Asked to harvest the state-of-the-art liquidity, crowding and fragility candidates — concept plus a verified open-source implementation for each — for the Axis-2 axis. Delivered 120 candidates plus 7 out-of-scope across six family sweeps, frozen into a cohort and handed downstream.
| Quantity | Value | Meaning |
|---|---|---|
| Candidates harvested | 120, plus 7 out-of-scope | from a 16-candidate seed |
| Firings | 6 (12 / 20 / 18 / 20 / 15 / 19 new per firing) | |
| Downstream verdicts written | 0 of 120 | the consumer never decided any of them |
| Cohort freeze | commit e6eaa6be | the exact object the consumer took |
| Last harvest firing | 2026-07-23 | the ledger's newest row |
| Firing | What it did |
|---|---|
| 1 | Swept order-flow toxicity and informed trading; 12 new |
| 2 | Swept systemic fragility and cascade risk; 20 new, two screened out for needing leverage/funding data we do not hold |
| 3 | Swept jump, tail and ignition detectors; 18 new, all computable from spot data |
| 4-5 | Swept price impact/capacity then crypto-integrity and volume-clock bars; 35 new, and caught that one popular library is proprietary rather than open source |
| 6 | Swept order-book depth and resiliency; 19 new — total reaches 120 |
Nothing the loop owes. Its page still reads LIVE and 'never treats the list as closed' — a stale banner, not work. Whether the plan's second half (harvesting the domain-specific measurement instruments) was separately closed is not evidenced in the campaign folder.
The harvest delivered a frozen, downstream-consumed cohort and has no completion criterion by design; more firings would only widen a list that is already frozen elsewhere. That is saturation, not a stop.
No stop marker exists in this folder. The nearest is its consumer's: 'stop the loop; settle overlap first; one-candidate throwaway pilot end to end; re-freeze only on a design that has run.'
Lifecycle status is process state, not a judgement of the findings. Results are stated as numbers with their uncertainty.
Campaign regime-invariant-orthogonality (2026-07-22) · the second orthogonal axis: liquidity, crowding & fragility · append-only, self-resyncing spoke · ← back to campaign reader
The claim: a feature is orthogonal only if it stays non-redundant and keeps its meaning across market regimes. Axis-2 is the liquidity / crowding / fragility half of that paradigm — a different information axis from price ("how fragile / how liquid" vs "which way will price go"), so it is orthogonal by construction to the price-microstructure panel.
Grounding Axis-2 requires two harvests (per axis-2-cascade-free/GROUNDING-PLAN.md), both filled by the append-only loop below. Each harvested candidate carries two co-equal deliverables: the state-of-the-art concept / matrix (stable paper ID) and its state-of-the-art FOSS implementation — the best actively-maintained open-source code that computes it (repo · commit · license · language · maintenance signal). Nothing is evaluated on our data yet.
Illiquidity/price-impact, effective-spread, range-vol, order-flow toxicity, microstructure-invariance, capacity, cascade/fragility measures — computed from spot trades/OHLC only.
The 23 grounded orthogonality matrices grade any feature column for free (they don't care price-vs-liquidity). What's genuinely missing is the domain-specific instruments — a capacity metric (Days-ADV), an impact estimator (Kyle-λ), a toxicity metric (VPIN), a microstructure-invariance scale (Kyle–Obizhaeva).
All Binance data is spot, no futures. The highest-signal crowding features — funding rate, open interest, liquidation heatmaps — are derivatives → out of scope. They are catalogued for completeness but flagged, never grounded. Every candidate is tagged spot · needs-L2 · deriv (oos).
SSoT: axis-2-cascade-free/CANDIDATE-CATALOG.md. Stable IDs + SOTA FOSS-implementation status re-verified/hunted by the loop before any promotion. PF=parameter-free · DIM=dimensionless · PP=principled-parameter.
| # | Name | Family | Spot? | Param | SOTA FOSS impl |
|---|---|---|---|---|---|
| S01 | Amihud illiquidity | illiquidity/impact | spot | DIM | UNVERIFIED |
| S02 | Kyle's λ | price-impact | spot | PP | UNVERIFIED |
| S03 | Roll effective spread | eff-spread | spot | PF | mql5#150/#151 |
| S04 | Corwin–Schultz | eff-spread | spot | PF | mql5#150/#151 |
| S05 | Abdi–Ranaldo (h_norm≈0 degenerate on our cells) | eff-spread | spot | PF | mql5#150/#151 |
| S06 | EDGE (Ardia et al. 2024) | eff-spread | spot | PF | mql5#150/#151 |
| S07 | Parkinson range-vol | range-vol | spot | PF | mql5#150/#151 |
| S08 | Garman–Klass | range-vol | spot | PF | UNVERIFIED |
| S09 | Rogers–Satchell | range-vol | spot | PF | UNVERIFIED |
| S10 | Yang–Zhang | range-vol | spot | PF | UNVERIFIED |
| S11 | VPIN | order-flow toxicity | spot | PP | UNVERIFIED |
| S12 | Order-flow imbalance (OFI) | order-flow | L2 | PP | UNVERIFIED |
| S13 | Kyle–Obizhaeva invariance | microstructure invariance | spot | DIM | UNVERIFIED |
| S14 | Days-ADV (capacity) | capacity/crowding | spot | DIM | NONE (trivial) |
| S15 | CVD / taker buy-sell skew | order-flow | spot | PF | NONE (trivial) |
| S16 | Hawkes self-excitation intensity | fragility/tail | spot | PP | UNVERIFIED |
| — | Funding rate | crowding | deriv | — | OUT-OF-SCOPE |
| — | Open interest | crowding | deriv | — | OUT-OF-SCOPE |
| — | Liquidation heatmap | ignition | deriv | — | OUT-OF-SCOPE |
--readonly=2), spare-only, watchdog-guarded.Honest limit carried in: every ORTHOGONAL verdict stays regime-scoped (ICP 0/92) — "orthogonal in regime R", never "forever".
