MARKET REGIME CALMDEPOSIT MIGRATION ALARMLEVERAGE ALARMSTABLECOIN RAILS WATCHBOND BOOK WATCHDEALER CAPACITY WATCHBONDHOLDERS WATCHBOND MARKET CALMCORPORATE TRANSMISSION CALMHOUSEHOLD CREDIT WATCHSHORT PRESSURE 0 FLAGSCRYPTO EXPOSURE 1 FLAGTBTF REGISTER 15 FLAGSNEWS WATCH NO READcaptured fallback · 9 Aug 2026; live fetch replaces itUS bank layer: the 2026 scoreboard, two flagged a year early, two misses published →Undertow: the cross market liquidity map, sealed calls and misses kept, verify the record →
Institutional failure, read in event time reviewed cut · 09 aug 2026

Ratings move last. The evidence moves first.

LiquiLens reads public filings, funding structure and market signals as a chronological record—then shows exactly where the warning holds, where it misses, and where the evidence stops.

Download radar data

VALIDATED_NOT_COMPLETE is the current Lab status. The execution pass is sealed; independent validation and retrospective eligibility are not implied.

latest replay / sealedRUN D34D2AE66F9F
EVENT FIELD5,273
EVENT CHECKPOINTS42,162
10 CONTRACTS421,620 evals
scored outcome genuine forward row context / abstention evidence blocked
Open every product result
early warning · replayed record
LiquiLens flags the record breaks median diagnostic lead
Drawn to the served median of the 48-institution construction-PIT diagnostic: 21.5-month median lead. A schematic of the historical record, not one lender's chart and not validated-backtest evidence.
Reviewed 9 August 2026 · ten-product replay

Every result is here. Including the gaps.

The newest sealed pass evaluates ten product contracts across four horizons under baseline and one-day added-lag timing conditions. Select a row to inspect what was scored, what abstained, and why none of the seven requested histories is presently eligible for retrospective backtesting.

VALIDATED_NOT_COMPLETE execution sealed · definition incomplete
event field5,273canonical events
evaluation field421,62042,162 checkpoints × 10 contracts
accounted gaps413,8280 unexplained kinds
robustness layer82,55614 claim blockers remain
event coverage32 / 47national jurisdictions
Read the exact integrity and coverage receipt +

One sealed ten-product replay pass contains 421,620 evaluations across 5,273 canonical events, producing 413,828 exactly bound gap rows and 51,870 root causes, alongside 82,556 robustness audit records. Named-event coverage is 32 of 47 national jurisdictions, with 4,191 verified event identities / 1,082 unresolved mappings. Three products have 4,146 genuine prospective forward records; 0 of 7 requested products is eligible for retrospective backtesting. 14 robustness claim blockers remain; there were zero model changes.

This is an integrity and coverage receipt, not a performance-improvement or promotion claim. Read the machine-readable receipt →

Outcome field captured correct rejection missed false positive not scored
30d90d180d365dbaseline + added lag
INSTITUTION RISK · PRODUCT 09

Undertow — bank legacy

RETROSPECTIVE: INELIGIBLE

The legacy bank record produced the largest scored slice in this replay. Its generic threshold sweep cannot reproduce the native ranking rule, and the panel does not contain one coherent as-published revision for every bank-quarter.

evaluations42,162
scored8,976
captured4,026
missed478
false positive791
gap rows34,455
What this establishes

An internally consistent evaluation record with exact outcome bindings and a visible case-review trail.

What it does not establish

A vintage-complete retrospective performance claim or independently validated model.

Each of the 421,620 outcomes is one contract evaluated at one generated event checkpoint; the total is not a count of unique product-event pairs, institutions, or backtests. “Captured” is displayed only where a score cut and matching outcome are present; absent evidence remains absent.

