OJK OFFICIAL monthly txn value (Rp T · regulator aggregates ALL licensed exchanges = the truth line):

our live aggregate vs OJK official (2026-04 = Rp23 T/mo): extrapolating 6 complete days ×30 BRACKETS it, and does not reproduce it: credible-core Rp13 T = 0.57× OJK (43% under) → total Rp53.9 T = 2.35× OJK (135% over). More days will not close this. The gap is definitional, not sampling: self-report inflates the total, and the biggest retail players (Pintu/Ajaib/Pluang) are invisible to us so the core under-counts. Also not like-for-like on time: OJK's latest is 2026-04 and this window is live, and the official series fell from Rp37.2T (2025-11) to Rp22.982T over that gap. OJK monthly = the anchor.

OJK + DJP official figures (regulator and tax authority, monthly cadence):

Indo vs GLOBAL volume trajectory (both indexed to 100 at 2025-07 · tracking or decoupling?):

method: Direct keyless exchange API where verified (Indodax, Luno-ID, Bittime). CoinGecko raw exchange-reported 24h volume (BTC->IDR) for Indo-domiciled exchanges whose direct endpoint is key-gated or geo-blocked. App-only + no-volume-API players excluded.

caveat: Indodax self-reports Rp262B vs CoinGecko Rp224B (self-report vs aggregator gap). Direct-endpoint audit 2026-07-12: Bittime wired DIRECT (openapi.bittime.com, keyless). Tokocrypto market-data is key-gated (code:3701), Bitwyre /public API is dead (404/web-app), Pintu keyless API has price but no volume field -> CoinGecko/excluded. Upbit-ID is Komdigi-blocked at the ISP (Biznet RPZ DNS-sinkhole of id-api.upbit.com -> HTTP 000; a Jakarta/Biznet IP TRIGGERS the block, does not fix it); a DoH bypass returns live data (verified 2026-07-13) but Upbit is kept on CoinGecko per operator call. Ajaib has its own anti-scrape WAF (HTTP 403 even with full browser headers). 2026-07-13: BTSE Indonesia wired DIRECT via api.btse.id (local OJK entity, keyless market_summary IDR-quote vol; global api.btse.com is Komdigi/Biznet DNS-blocked). Reku wired DIRECT via api.reku.id /v2/ticker (its ticker carries IDR-quote vol; was wrongly excluded as market-data-only). 2026-07-17: Mobee wired DIRECT+AUTH via open-api.mobee.io /v1/markets/summary (authenticated, read-only); prior 'app-only, no public API' belief was stale (Mobee shipped an Open API). base_volume on IDR_ pairs = IDR turnover (verified quote_volume*last==base_volume). IDR-QUOTE PAIRS ONLY: Mobee's USDT book is wash-dominated (USDT_AMZNX alone self-reports ~$1.9B/24h = 97pct of its USDT volume, vs ~$160k on every other USDT pair) so USDT/USDC pairs are excluded; IDR-quote sum stays >Indodax so the wash-guard still keeps Mobee out of credible_core. BTSE ID, Bittime, Mobee report volume exceeding Indodax (the market leader), likely wash/incentivized-inflated. total is an UPPER BOUND. credible core (ex-BTSE ID/Bittime/Mobee) = Rp433B.

excluded: Pintu Pro (keyless api.pintu.co.id has price, no 24h-volume field); Bitwyre (documented /public API dead, host serves a web app); Ajaib Kripto (Cloudflare WAF 403 even w/ browser headers on BOTH the web app AND the documented API host api.ajaib.co.id; the /coin/internal/v1/price endpoint also needs ECDSA-signed auth + Ajaib exposes NO public 24h-volume endpoint. Adding = CF-bypass + API keys + request-signing for price-ONLY/no-volume. Checked 2026-07-13, not worth it); Upbit ID (Komdigi/Biznet DNS-blocked, still nebeng global/KR infra; shown via CG)

Two different quantities, and they are not comparable. The GAP is how far apart venues price the same asset. The SPREAD is the bid-ask width inside one venue. A market can have a wide gap and tight spreads, or the reverse. They were two panels until 2026-07-28, both headed "spread", which is how a reader ended up with two numbers and no way to reconcile them.

green=cheapest, gold=priciest venue per pair; AVG+GAP across live liquid quotes; grey*=stale/thin book, excluded. USDT/IDR ~Rp18,021 across all venues is ~AT PAR with interbank USD/IDR (Rp18,045, -0.13%) = no meaningful stablecoin premium right now (IDR itself is just weak, ~Rp18K/USD).

