Questions

No. Ambang is a research instrument for practice trading. Nothing it produces is an order, a signal to act on, a recommendation, or financial advice, and no part of it is tuned to make money. A receipt is a record of what a stated rule concluded from stated evidence at a stated time, and it is worth exactly what that rule and that evidence are worth.

No, and it could not if it were asked to. Every decision is on paper. The credential Ambang holds for RYO has the scope tools:read, which grants reading market tools and nothing else, so there is no order path to misuse and no key that could sign one.

The wallet connection is the same story from the other side. It exists to claim authorship of a strategy and to sign its calibration attestation, which is a message signature and never a transaction. The product connects, asks the server for a nonce, asks the wallet to sign that message, and verifies the signature server-side. No transaction is ever requested, which is why the button says so permanently.

From RYO's hosted market tools, read live. Six tools exist on the live catalogue: market_overview, scan_market, analyze_token, deep_analysis, compare_tokens and monitor_market_sentiment_shift. Ambang reads them, freezes the responses exactly as they arrived, and evaluates every strategy in a cycle against that same saved snapshot, so two strategies can never disagree merely because they read the market a moment apart.

Every number on the product carries where it came from: the tool, the timestamp, and whether the source was live, recorded, or simulated. That label sits beside the number, never in a footnote, because a footnote is where provenance goes to be ignored. Coverage lists every logical field and the exact path in the RYO response it is read from.

RYO has no history endpoint. Every read is point-in-time, and the only backward-looking tool is fixed at seven days, so all the history this product has is the snapshots it stored itself. A first visit has nothing to compare against, and the product says so instead of filling the gap.

The field stays missing. It is never zero-filled, never carried forward from an earlier read, and never guessed. A rule that depends on it becomes not evaluable, its decision is marked degraded, and the row that would have drawn it is replaced by a ruled blank with RYO's own reason printed inside it. A missing measurement is information; a zero would be a lie about it.

This is not hypothetical. Of the fourteen logical fields Ambang defines, thirteen resolve from live RYO responses and oi_change_24h does not, because RYO marks its derivatives block unavailable. It is shown as unresolved everywhere it appears rather than quietly dropped.

A receipt is one strategy's run against one saved, unchanged market snapshot, written down in full: the evidence it used, the practice decision, the stated confidence, every computed flip condition, the review, and the hash of the receipt written before it. Because each hash covers the previous hash, the receipts form a chain, and editing any earlier one changes every hash after it. The record cannot be quietly rewritten; it can only be visibly broken.

That is a claim about tampering, not about truth. A hash chain proves that a record has not changed since it was written. It says nothing about whether the evidence was any good or the rule was any use.

Reading the chain head…

The head of the receipt chain is stamped through OpenTimestamps, which folds the hash into a transaction that lands in a Bitcoin block. Once that block is buried, the stamp proves the chain head existed no later than that block's time, and rewriting history to match it would mean rewriting Bitcoin. That is the part a hash chain alone cannot give: not only that the record was not edited, but when it existed.

It proves nothing else. It does not make a decision correct, does not certify the evidence, and does not show that any receipt was written when it claims to have been, only that the whole chain existed by the time of the stamp. The proof is a detached OpenTimestamps file over the head hash, which any OpenTimestamps client can check against Bitcoin without asking this server anything.

Partly, and the boundary is worth stating exactly. You can recompute the chain yourself: every receipt carries the hashes it was built from, and verification rehashes the whole chain and reports the first receipt that fails, if any. You can replay a receipt, which re-resolves the fields from the stored snapshot and re-runs the stored strategy spec rather than repeating a cached answer, so a replay can genuinely disagree with the record. You can check the Bitcoin stamp with a client that has never heard of this product.

What none of that establishes is that the saved snapshot matches what RYO really returned at that moment. Only RYO could confirm that, and it keeps no history to confirm it against. The honest position is that Ambang can prove its own record is internally consistent and old, and cannot prove its source was.

Empty. No receipt has been resolved into an outcome yet, so every strategy stands at n = 0: no hit rate, no Brier score, no calibration curve. The track record page draws that emptiness rather than filling it, and says what would fill it and when.

When outcomes do exist, the only score kept is calibration, grouped by market regime, phase, and stated confidence. There is no profit-and-loss leaderboard, because a run of luck reads exactly like skill on one, and calibration does not.

No. The decision, the confidence, and every flip condition are computed by code from a stored strategy spec and the saved market snapshot. A language model appears at exactly one point, as a review of the finished decision, and its authority runs one way only: it may lower the confidence or veto the side, and it may not raise confidence, invent a side, or introduce a number that is not already in the evidence. A response that tries to is discarded and the receipt records that it was.

Ambang is Indonesian for threshold, which is the mechanism the product is built around. Every receipt carries the smallest single-field change that would flip its decision, computed from that cycle's own values rather than asserted once in a sentence. When no single field can flip a decision, that is printed as a result too.

RYO Digital Limited. Ambang was built for the RYO-CHAN Virtual Hackathon 2026, and under the official rules all original work created for and submitted to that hackathon is owned by RYO Digital Limited, with each participant assigning the transferable rights in it. Submissions are proprietary and are not open source unless RYO Digital Limited approves otherwise in writing, so the repository is private.

The same rules require every AI tool, third-party library, dataset, external service, template, and piece of pre-existing code to be disclosed. Ambang's disclosures are listed in the repository README rather than summarised here, so there is one list to keep correct rather than two.