SEALED run_d34d2ae66f9f · receipt sha256:2a726eeec94d… 4,191 identities verified · 1,082 unresolved · zero model changes Machine-readable replay atlas →

The warning period, to scale

  1. 01 / 06 41 months

    Supervision moved before the headline.

    In the published construction-PIT diagnostic, the RBI's own PCA tripwires fired a median of 41 months before the regulator moved.

    RBI PCA · CONSTRUCTION-PIT
  2. 02 / 06 27 months

    The funding structure broke first.

    Funding fragility flagged Altico 27 months out in the same diagnostic, while reported bad loans still sat near zero.

    ALTICO · FUNDING FRAGILITY
  3. 03 / 06 21.5 months

    The cohort clears the noise.

    The 48-institution construction-PIT diagnostic records a 21.5-month median alert across the replayed warning set.

    48 INSTITUTIONS · MEDIAN ALERT
  4. 04 / 06 17 months

    The institution score leaves investment grade.

    Yes Bank moved below investment grade on the LiquiLens score 17 months before the moratorium weekend.

    YES BANK · SUB-INVESTMENT GRADE
  5. 05 / 06 8 months

    The network catches the strain.

    The same public record produced a DHFL network signal eight months out, before the outcome reached the headline.

    DHFL · NETWORK SIGNAL
  6. 06 / 06 the record breaks

    The evidence becomes the event.

    The public record breaks. The warning horizon closes; the diagnostic is preserved with its misses and construction limits intact.

    T−0 · PUBLIC OUTCOME

Every marker is a published number from the historical diagnostic. This sequence is evidence of warning behavior inside a construction-PIT replay—not validated-backtest or real-money eligibility.

Historical evidence · served claim boundary

The number never travels without its boundary.

The public API serves current screening scores and three separate historical records. Each record carries its own cohort, construction method and eligibility flags. Select or focus a passport to expose its evidence state.

current screening historical diagnostic Lab / research
IN / 01
India · 48 institutions

Institution diagnostic

88.9% recall21.5 mo median lead
PERIOD_END_PROXY_CONSTRUCTION_PIT

Filing availability is proxied rather than reconstructed from a complete as-published archive.

Validated backtest NOReal money NO
US / 02
United States · 552 failures since 2008

Bank-failure diagnostic

72.8% recall0.854 pooled AUC21.7 mo median lead
CURRENT_AMENDED_CONSTRUCTION_PIT

The replay uses currently available amended call-report vintages, with misses named.

Validated backtest NOReal money NO
EU / 03
Europe · 7 institutions

Named case files

7 named records0 cohort estimates
NAMED_CASE_FILES_CONSTRUCTION_PIT

Northern Rock through Credit Suisse: auditable files, not a population performance estimate.

Validated backtest NOReal money NO
AUTHORITATIVE SURFACEapi.liquilens.in/api/evidence/markets ↗npx liquilens --recordIf prose and payload differ, the payload wins.
The gap

Public filings can surface deterioration before ratings change.

The IL&FS, DHFL, Yes Bank and Altico records contain changes in funding structure and balance-sheet quality before their public outcomes. LiquiLens standardizes those disclosures into chronological signals. The historical API labels the replay PERIOD_END_PROXY_CONSTRUCTION_PIT because filing availability is proxied rather than fully reconstructed from as-published vintages.

Regulatory databases provide the underlying filings, but teams still have to reconcile dates, definitions and missing periods. LiquiLens performs that reading systematically and publishes both successful and missed historical alerts. Its public record includes the two 2026 US balance-sheet failures, flagged 13 and 19 months in advance.

The numbers · the first market

Market context

25 lakh crore

India's MSME credit gap, the shortfall between credit demanded and credit supplied by formal lenders, as of March 2025.

Deloitte analysis, 2025 · Standing Committee on Finance put it at ₹20 to 25 lakh crore
14%

Share of Indian MSMEs with access to formal, institutional credit. The rest borrow informally or not at all.