Top-of-book bid/ask width per venue = what a taker pays to cross the spread right now. This is execution cost, NOT the cross-venue gap in section 1 above. green=tightest, gold=widest live book per pair; grey*=dislocated (quote is real but the ladder is effectively empty, so it is excluded from best/worst ranking). Quotes that are crossed or wider than 25% are dropped as no-book rather than reported. A tight top-of-book here does not prove size can be filled: see the depth panel below, where the tightest quoted venue is not the cheapest to actually trade.

What a market buy actually pays, walked through the live orderbook, versus mid. Sizes are Rp10M / Rp100M / Rp1B because a depth number without the size it was measured at means nothing. green=cheapest at that size, gold>0.25%, red>1% or unfillable. "dust" marks a venue whose quoted spread sits on under Rp1M of size, i.e. a tight-looking price you cannot trade. Mobee publishes no orderbook, so its depth is UNKNOWN and is deliberately left blank rather than estimated from its spread. This is what the VISIBLE LADDER implies, not a guaranteed fill. Venues commonly apply market-order protection or price bands and would reject a large order rather than fill it at the far end of a thin book, and the ladder itself moves between snapshots. Read a large-band figure as a measure of book thinness, not as a price anyone actually paid.

No composite score, on purpose. Averaging measured and unmeasured dimensions would produce one confident number whose precision is invented, and would rank venues by which dimensions happen to be observable rather than by which venue is better. Compare WITHIN a column only. SPREAD is per-venue across its own pair basket and is therefore not like-for-like between venues, and not the same measurement as DEPTH (which is BTC/IDR at Rp100M). *=dust top-of-book, ‡=grouped API assessment covering several venues. Still UNMEASURED for every venue: fiat rails, deposit/withdrawal reliability, because no per-venue source exists yet; those are blank rather than assumed fine. UPTIME is derived from 213 archived cycles and only accrues forward, so treat it as "has not missed recently" rather than a reliability record.

This is a FLOOR, not a total. Platform fee is the venue's own pricing. Round trip adds the exchange levy and the tax, and the tax is sell-side only, so a round trip is not twice a one-way fee. Ranked only against like: Rankings compare only venues measured on the same components. Venues with missing components are NOT ranked, because a cost built from fewer parts is mechanically smaller and would win on ignorance. Fully covered and ranked: Reku, Mobee, BTSE-ID, Luno, Bittime, Indodax. Partial coverage, not ranked: none. Fee page unreadable: none. Still excluded for every venue: min_trade, network_fee, time_to_cash. No Indonesian venue publishes a fee endpoint, so every number here was read off a page by hand and carries its own source and date (checked 2026-07-29). The real cost is higher than every figure here, never lower.

Venue-sourced risk notices from Indonesian exchange Telegram groups, where impersonation warnings and rail changes surface first. These are detections, not verdicts. A row means a venue said something about risk, never that we confirmed a risk is real, so every row carries its source link and whether the venue or a member said it. Filtered 57 raw detections down to 8 by requiring an actual assertion of a threat or a state change: support replies explaining what delisting means, member questions and group rules are excluded, because keyword match finds the TOPIC and is silent on whether anything happened. "standing xN" means one notice posted N times, which is diligence by the venue, NOT N separate incidents. 🔒 Public group messages only, no member profiling, no individuals named.

