Hyperliquid first exhibits up as a value story.
In a weak crypto tape, HYPE saved attracting consideration whereas the standard majors appeared heavier. That doesn’t show something concerning the venue’s market construction. A token chart will not be an audit path; value efficiency compresses flows, incentives, listings, buybacks, leverage, and narrative into one line.
But it surely does clarify why the venue is price opening.
The higher query will not be whether or not HYPE went up. It’s what sort of market object sits beneath that focus, and whether or not the general public e book exposes sufficient state for an outsider to reconstruct the shifting floor of the market, somewhat than simply one other change UI.
A pool snapshot and a e book snapshot reply totally different questions, and this publish is about what occurs after the snapshot. A single l2Book pull is a measurement. An audit path is a sequence.
That distinction got here out of prior AMM work, the place the recurring blocker was less complicated: you can not cause ahead from a market state till the state object is outlined. If the thing is unstable, over-abstracted, or blended in with hidden assumptions, a later monitor, rule engine, or studying loop dangers becoming artifacts as an alternative of mechanism. Hyperliquid turned the following check case as a result of its public CLOB floor appeared archivable sufficient to strive.
The AMM aspect has its personal model of this drawback. Bunni is a compact hook-side instance: its autopsy traced the failure to Bunni’s personal accounting logic, to not Uniswap v4’s core swap math. As soon as a pool provides customized protection logic on high of the AMM, that logic turns into a part of the state machine an auditor has to comply with, and it wants sequence-level audit, not a single steadiness verify.
Hyperliquid fingers the identical drawback a unique object. There aren’t any pool reserves or hook accounting right here. The state is a shifting e book: posted depth, trades, funding, open curiosity, mark-oracle premium, and reference mids pulled from different venues.
Hyperliquid begins from a unique object
The static distinction is acquainted: callable stock versus posted depth.
An AMM pool can normally be inspected as a compact state object: reserves, liquidity vary, charge tier, swap logs, pool value. A CLOB needs to be watched as a shifting floor as an alternative — e book updates, trades, funding, open curiosity, mark-oracle context, and foundation in opposition to reference mids. Cancels and reposts typically present up solely as depth that moved between snapshots, with no occasion marking the change.
On the AMM aspect, the market object continues to be a pool: reserves, curve state, LP accounting, charge logic, and now hook state. Hooks and dynamic charges are an try and make that passive stock extra conditional, which is nearer to what a CLOB order already is by default — posted at a value, for a dimension, till the maker cancels, till it fills, or till danger modifications.
On the CLOB aspect, the thing is posted depth backed by energetic makers, margin guidelines, funding, and a danger engine, not a pool defending passive stock. The seen stress indicators comply with from that: unfold, seen depth, signed movement, funding, OI, mark-oracle premium, and foundation. Neither floor is being scored as safer right here; the thing being inspected is simply totally different.
Hyperliquid paperwork the e book as HyperCore / L1 state with price-time precedence matching. The general public information API and WebSocket path are the outsider observer path – twenty seen ranges per aspect, callable from exterior, which is what makes the e book L1-settled, API-observable, and self-archivable within the first place. Archive the general public streams, normalize the rows, and label what by no means entered the artifact.
What the structure makes observable
This issues as a result of the general public stream will not be the identical factor as HyperCore itself. HyperCore consists of the order books internally; what the API exposes is an observer path, not a byte-for-byte replay of that inside state.
Three layers fall out of that. Public feeds — l2Book, trades, metaAndAssetCtxs, exterior reference mids – are the primary. Derived metrics come second: unfold, seen depth, imbalance, signed movement, funding/OI motion, mark-oracle premium, cross-venue foundation. The third layer stays hidden no matter how lengthy you acquire: maker intent, true queue precedence, non-public latency, account-level margin state, the total liquidation path, risk-engine thresholds.
Reconstructing the shifting e book
There isn’t any level reasoning ahead from a e book state if the e book state itself is just a screenshot, so the primary job right here is reconstruction, not prediction.
The workflow is apparent: archive public streams, normalize them into book-timestamped rows, be a part of nearest-prior perp context and reference mids, compute derived fields, and mark gaps as an alternative of filling them.
An area collector on BTC, ETH, and SOL pulls WebSocket l2Book and trades, periodic metaAndAssetCtxs, and Binance/Bybit perp mids as exterior anchors. Foundation here’s a reconciliation sign, not a mispricing verdict. The output is a partial outsider artifact – one row per e book statement with unfold, seen depth bands, sixty-second signed movement, joined meta fields, cross-market foundation, and hole flags.
Any future monitor, rule engine, or agent that makes use of this knowledge inherits the identical observable boundary: it can’t act on maker intent or hidden margin state, as a result of these by no means entered the artifact to start with.