Deloitte, 2025
~30% of GDP

What MSMEs contribute to India's economy, along with roughly 45% of exports, across more than 6 crore registered enterprises.

Ministry of MSME, via PIB, July 2024
27.25 lakh crore

Bank credit outstanding to MSMEs as of March 2024, growing over 12% a year. The book is large, and it is getting larger faster than the tools that watch it.

RBI Deputy Governor Swaminathan J, November 2024

And the rupee that is already in the bank earns nothing while it waits: RBI's Master Direction on Interest Rate on Deposits prohibits banks from paying interest on current account balances. That is the idle side of the same gap.

What LiquiLens is

Bank and lender early warning, with public evidence boundaries.

LiquiLens converts public filings, funding structure and market signals into institution-level risk screens. Its construction-PIT diagnostics report 88.9% recall across 48 Indian institutions and 72.8% recall across 552 US failures since 2008; Europe is seven named case files, not a cohort statistic. Both eligibility flags are false for all three records. Environment data such as monsoon conditions and market stress is displayed as context but kept outside institution scores. Controlled pilots evaluate cash forecasting, liquidity risk and credit workflows using each customer's own data.

One evidence discipline, market-specific models

A common claim vocabulary, fitted to each market.

LiquiLens began with Indian NBFC diagnostics, extended into US bank work, and now develops market-specific studies across additional countries and institution types. The portable layer is the evidence discipline: chronological evaluation, source tracking, coverage limits, status and eligibility flags. Disclosures, regulatory definitions and fitted parameters remain market specific.

The engine room

Ten engines, each grounded in published research.

01

Forecasting

Multi horizon cash and liquidity forecasting, so the buffer is sized before the gap appears.

02

Liquidity at Risk

Monte Carlo simulation of shortfalls across thousands of paths over a rolling horizon.

03

Liquidity Credit Score

An LCR and NSFR driven score that reads buffer adequacy, funding stability and concentration.

04

Contagion Radar

Predicts which counterparties will delay payment before it lands, and sizes the line to pre arrange.

05

Liquidity Mesh

Matches surplus to deficit and prices it inside the bank spread, so both sides keep what the spread used to take.

06

Early Warning

A five signal lender suite. Trajectory, contagion, stacking velocity, fraud rings and single party concentration, one watchlist under one false alert budget.

07

Failure Probability

A probability of failure for every institution with a vetted filing, fitted on the real collapses in the record and served with its confidence intervals, not just a point.

08

Regulator Tripwires

Distance to the regulator's own action zones, the RBI's PCA and SAF thresholds in India, in plain percentage points. In the construction-PIT diagnostic, those public tripwires fired a median of 41 months before the regulator moved.

09

Funding Fragility

Reads the liability side for deposit runs and rollover freezes. In the same diagnostic it flagged Altico 27 months out while reported bad loans sat near zero.

10

Market Distance

A daily market implied distance to default for 25 listed lenders, so the board reprices between filings instead of waiting for the next quarter.

And the bench keeps refilling. A July 2026 research pipeline implemented thirteen recent papers as deterministic engines and research surfaces: survival based PD term structures, conformal alarms with finite sample guarantees, solvency and liquidity run decomposition, multilayer contagion analytics and a depositor run simulator among them. Each one stays display and research only until it passes a pre registered product gate study, the same bar that keeps monsoon data out of the scores today. The first has now crossed that product gate. The conformal alarm joined the watchlist tier in July 2026, wired deliberately narrow: it can lift a name off green, never demote one, never escalate past what the other signals justify, and the sign off that admitted it is published at api.liquilens.in/api/evidence/gates. A product gate controls scoring behavior; it does not change the separately served historical-evidence eligibility flags.

Research grounded · calibrated per market · built to survive a real risk review.
The regulatory frame the engines compute against: Basel III LCR standard (BIS, 2013) · RBI liquidity risk framework for NBFCs (2019) · RBI Scale Based Regulation (2021).

