Arunights Research — Risk Grade Methodology v0.2.1
Effective 2026-07-11 (v0.2), amended 2026-07-16 (v0.2.1, additive). This document is versioned; grades cite the version they were produced under. Changes are logged in the changelog at the bottom.
What these grades are
An Arunights Risk Grade is a published, evidence-linked opinion about the resilience of an onchain asset or entity to loss events: depegs, redemption failures, counterparty failures, and collateral contagion. It answers one question: if stress arrives, how likely is a holder to get their money back at par?
These grades are not credit ratings, not investment advice, not an offer or solicitation, and not produced under any regulatory license. No issuer pays for a grade. We hold no positions in rated assets and do not trade them. Every input to every grade is published alongside the grade.
The scale
One universal scale across all asset classes, so a B+ vault and a B+ dollar token express the same judgment about loss resilience.
| Grade | Score | Meaning |
|---|---|---|
| A+ | ≥92 | Highest resilience. Verifiable backing, proven redemption under stress. |
| A / A− | 85–91 / 78–84 | Strong. Failure requires multiple simultaneous breakdowns. |
| B+ / B / B− | 71–77 / 64–70 / 57–63 | Sound but with identified structural dependencies. |
| C+ / C / C− | 50–56 / 43–49 / 36–42 | Fragile. Holders are exposed to single points of failure. |
| D | 25–35 | On watch. Structural red flags consistent with pre-failure patterns. |
| F | <25 or realized event | Failed. Realized loss of peg/principal. Tombstone grade. |
Every grade carries two companion labels:
Confidence (high / medium / low) — the quality and verifiability of the data behind the grade. A B with low confidence is a different statement from a B with high confidence. Incumbent raters do not publish this; we consider hiding it dishonest.
Status — draft (internal, unverified), provisional (published with open verification items), published (fully verified, right-of-reply completed), tombstone (post-failure record). Drafts are never distributed as conclusions.
Pillars and weights by asset class
Yield-bearing / synthetic dollars (yield_dollar)
| Pillar | Weight | What is scored |
|---|---|---|
| Backing & reserves | 25% | What actually backs the token; attestation cadence and auditor quality; proof-of-reserve verifiability; whether backing is exogenous (T-bills) or endogenous (own ecosystem assets). |
| Redemption & exit | 20% | Direct redemption rights and their gates/cooldowns/minimums; secondary market depth relative to circulating supply; behavior of exits under past stress. |
| Counterparty & custody | 15% | Custodians, prime brokers, exchanges, and external managers in the loop; bankruptcy remoteness; the Stream lesson — discretionary external managers are the deadliest single flag. |
| Peg & market infrastructure | 15% | How lending markets price it (hardcoded 1:1 oracles are a systemic amplifier); depeg history; liquidity fragmentation. |
| Leverage & rehypothecation | 15% | How much of supply is looped as collateral elsewhere; recursive minting; whether the asset is a building block in leverage towers (the xUSD contagion pattern). |
| Transparency & governance | 10% | Named legal entity and team; docs quality; admin keys, upgradability, timelocks; disclosure of strategy changes. |
Curated vaults (vault)
Collateral quality 25%, exit liquidity 20%, concentration 15%, oracle risk 15%, curator track record 15%, transparency 10%.
Curators as entities (curator)
Realized loss history 25%, risk process 20%, disclosure quality 15%, concentration of strategies 15%, tenure & scale 15%, alignment & conflicts 10% (who pays them, fee structures, whether they rate/curate their own products).
Tokenized RWA wrappers (rwa_wrapper)
Legal structure & investor recourse 25%, underlying asset quality 20%, redemption mechanics 20%, counterparty & custody 15%, onchain implementation 10%, transparency 10%.
Structural overrides (grade ceilings)
Weighted averages hide fatal flaws, so specific findings cap the grade regardless of score:
| Flag | Ceiling | Rationale |
|---|---|---|
realized_loss_event |
F | A realized loss of peg or principal is a fact, not an opinion. |
undisclosed_external_manager |
C | The exact Stream Finance failure mode. |
unverifiable_reserves |
C+ | If backing cannot be verified, resilience claims are faith. |
hardcoded_oracle_collateral |
B | Assets priced at hardcoded $1.00 in lending markets while used as collateral at scale amplify contagion. |
no_redemption_thin_liquidity |
B− | No direct redemption plus shallow secondary markets means exits fail exactly when needed. |
admin_keys_no_timelock |
A− | Uncapped upgrade power is a tolerable but real risk. |
Positive structural flags (v0.2.1, no ceiling — labeled evidence, not a mechanical score change):
| Flag | Effect | Rationale |
|---|---|---|
regulated_emoney |
None — informational | The asset is issued under a statutory e-money regime with an enforceable par-redemption right (e.g. a MiCA Electronic Money Token, or a MAS-regulated Singapore stored-value facility). This does not add points automatically — the engine does not auto-score prose evidence — but it is a fact the analyst is instructed to weigh positively when hand-setting the Redemption & exit pillar score, the same way every other piece of qualitative evidence already works in this system. It exists so the fact is labeled and evidence-linked consistently across every asset it applies to, rather than buried inside free-text pillar notes. |
Every override — ceiling or informational — appears on the asset page next to its evidence, exactly like every other flag; there is no separate, less-visible category for the positive one.