Every foreign exchange in this data entered Indonesia by ACQUIRING a licensed local entity rather than by applying for a licence. The licence is the asset being traded. Stated as an observation about these rows, not as a claim that no other entry route exists.
5 source-backed edges across 13 entities, 3 of them a foreign venue entering via a licensed local shell. No ownership percentages, no inferred relationships, no scraped registry: each edge keeps the status word its source gave, so awaiting OJK clearance is never rendered as done and an investment is never flattened into an acquisition. Edge types we cannot source yet, listed rather than left absent: bank / fiat rail, custody, VC funding, listing agreement, university / talent. A sparse graph shown without its gaps reads as a complete map of a small ecosystem instead of a partial map of a larger one.

RESEARCH CHECKLIST, NOT LEGAL ADVICE. Utopia sells market-entry advisory, not compliance or legal work. Every row is a QUESTION plus where its primary source lives. Nothing here says which licence binds you or whether you are exempt, because that is legal advice, we are not qualified to give it, and a confident wrong answer costs someone their entry.
10 questions, of which 2 are answerable from live data on this page and the rest point at a primary source or a professional. Each row states what we CAN see versus what someone qualified has to answer, so the boundary is visible per row instead of buried in a footer disclaimer.

One story in two steps, and previously two panels showing the same number twice. Step 1 is whether the stablecoin rate sits off the official rupiah rate, which is the cleanest read available on capital wanting out. Step 2 compounds that with execution cost to give what the Indonesian leg of the corridor actually costs. The full journey always costs more, never less.

USDT/IDR + USDC/IDR spot = avg across the direct CEX; vs official USD/IDR (Yahoo interbank). Positive = stablecoin premium (capital-flight/demand signal), negative = discount. USDC ~Rp18,020 (-0.14% vs official) tracks USDT, same at-par read. Currently ~at par: IDR is weak (~Rp18K/USD) but no stablecoin-premium stress. 30d lines: USDT + USDC (CoinGecko) vs official (Yahoo).

INDONESIAN LEG ONLY. This is IDR to USDT on an Indonesian venue. The destination leg (USDT to SGD/MYR/THB) is NOT measured, so the full journey costs MORE than every figure here, never less. Crypto as a payment instrument is banned in Indonesia (PBI 18/40/2016); this describes asset conversion, not a payments product.
Leg 1 compounds two costs that are normally quoted separately: walking the real orderbook for a given size, and the stablecoin's own deviation from the official USD/IDR rate. They COMPOUND rather than add, because the deviation applies to the price you already paid slippage on. A venue whose book cannot fill the band is reported as unpriced rather than extrapolated. The five unmeasurable legs are listed by name so a one-leg number is never mistaken for a corridor total.
Reference: The comparison is to the INTERBANK mid rate, which is the best price in the market and one no retail buyer is ever quoted. A retail bank or money changer charges a spread on top, so the gap versus what a person actually pays at a counter is WIDER than the number shown, in the stablecoin rail's favour. We do not measure retail bank rates, so that wider gap is asserted in direction only, never in size.
Not measured (1 of 2 legs): destination conversion, bank wire comparison, remittance operators, payment pilots + ASEAN interop, offshore access / blocking

Counts and aggregate value only. No individual address is recorded, ranked, labelled or returned. The participant figure is a set CARDINALITY and identifies nobody. No nationality is inferred from any wallet.
Each deployment is queried from TWO different providers over the SAME block range and only ships when they agree within 2%. This is not caution for its own sake: one provider 403s a query another answers, and another silently caps eth_getLogs at 50 blocks, which would return a tenth of the true count from an identical request. Disagreement is PUBLISHED as disagreement. A zero is a real observation, not proof a token is unused: activity may live on another chain, in a bridge, or inside an exchange ledger where no on-chain transfer occurs. Deployments outside the measured chains are listed by name, because a silently partial sweep reads as a complete one.
Window: 100 blocks per chain. Not swept: IDRX/Gnosis, IDRX/Lisk, IDRX/WorldChain, IDRX/Etherlink, IDRX/Kaia, IDRX/Monad, IDRT/Harmony, IDRP/Kaia.
SHORT WINDOWS. Provider range limits force a 100-block window, which is only seconds to a few minutes on fast chains. A zero over a window that short is weak evidence and must not be read as a token being idle. Ethereum is the only chain here whose window spans a meaningful period.
THE WIDER COLUMN IS UNCONFIRMED, AND THAT IS THE POINT. The confirmed window is pinned at 100 blocks because the strictest keyed provider caps eth_getLogs at 10 blocks, and at that width every row reads 0, which is true and says almost nothing. The wider column is ONE provider over 5000 blocks, so it is context for the confirmed figure and never a substitute for it. Where the two differ, the confirmed number is the claim and the wider number is the reason to keep watching.
COVERAGE IS NOT A FINDING. How many rows reach two-provider agreement depends on free-endpoint availability at that moment, not on the tokens. Identical code 2.5 minutes apart returned 7 of 11 and then 1 of 11. Read the per-row states, never the coverage count, and never compare that count across runs.