The environment layer

Institution signals and external conditions remain separate

An Indian microlender's repayments live and die with rural cash flows, and rural cash flows live and die with the kharif harvest. So the board now watches the world around the institutions, not just the institutions: monsoon and ENSO, food and fuel prices, funding conditions and political stress, refreshed every six hours from free public sources. Coverage is honest by construction. When a feed goes dark, the board says a feed is dark. Absence is never rendered as calm.

The public signals layer

What the public record can show

The public layer tracks a growing catalog of evidence, including bond marks, disclosed short positions, market-liquidity providers, sovereign holdings, deposit movement and systemic-importance designations. Each threshold includes its source in the API payload. These observations sit beside institution scores and do not become private-book or supervisory data by implication.

CALM Market regime · daily

The tape, read for turns

A sequential change detector watches how the bank index moves against the broad market, the rupee, crude, US yields and volatility, and calls the regime. The captured fallback is CALM as of 7 August 2026; the browser replaces it from the live pulse endpoint when available.

Request the regime card →
NO READ News watch · per institution

The press, read for pressure

Every covered institution gets a news attention surge score against its own baseline, the tone of that coverage, and the share of it using distress language. Coverage today is partial, and the board says exactly which names it covers and which it cannot see yet. Thin coverage is flagged, never smoothed.

Read the live feed →
WATCH Stablecoin rails · daily

The dollar rails, watched at the issuer

Issuer level early warning for onchain dollars: peg discount, redemption run velocity, and how concentrated each coin's chain footprint is, under published thresholds. Built for the treasury migration nobody's bank statement shows yet. The board opened at WATCH on its first read and said why.

Read the live feed →
1 FLAG Crypto exposure · cited register

Which banks touch crypto, and how hard

A cited per institution register of crypto exposure, channel by channel, joined to the US run risk board. The compound flag fires where an active crypto deposit channel meets a fragile funding base, and it fired on day one. Meme coin holdings are served as a stated absence, because nobody can observe them honestly.

Read the register →
WATCH Bond book · per bank, quarterly

The SVB arithmetic, run on every bank

Unrealized bond losses against tier one capital for 1,487 FDIC banks, straight from the public filings, compounded with how runnable each deposit base is. Neither number alone killed SVB. Both together did. A rate nowcast marks the book to today's yield and wears an estimate label, because that is what it is. Where the arithmetic puts a named bank's unrealized losses above its tier one capital, the payload's flag says that and nothing more: the bank is operating and FDIC insured, the figure is a mark to market subtraction on filed numbers, and it moves no score.

Read the board →
0 FLAGS Short pressure · per institution

Who is betting against which bank

FINRA short interest and daily short sale flow per US bank, joined to run risk. In Europe the registers name names: which fund is short which bank, above half a percent, by law. The rule is rising pressure on a fragile name, never a level alone. SVB's shorts were flat while it died, and the board says so.

Read the board →
WATCH Dealer capacity · weekly

The desks everyone assumes are deep

How much spare intermediation capacity the primary dealers have left, from the New York Fed's own weekly numbers, next to a cited register of how few firms actually make the markets. When the warehouse runs low, the exit gets expensive for everyone at once. It read WATCH on arrival.

Read the board →
ALARM Hedge fund leverage · weekly + quarterly

The crowd, measured against itself

Hedge fund leverage from the OFR's Form PF aggregates, and gross short positioning across fourteen futures markets scored against each market's own history. Six sat past their 95th percentile stretch on the first read, so the board opened at ALARM and listed every one. Sterling 1992, the yen unwind of 2024 and Archegos ride along as case files.

Read the board →
WATCH Bondholders · who holds the debt

The basis trade, sized and clocked

Leveraged fund Treasury shorts from CFTC positions plus the Cayman custody line, where book size alone can only ever reach WATCH and only a fast unwind can ALARM, calibrated to the pace the Bank of England documented in the gilt crisis. The UK leg is a cited register because no live feed exists, and it says so.