Each firing of the loop appends one row here and emits one self-contained page index_iter_<N>_<slug>.html in this folder. Rows and pages are never edited or deleted — corrections add a new sibling row citing the parent. Mirrors the markdown SSoT axis-2-cascade-free/LEDGER.md.
| Iter | Date | Family / angle | #New | Running total | Page | Note |
|---|---|---|---|---|---|---|
| 6 | 2026-07-23 | order-book depth / imbalance / resiliency (spot L2) — OBI/queue-imbalance · micro-price (Stoikov)/weighted-mid · MLOFI + integrated/cross-impact/deep OFI + DeepLOB · Næs-Skjeltorp book-slope · XLM/cost-to-trade · Obizhaeva-Wang & Large resiliency · Bouchaud-Mézard-Potters LOB-shape · queue-reactive models · TAQ 23-measure family · order-book entropy | 19 | 120 (+7 oos) | iter 6 → | L2 spot-legal under principle #8. FOSS: frds, highfrequency::tqLiquidity, LOB-feature-analysis, hftbacktest. Reference-only (no license): microprice, DeepLOB |
| 5 | 2026-07-23 | crypto-spot liquidity-quality & integrity + volume-clock bars — Cong et al Benford 1st/2nd-digit + roundness-clustering + power-law-tail wash tests · Aloosh-Li validation · Deng-Zhou liquidity jump/diffusion · Easley-LdP-O'Hara volume clock · AFML tick/volume/dollar + imbalance + run bars · P&D detection · CUSUM/Bai-Perron · Nigrini baseline | 15 (+2 oos) | 101 (+7 oos) | iter 5 → | Crypto-native. mlfinlab is proprietary → use mlfinpy. Aloosh-Li: power-law wash test misleading vs ground truth → WATCH. OOS: DEX on-chain, order-book |
| 4 | 2026-07-23 | illiquidity & price-impact + market-impact laws + capacity/turnover — Amihud-Mendelson/Amivest · Hasbrouck-IS & Gonzalo-Granger-CS price-discovery · Gibbs/FHT/LOT-Zeros/effective-tick LF spread proxies · Pástor-Stambaugh/Sadka · √-law/Almgren-Chriss/Kyle-Obizhaeva-W · VNET/MEC resiliency · crypto SOTA Brauneis/Donier-Bonart/Maitrier | 20 | 86 (+5 oos) | iter 4 → | Anchor = eguidotti/bidask (MIT). Corrections: LiquidityMeasures/liqmetrics R pkgs do not exist; mlfinlab not FOSS. Maitrier 2025 metaorder reconstruction makes 5 metaorder-gated entries spot-legal |
| 3 | 2026-07-23 | jump / tail / fragility ignition — BNS/Huang-Tauchen/Lee-Mykland/Aït-Sahalia-Jacod/Jiang-Oomen jump tests · Mancini-threshold/MedRV-MinRV/preavg jump-robust vol · BNKS signed semivariance · realized skew/kurt · Hawkes branching-ratio & Filimonov-Sornette reflexivity · Hill/Pickands/POT-GPD tail-index EVT · LPPLS crash singularity | 18 | 66 (+5 oos) | iter 3 → | All 18 spot-legal (no quotes/L2/derivatives). SSoT = R highfrequency; tick (Hawkes); evir/POT/extRemes (EVT). Endo-exo caveat: n→1 branching ratio biased up by intraday non-stationarity |
| 2 | 2026-07-23 | liquidity-cascade / fire-sale / feedback-loop / systemic-fragility — ΔCoVaR/MES/LRMES/CATFIN tail · Diebold-Yilmaz & Baruník-Křehlík connectedness · Billio Granger-net · Absorption-Ratio/RMT/Turbulence/AutoencoderRR · DCC/DECO/Correlation-Surprise/CISS · CSAD/CSSD herding · commonality-in-liquidity/LVaR spirals | 20 (+2 oos) | 48 (+5 oos) | iter 2 → | Spot-legality was the load-bearing screen: SRISK (needs leverage) + Brunnermeier-Pedersen spiral (needs funding) flagged out-of-scope; return-only legs kept |
| 1 | 2026-07-23 | order-flow toxicity & informed-trading — PIN/MPIN/AdjPIN/iVPIN · MRR/Glosten-Harris spread decomposition · BVC/Lee-Ready/tick signing · Lillo-Farmer long-memory/propagator · Sato-Kanazawa crowding | 12 | 28 (+3 oos) | iter 1 → | PIN SSoT = monty-se/PINstimation (R); Py = frds + Trade-Classification-Algorithms. Only 2023+ spot-computable SOTA = Sato-Kanazawa PRL 2023. Toxic-flow ML (FX/quotes) out-of-scope |
| 0 | 2026-07-22 | seed across 6 families (illiquidity · eff-spread · range-vol · order-flow · microstructure-invariance · capacity) + 3 deriv flagged | 16 (+3 oos) | 16 seed | — (kickoff; markdown SSoT) | seed catalog only — not harvested; loop begins at iter 1 |
findings/evolution/audits/2026-07-22-regime-invariant-orthogonality-campaign/axis-2-cascade-free/ — CANDIDATE-CATALOG.md · LEDGER.md · GROUNDING-PLAN.md · LOOP-PROMPT.mdaxis-2-cascade-free/GROUNDING-PLAN.mdaxis-2-cascade-free/LOOP-PROMPT.mdrsync of this folder only (never dashboard:deploy) · repo = SSoT · docs-only; deploy to the nasimubd site is a separate gated step.