Delisting notices, official venue announcements only. The withdrawal deadline is the actionable field, so a notice whose deadline could not be parsed still shows, saying so:

What this panel cannot answer yet, and why. The gap is retention, not difficulty: the archive recorded per-pair spreads for a fixed 8 and venue-level volume, never the pair INVENTORY, so listings and delistings left no trace. Capture starts now, which makes these answerable going forward and never retroactively. 86 cycles retained so far.

• Did liquidity persist after a listing? The archive never recorded the pair INVENTORY, only per-pair spreads for a fixed list of 8 plus venue-level volume. Retention starts with this build, so the answer becomes available going forward. 86 cycles captured so far.
• When was a pair first listed? Same cause. A first-seen date only exists from the first cycle we retained.
• Is a token CFX-eligible? The bourse publishes the eligible-asset list and we do not hold it. Inferring eligibility from what venues list would be backwards: a venue listing a token is evidence about the venue, not about eligibility.

Two different measurements, deliberately shown together. OJK counts REGISTRATIONS with licensed exchanges, which is an account count and includes dormant ones. Chainalysis ranks ESTIMATED ON-CHAIN ACTIVITY, which is a different population and a relative position, so it moves when other countries move. Neither is "users". Read the registration line for scale and the rank for momentum.

Chainalysis: Indo highest APAC YoY growth. OJK optimistic path ~26-27M by end-2026.

Indo #7 overall 2025 (was #3 in 2024). Drop is methodology: Chainalysis added an institutional >$1M sub-index favoring ETF-heavy markets (US/India/Brazil). Indo's retail+DeFi is still massive: #4 in the world on DeFi.

* = investment, not an acquisition. The starred party took a stake; it did not buy the licence holder. On a board about acquiring your way in, that is a different route with different control, and reading it as an acquisition overstates what changed hands.

8 of 29 licensed exchanges expose full public APIs; 1 market-data-only; app-first players publish none

defunct: Zipmex wound down (off current OJK list)

on-chain read (keyless, multichain): IDRT $5.02M but CoinGecko claims $9.63M = over-stated ~43%. IDRP $1.95M is 99.6% on Kaia. IDRX $1.01M spread thin over 10 chains. XIDR (StraitsX) $117k across Ethereum/Polygon/Zilliqa, ~94% on Ethereum; its Zilliqa leg is a FiatToken proxy, so supply is read off the proxy and not the implementation address the public explorers link to. IDRT near-zero daily volume. The entire rupiah-stablecoin rail is ~$8.11M.

Dates only, no forecasts. Every row restates a date the issuing body published; nothing here predicts what a regulator will do. Where a field carried several dates the LAST is used, because the transition or compliance deadline is what bites, not the signing. 2 upcoming, 22 already in force (6 most recent shown), 1 with no machine-readable date, SHOWN not dropped because the messiest entries encode staged or conditional timing and are the ones worth reading. year-only means the source gave no day and the row is sorted at year-end, not that we know the day. approx means the source hedged and that word is preserved.