Read the board →
ALARM Deposit migration · India, fortnightly

Where the deposits are walking

Credit growing five points faster than deposits is the RBI's own flagged condition, and the board opened at ALARM on exactly that. Mutual fund flows against deposit growth, deposit rate chase, margin compression, each channel honest about its lag. The one series with no public source ships as a stated gap, not a guess.

Read the board →
15 FLAGS TBTF register · annual cadence, stated

Too big to fail, on the record

Every G SIB with its bucket and surcharge, India's D SIBs, Europe's sample, each joined to whatever live signal is firing on that name. No live TBTF score is served because none exists to compute, and pretending otherwise would be theater. Credit Suisse sits in the register permanently. Designation is not protection.

Read the register →
By design Display only · stated in every payload

Warning, not verdict

None of these display-only layers moves a score or tier by so much as a decimal until a pre-registered lift study on the crisis record says it has earned the right, the same bar that kept monsoon data out. Every payload states this about itself, so nothing downstream can mistake context for a score.

Query the layer →
Historical diagnostic on the public record

A 48-institution construction-PIT study

Rules are frozen before scoring and post-default periods are walled off, but filing availability is proxied rather than reconstructed from a complete as-published archive. The served status is PERIOD_END_PROXY_CONSTRUCTION_PIT; validated-backtest eligible: NO; real-money eligible: NO.

16 of 18

Failures flagged ahead of the record, excluding books concealed by outright fraud, at a median lead of 21.5 months. Significant at p < 0.001, discriminating at AUC 0.713 against healthy controls, and robust across a broad sweep of thresholds rather than one lucky setting.

48-institution construction-PIT diagnostic across two decades and three cohorts
Two dials

Conservative mode caught 11 of 16 with zero false alarms on healthy controls. Sensitive mode adds the funding lens and caught 16 of 18, at the price of three healthy names warned. Both settings ship together, so a risk head chooses the trade instead of inheriting ours.

Alert modes served live at api.liquilens.in/api/evidence/alert-modes
0 false alarms

On the 2025 microfinance cycle, a chronologically later replay covered 11 institutions after the rules were frozen. It remains inside the same filing-availability-proxied construction-PIT boundary.

2025 cycle tranche, vetted primary filings, construction-PIT
17 months

Lead on Yes Bank in the earlier network signal study, where 7 of 8 NBFC defaults of 2018 and 2019 were flagged at a median of 6.5 months, DHFL at 8 and Altico at 11.

Rating timeline case study, phases 1 and 2, predictions pre stated

How far ahead the flag flew

Named leads from the rating timeline study, in months before the public record moved. Predictions pre stated, sources cited in the Evidence record.

Yes Bank
17 mo
Altico
11 mo
DHFL
8 mo

Median across the study: 6.5 months on 7 of 8 defaults of 2018 and 2019. The separate 48 institution balance-sheet diagnostic reports a median of 21.5 months. Full record, misses included, in the Evidence tab, which is granted on request.

These are studies on public data, not claims about a customer's private book. One test covers a stressed lender that transferred bad loans to an asset reconstruction company and then reported a sharp recovery; the method prevents that transfer alone from producing an improved grade. Paid pilots run a separately pre-registered evaluation on the lender's own data and infrastructure.

The live board

The live public signals layer

Below is the MFI Risk Board from the public screening API, fetched as this page loads. The same ensemble that replayed two crisis cycles is pointed at today's filings: every registered NBFC MFI with a vetted dossier is scored on its latest public filing, with the score trajectory drawn quarter by quarter. Coverage is stated, never implied. A row falls off when its filings go stale instead of sitting there looking current.

connecting to the public board…
The board answers from the same servers that run the product. If nothing renders here, the honest reading is that the API did not answer from your network, and the full board is in the demo, which is granted on request.

