a display artifact.
• Settlement is per-city heterogeneous: mostly Wunderground at a named airport station (NYC=KLGA, London=EGLC, Tokyo=Haneda, Chicago=O’Hare, Shanghai=Pudong), but e.g. Hong Kong uses the HK Observatory “Absolute Daily Max”.
• Books thin outside the 1–2 favorite buckets ($3–90 within 5c on tails).

Strategic conclusion

1. The 5m/15m Chainlink-settled crypto family is structurally unique on Polymarket — the only automatic, dispute-free, minutes-scale settlement on the venue. Every expansion candidate died on settlement/velocity or competition, not on missing truth feeds. Best growth axis = depth in the current lane (the report names the pre-registered 15m BTC arm as the highest-value expansion; note the 15m observe read currently gates any 15m money lane — efficiently priced, no τ pocket).
2. Watchlist (passive, zero build): HYPE 5m/15m liquidity, 4h cadence volume, hourly series if UMA settlement moves to Chainlink.
3. Do not build: touch barriers (refuted), sports/esports live (pipeline mismatch), econ prints (capacity/cadence), tweet counters (occupied), structural arb (null result, now documented).

Provenance

Workflow run wf_3a86c588-6be — 10 Opus subagents (8 family scanners → 2 deep-dives; adversarial verify never triggered, nothing reached “promising”), 116 tool calls, ~22 min, 745k tokens. Per-agent evidence + full deep-dive JSON in the session workflow journal (subagents/workflows/wf_3a86c588-6be/journal.jsonl). All prices/volumes point-in-time live pulls, 2026-07-09 ~16:45–16:58 UTC.

Related

• crypto-shortterm-15m-observe-nogo-2026-07-06 — the 15m arm read (efficiently priced, no tradeable τ pocket); frames what “depth in the current lane” can and cannot mean today.
• crypto-shortterm-competitor-wowitsamazing-refuted-2026-07-03 — competitor analysis on our own family; the fee term 0.07·x·(1−x) validated there is the same drag any expansion lane must clear.
• crypto-shortterm-bnb-expansion-eval-2026-07-05 — established “5m universe = exactly 6 assets” (2026-07-05); the HYPE listing is the first change to that fact.
• crypto-shortterm-data-sources — settlement mechanics of our lane (on-chain Chainlink via Polygon/Alchemy; UMA ≥2h lock-up noted there).
• Crypto 5m Up-Down — Oracle-Lag Strategy (Canonical Spec) — the validated 5m lane every candidate was measured against.
• crypto-shortterm — project index.