RECORDKEEPING, NOT TAX ADVICE. This produces records and a gap list. It computes no liability and hard-codes NO tax rates, because rates change, differ by taxpayer and instrument, and a confident wrong rate is worse than a blank field: someone files on it. Where a rate belongs, the primary regulation is named instead. A qualified professional determines liability.
The expensive part of crypto tax is rarely the arithmetic. It is that the records are incomplete and nobody notices until a deadline. So the gap list is produced every run rather than a total resting on silence.
Canonical export columns: ts, venue, side, asset, quote_ccy, asset_amount, quote_amount, fee, fee_ccy, ref

read: Pattern: Indo is EARLY on classification/tax/AML (crypto slotted under financial-sector law, CARF, Travel Rule) but LATE + restrictive on openness (single-bourse gate, 40% foreign cap, no stablecoin framework, payment banned). The world is opening stablecoins (US/HK/SG); Indo is gating them.

This is a research matrix, not a ranking. No score and no best-jurisdiction verdict: the weights would be ours, and the same regime is an opportunity or a wall depending on whether you are a token, an exchange or a payments business. 16 cells are RELAYED from a named secondary and have not been checked against the instrument, so treat them as leads. 21 are blank because nobody has looked, which means unknown and never "no requirement". Peer source: Tiger Research, Retail Investor Landscape Across 9 Asian Markets, 2026-03-31. HTX-funded, zero citations in the document.

The empty rows are the point. 6 sectors were searched and returned nothing at all (Payments / on-ramp, Custody / wallet, DeFi protocol, DePIN, Identity / KYC, Data / analytics). 1 returned names whose domains no longer resolve (Gaming / metaverse). Those are different findings: the first says nobody built here, the second says somebody did and it is gone or moved. Payments is the sharpest result: the sector is well served and every operator the pass returned was foreign, so it is a gap in local ownership rather than in supply. What this is not: a census. The roster is one round of search on 2026-07-28 plus the venues we already measure, so exchange coverage is strong because that is what this dashboard does, and everything else is thinner by construction. Our own API census puts 29 licensed exchanges in the market against the 14 listed here. Status words are narrow: quiet means we hold no recent evidence of operation, NOT that a company is gone, and wound-down is only ever used with a link. A site that fails our probe becomes unknown, never dead.

Quick reference (the stack at a glance, full text per law below):

25 laws, sourced from Logseq Regulation/ + primer. click any law to expand its full summary. Links are official source locators, not asserted document deep links.

Indonesian coverage comes from regulators, not media. 4 Indonesian item(s) above are primary regulator documents (OJK, Bank Indonesia, Kemenkeu, Komdigi, CFX). The Telegram capture contributed 0 Indonesian items of 8 crypto-relevant across 7 channels, because it holds no Indonesian sources and Indonesian crypto media has no Telegram presence. That is a TRANSPORT gap, not evidence nothing happened locally. Telegram rows below are global context.

0.0% of the pool is the same story arriving more than once (14 claims, 14 distinct). A further 63 within-channel repeats are excluded entirely: one warning posted N times is one warning. Two dimensions are not measured and are not scored as zero:

• Correction history: UNMEASURED. We have never tracked corrections, and defaulting to 'none' would read as a clean record rather than an absent one.
• Reached Indonesian channels: UNMEASURABLE from our pool. Zero of our monitored channels are Indonesian, so a 'no' here would be a claim about Indonesia when it is a fact about our sensor.

No reviewed incidents yet, and that is a statement about this archive rather than about the market. Retention began 2026-07-25; nothing before that date was kept. Backfilling from memory or from search summaries would produce a confident history assembled from exactly the sources this dashboard tells you not to trust.

Candidate queue (auto-proposed from official notices, regulatory deltas and sustained anomalies). A candidate is not an incident. It is a thing we captured that looks like one and has not been reviewed:

The two layers are on DIFFERENT WEEKS. Term rates are 2026-W30, the reviewed semantic layer is 2026-W29. They come from separate aggregation runs, so do not read a term rise and a topic as the same week of evidence. The semantic layer lags because it is reviewed before it ships.

Reviewed semantic layer (meaning-level, sits BESIDE the raw term counts, never replaces them):

method: panel 8/14 INDO groups present in W29 and all baseline weeks; rate, not count. Public redacted chatter only. Human-reviewed static draft; no automatic semantic generation in dashboard refresh.