Scores are disclosure based reads of public filings, not credit ratings, and every row names exactly which factors scored it. Institutions the ensemble cannot score honestly are excluded with the reason stated, not hidden. The full board, with dossier notes, sources and exclusions, is in the demo's Universe tab.

The failure radar

One risk board across lender types

Below is the Failure Radar from the same public screening API: every institution with a fresh vetted dossier, whatever its licence type, read four ways at once. A failure probability fitted on the real collapses in the record. Distance to the RBI's own action zones. Funding fragility from the liability side. A market implied default distance for listed names, repriced daily from live equity prices. The watchlist tier is a published rule over published components, so "why is this name orange" always has an answer.

connecting to the public radar…
The radar answers from the same servers that run the product. If nothing renders here, the honest reading is that the API did not answer from your network, and the full board with every component is in the demo, which is granted on request.

Failure probabilities are fitted on the vetted record and corrected to a stated population base rate; they are screens, not ratings. The market column is a Merton distance to default from live equity prices, shown only for listed names, and the board says why any lens is dark for a given row. Method, the construction-PIT diagnostic and the model card are public: historical diagnostic · model card.

Exposure watch

Overlay public signals on your exposures

A treasury or credit team records its exposure to partner banks, reserve custodians, borrower NBFCs and MFIs. Exposure Watch joins those amounts to the relevant public risk screen, compares each daily reading with its stored baseline and routes changes through governed alerts. It does not calculate a new institution score. An exposure that lacks public coverage is marked unseen rather than calm, and each alert is retained in the verified watch log.

Tenant scoped and fail closed, like everything else that touches a book. Available to pilot tenants today through the product API.

Who it is for
For lenders

NBFCs and banks

The early warning, credit scoring and drift monitoring that the RBI's push on early warning systems and expected credit loss now expects, ready before the regulator asks.

For businesses

MSMEs sitting on idle cash

The treasury function a large corporate has and a small business never could, so working capital earns overnight instead of resting in a current account.

Enterprise grade

Security and deployment controls

A lender's regulatory book is not demo data. So the platform is engineered like the infrastructure it claims to be, and the claims below are enforced by tests that run on every commit, not by a policy document.

ISOLATION

Every tenant sealed off

Each lender's CRR, SLR, ALM and reporting book is scoped to that lender alone. Cross tenant isolation is proven by automated tests on every commit, and a probe for another tenant's report returns nothing, not even confirmation it exists.

SECURITY

Locked down by default

Short lived tokens with rotation and role based access, account lockout, HTTPS enforced everywhere with HSTS, and dependency and secret scanning in CI and weekly scheduled sweeps. Deployment builds refuse to boot in an unsafe configuration.

PROVENANCE

Provable to a regulator

A hash chained point in time ledger records every watchlist and alert as it was published. When a review asks what the engine said and when, the answer is cryptographic, not archaeological.

CONTINUITY

Observable and recoverable

Deployment builds include live metrics, error tracking, repository-specific release gates and nightly database backups, with a self-hosted option that keeps a lender's book inside its perimeter.

Try a piece of it

When does the RBI's LCR clock start for you?

The RBI framework applies Liquidity Coverage Ratio requirements to all deposit-taking NBFCs and, subject to RBI exclusions, to non-deposit-taking systemically important NBFCs with ₹5,000 crore or more in assets. Excluded classes include CICs, Type I NBFCs, NOFHCs and SPDs. This calculator is a planning screen for that published threshold, not a regulatory determination.

"…sufficient High Quality Liquid Assets to survive any acute liquidity stress scenario lasting for 30 days."

What the buffer is for, in the regulator's own words. RBI, Liquidity Risk Management Framework for NBFCs and CICs, 4 November 2019

Runs entirely in your browser. Nothing you type leaves this page. Educational, per the RBI liquidity framework for non deposit taking NBFCs; deposit taking NBFCs carry LCR at any size. The full LCR, HQLA and structural liquidity stack is what LiquiLens computes live. This widget is the doormat, not the house.