The figures under use a 7-day (168 h) assortment window (2026–07–13T00:00Z-2026–07–19T23:59Z; 112,626 BTC e book rows) — the longest totally steady stretch in a roughly two-week assortment run, with no reference-feed or WS outages inside it. One joined row from the center of the run exhibits how the fields line up: at 11:59:44 UTC on Jul 16, unfold 0.16 bps; seen 10 bps depth ~$10.6M; sixty-second signed movement ~−$79k; funding joined from the nearest-prior meta ballot; HL mid inside a couple of bps of Binance and Bybit mids.
This row doesn’t clarify maker intent, queue precedence, or liquidation causality. What it does present is narrower: the e book, commerce tape, perp context, and reference mids be a part of into one defensible outsider market-state artifact.
Liquidity, leverage, and pricing stress
The joined row above reads by way of 4 surfaces. Liquidity: sub-bps unfold with thick seen near-mid depth. Stress: heavy sell-side signed movement, with the seen e book leaning ask-heavy whereas takers hit the bid. Leverage: funding from the joined meta ballot, with no apparent crowding sign in a single row. Pricing stress: HL mid monitoring exterior mids somewhat than drifting broad.
Rule-based flags on the ledger are stress screens, not accusations. Over the 7-day window, percentile flags relaxation on a firmer baseline than the unique 32 h pilot, however nonetheless in need of the 30-day goal; a 30-day archive stays the precise goal for any market-structure declare.
What stays hidden
Even with a working collector, the outsider nonetheless doesn’t see maker intent, true queue precedence, non-public latency, account-level margin distribution, the total liquidation path, or risk-engine thresholds.
Public depth can disappear with no commerce. Public fills carry no maker labels. Cross-venue foundation diverges for causes that aren’t robotically mispricing or routing alpha — the CPI row above is one occasion of a foundation spike with a believable however unverified clarification.
That incompleteness will not be a footnote to wash up later; it’s a lot of the level of conserving the ledger in any respect. What the ledger is sweet for will not be seeing every little thing, however recording precisely what entered the artifact and what stayed exterior it, and doing that repeatedly sufficient that the boundary itself turns into measurable over time as an alternative of guessed at from one screenshot.
Reconstructing the general public e book is beneficial not as a result of it restores the venue’s full inside fact, however as a result of it turns an opaque stream right into a state object that may be monitored, in contrast, and stress-tested — the unfold and depth spikes in Fig. 2 double as alert thresholds, and the flat-OI-plus-volume-burst learn in Fig. 4 is one occasion of narrowing a proof, not simply describing a chart. That is sufficient to assist alerting, occasion attribution, and execution or danger choices, with no need a accomplished image of maker intent behind them.
Closing
The purpose was by no means Bunni itself. It’s the object an auditor has to comply with.
On the AMM aspect, the dwell questions are dynamic charges, hooks, and LP loss surfaces — state that solely is sensible learn throughout a sequence. On the CLOB aspect, the dwell questions are a shifting e book, funding and OI, mark-oracle premium, foundation, and the hidden venue state behind them. Each lanes fail the identical approach when the audit object is left imprecise: reasoning ahead from a screenshot.
A public e book doesn’t remedy the audit drawback a lot as relocate it. The following verify on this knowledge will not be a much bigger dashboard — it’s whether or not a 30-day model of the identical ledger holds up the CPI-basis learn above as a repeatable sample (macro print → quantity burst → foundation spike → flat OI) somewhat than a one-window coincidence.
Appendix: assortment and reproducibility
- Window: 2026–07–13 → 2026–07–19 UTC (7 days, 168 h). Cash: BTC, ETH, SOL — roughly 112,600 rows every. This was the longest stretch in a ~2-week assortment run with no reference-feed or WebSocket outage inside it; these gaps present up elsewhere within the run as lacking rows in Fig. 4, not as a fabricated foundation studying.
- Limitations: seen l2Book ranges solely; nearest-prior be a part of for perp meta and reference mids, dropped somewhat than carried ahead previous 5 minutes of staleness; a 7-day run, not but the 30-day market-structure research this factors towards.
- Replica: Gist appendix (archive_io.py, collect_ws_ledger.py, collect_ref_mids.py, build_ledger.py, build_figures.py).
- Bunni autopsy: weblog.bunni.xyz
- Jul 14 CPI/BTC context: CoinDesk
This publish was initially revealed on my private weblog: https://egpivo.github.io/2026/07/26/when-the-book-keeps-moving.html
When the Guide Retains Shifting. What Hyperliquid Lets Outsiders Reconstruct. was initially revealed in The Capital on Medium, the place individuals are persevering with the dialog by highlighting and responding to this story.