Regional pegs, non-USD instruments (v0.2.1, additive)
Coverage extends beyond USD-pegged instruments to regional fiat pegs (SGD, AED, AUD, EUR, and others as coverage grows). Every asset carries a peg_target (defaults to USD for every asset graded before this version — no existing grade is affected). Two mechanical consequences, both computed by the quant model-signal layer, never by hand and never moving a published grade on their own:
- Peg deviation is measured against the target fiat, not against the US dollar. A SGD-pegged instrument's deviation is computed against the daily USD/SGD rate (ECB reference rates via frankfurter.dev, snapshot-cached like every other model-signal input), not against $1.00. A perfectly-held SGD peg reads as ~0 deviation; a real depeg still reads as a real depeg — the FX adjustment corrects for currency, it does not launder actual instability.
- Thin-DEX assets get a documented liquidity fallback. Where no reliable DEX depth/turnover data exists — common for regional pairs that clear mainly on CEXs or via direct issuer redemption — the exit-capacity model signal falls back to CEX listing breadth plus the analyst-supplied redemption-gate terms (direct redemption? cooldown, minimums, whitelist). That fallback signal's own confidence is auto-capped at medium, regardless of how strong the underlying CEX/redemption data looks — a proxy for a proxy is never treated as high-confidence by policy, not analyst discretion. This caps only the model signal's internal annotation; it never overrides the analyst-set, human-owned
confidencefield published with the grade.
Full technical detail: pivot/09_REGIONAL_EXPANSION_SPEC.md.
Process
Grades are produced from public, primary evidence: onchain data, official documentation, attestation reports, and reputable reporting — every pillar score links its sources. Before an asset moves from provisional to published, the issuer receives a right-of-reply (72 hours minimum) with our draft findings; factual corrections are incorporated, opinions are not negotiable. Every published grade includes a falsifier: the specific evidence that would change it. Errors are corrected publicly and logged; grades are re-reviewed on material events and at minimum quarterly.
Calibration
A ratings scale where everything clusters at C is as useless as one where everything gets an A. Target distribution across mature coverage: roughly 15% A-range, 40% B-range, 35% C-range, 10% D/F. Grades are relative to the universal scale, not graded on a curve within a class — if the entire class is fragile, the class scores fragile.
Haircuts (v0.2)
A haircut translates a grade into a collateral-management number: the discount applied to an instrument's market value when counting it as collateral or measuring risk-adjusted exposure. Arunights haircuts are illustrative policy inputs, not margin requirements — a design partner calibrates them to their own risk appetite. The derivation is fully mechanical from published grade components, so every haircut is reproducible:
Base haircut by grade band:
| Grade | A+ | A | A− | B+ | B | B− | C+ | C | C− | D | F |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Base | 2% | 4% | 6% | 10% | 14% | 20% | 30% | 40% | 55% | 75% | 100% |
Confidence multiplier (data quality is a risk): high ×1.00, medium ×1.25, low ×1.50.
Class adjustment (structural liquidity of the wrapper): yield dollars +0pp, RWA wrappers +2pp (redemption/settlement lag), vaults +3pp (exit-queue risk). Curators are entities, not instruments — no haircut; curator grades enter as counterparty flags on positions they manage.
Flag add-ons: no retail redemption +3pp; hardcoded-par oracle +5pp; unverifiable reserves +10pp; realized loss event → 100%.
Final haircut = min(100%, base × confidence multiplier + class adjustment + flag add-ons), rounded to the nearest 0.5pp. Sanity anchor: this places high-grade regulated stablecoins near the 2–5% range and mid-B instruments in the 15–25% range — the same neighborhood as prevailing DeFi money-market LTVs — while pushing unverifiable or realized-loss instruments toward exclusion. Calibration is explicitly v0: the falsifier for the whole schedule is realized loss-given-stress data as it accumulates.
Independence
No issuer fees for grades, ever. Monitoring or data-access services, if ever offered, are disclosed on the asset's page and never influence grades. Funding sources for this research are disclosed publicly. The analyst holds no positions in rated assets.
Changelog
- v0.2.1 (2026-07-16) — Additive regional-expansion amendment:
peg_targetfield (defaults to USD, no existing grade affected) with FX-aware peg-deviation model signal for non-USD pegs; documented CEX-breadth + redemption-terms liquidity fallback for thin-DEX assets, auto-capped at medium confidence; new informational (non-ceiling)regulated_emoneyflag as a positive, evidence-linked input the analyst weighs in the Redemption & exit pillar. No changes to grades, scores, pillar weights, or any existing flag's ceiling. Full spec:pivot/09_REGIONAL_EXPANSION_SPEC.md. - v0.2 (2026-07-11) — Added haircut derivation (grade-band base, confidence multiplier, class adjustment, flag add-ons). No changes to grades or pillar weights.
- v0.1 (2026-07-10) — Initial methodology. Four asset-class modules, universal scale, override system, confidence and status labels.