Why now

Public data is available; systematic monitoring remains uneven

India built the payment rails and the regulator is rewriting how lenders manage risk. The plumbing to put idle cash to work and to watch a live book for trouble is the part still missing for everyone but the largest players. That is the gap LiquiLens was built to close.

One desk, two instruments

Private product, public evidence

The same team maintains Seiche, a free and open-source monitor for US dollar funding stress. Seiche publishes forecasts before outcomes, stores them in a hash-chained record and reports both successful and missed alerts. It provides a public research record alongside LiquiLens's private lender workflows; customer data and books remain separate.

The product · private by design

LiquiLens

Treasury and early warning for Indian lenders and businesses. The reviewer-facing product is preproduction and available for controlled, self-hosted pilots. The construction-PIT crisis diagnostic sits in the demo's Evidence tab, institution by institution, misses included, and access to it is granted on request.

The public record · free forever

Seiche

connecting to the live board…
··
The reading you see here is fetched from the live terminal as you load this page. If it does not appear, the terminal itself is one click away.
The desk, live everything in this panel is fetched from api.seiche.info as you read

Why show you a second product we give away? Because an early warning vendor asking for trust should publish both forward-issued calls and historical evidence limits. Seiche is ours. Its live record is sealed and checkable; its reconstructed history is explicitly construction-PIT. Frozen rules, misses shown and status-bound claims are what LiquiLens brings to a lender's book.

The founder
M

Built by a lawyer who taught herself to build. LiquiLens is a working product, not a prototype, made solo and from research up, because infrastructure a regulated lender can trust has to be real before it is sold. The thesis came first. The engines came next.

Mrinal · founder, LiquiLens

Questions

Plain answers.

What is LiquiLens?

LiquiLens is an early warning system for bank and lender failure. It reads what institutions and regulators publish, funding structure, filings and market signals, and flags names drifting toward distress. Historical diagnostics cover 48 Indian institutions, 552 US failures since 2008 and seven European case files, with misses published and exact status tokens beside the figures. Ten engines, each grounded in published research, do the watching, and the same platform carries treasury, liquidity risk and credit scoring for lenders and businesses. The public diagnostics are not validated-backtest or real-money eligible.

Who is it for?

Risk and treasury teams reviewing India's regulated lenders or FDIC insured US banks, plus Indian businesses managing idle cash. The current public boards cover India's regulated lenders and every FDIC insured US bank; private-book workflows remain part of controlled pilots.

Is the Failure Radar a credit rating?

No. It is a disclosure based screen over public filings: a failure probability fitted on the real collapses in each market's record, distance to the regulator's own action thresholds, funding fragility read from the liability side, and a market implied signal for listed names. Every row names the factors that scored it, the tier rule is published with the data, and institutions the engines cannot score honestly are excluded with the reason stated. Ratings come from licensed agencies. This is the screen a risk head runs between their updates.

Is LiquiLens only for India?

LiquiLens began with Indian NBFC diagnostics, then extended into current run-risk screens for FDIC insured US banks and a current-amended construction-PIT diagnostic over 552 failures since 2008. Work across additional countries and institution types is market specific; the current European evidence consists of seven named case files with no cohort model. The portable layer is the evidence discipline, and all three historical records currently report both eligibility flags false.

Why does the money market matter for an ordinary business?

The money market is where banks and large corporates park cash overnight and borrow short term. It is the deepest, safest pool in any financial system, and in most markets it is effectively closed to everyone below a certain size. LiquiLens exists to open that access, so a mid-size company's cash can work through the night the way a bank's does.

What is the RBI LCR change and why is there a countdown on this page?

Under the RBI framework, LCR applies to every deposit-taking NBFC and, subject to RBI exclusions, to non-deposit-taking systemically important NBFCs with assets of 5,000 crore rupees or more. Excluded classes include CICs, Type I NBFCs, NOFHCs and SPDs. The calculator estimates when a growth path reaches the published asset threshold; it does not determine regulatory classification or continuing applicability.

How big is the gap LiquiLens is built for?

Large, and documented. India's MSME credit gap was estimated at about ₹25 lakh crore as of March 2025, with only 14% of MSMEs holding formal credit (Deloitte, 2025). MSMEs contribute roughly 30% of India's GDP and about 45% of exports (Ministry of MSME, 2024). On the cash side, RBI directions prohibit banks from paying interest on current accounts, so idle working capital earns zero by rule, not by accident.

Is LiquiLens live? How do I get access?

The public screening API and boards are live. The reviewer-facing product remains preproduction and is available for controlled pilots. The walkthrough at demo.liquilens.in sits behind an access request. Its Evidence tab shows three status-bound historical records on real public data; the treasury console uses synthetic operating data. A paid pilot evaluates private-book workflows with no production decisions during the proof.

Has any of this been tested on real data?

Yes, as reproducible historical research with explicit limits. The India diagnostic covers 48 institutions and currently reports 88.9% recall with a 21.5-month median lead; its status is PERIOD_END_PROXY_CONSTRUCTION_PIT. The US diagnostic covers 552 failures since 2008, reports 72.8% recall and AUC 0.854, and is CURRENT_AMENDED_CONSTRUCTION_PIT. Europe is seven named case files, not a cohort estimate. These are public-data diagnostics, not a validation on a live customer book, and all three currently report validated-backtest eligible: NO and real-money eligible: NO. A paid pilot pre-registers a separate evaluation inside the lender's environment.

What is Seiche, and why does a treasury company run a free terminal?

Seiche is our free, open source terminal for funding stress in the US money market. It runs on public data, publishes forward calls as issued, and serves a construction-PIT historical diagnostic with misses and explicit eligibility flags at seiche.info. We run it because an early warning vendor should publish its evidence boundaries, and because the research that watches the world's deepest money market informs the same lab. Seiche stays free forever.

What is the MFI Risk Board on this page?

It is the live public screening board for India's registered microfinance lenders, rendered on this page from the public API. The RBI's official register lists over 9,000 NBFCs, 95 of them NBFC MFIs. Every MFI with a vetted dossier of primary filings is read by the deterministic screening policy, and the trajectory next to each name is its score drawn quarter by quarter. Scores are disclosure-based screens, not credit ratings. The separate history is construction-PIT and ineligible for validated-backtest or real-money claims. Coverage is published, exclusions carry a stated reason, and a row is dropped when its filings go stale.

Does the weather really predict bank failure?

We did not assume so. We tested it. A pre registered study replayed the crisis record with two decades of sourced monsoon data overlaid, hypotheses, thresholds and the pass bar all frozen before scoring. The honest answer on today's record is that the data is too thin to prove monsoon adds early warning, so environment readings stay on the board as context and never touch a score. The verdict, the rainfall data and the exact condition that would reopen the question are public at api.liquilens.in/api/environment/gate. When a vendor tells you their score blends in alternative data, ask to see their version of that study.

What happens to our data? How is it protected?

Every lender's book is isolated to that lender. The regulatory tables, CRR, SLR, ALM and reports, are scoped per tenant and the isolation is enforced by automated tests on every commit, so a request for another tenant's data returns nothing at all. Access runs on short lived tokens with rotation and role based permissions, HTTPS is enforced everywhere, dependencies and secrets are scanned continuously, and a hash chained ledger records every alert as it was published so the history cannot be quietly rewritten. For pilots that cannot let data leave the building, the entire platform runs on the lender's own servers.

Prove it on your data

Run a six-week paid proof pilot

₹2.5 lakh fixed fee · credited in full toward production · self-hosted by default · no production decisions during the proof. Annual deployment starts at ₹12 lakh; new connectors and additional books are scoped separately. For investor conversations: [email protected].