How this monitor works — and how it can be wrong
8 indicators measured against thresholds registered in advance, one strength index with published anchors, a separate trigger verdict, and a public log of every amendment. Nothing here is discretionary at publication time: the rules below were registered before the data came in.
1 · Registered thresholds
Each indicator has one threshold, registered before observation. Crossing it sets the indicator to firing. Thresholds change only by the amendment procedure in §5 — never in response to the current reading.
| # | INDICATOR | THRESHOLD | CLASS | SOURCE |
|---|---|---|---|---|
| 1 | Oil & gas budget revenues — “Oil and gas money coming in” | < 50% | 2 | minfin.gov.ru |
| 2 | Liquid NWF — “Rainy-day fund left” | < $20 bn | 2 | minfin.gov.ru |
| 3 | Budget deficit, % GDP (annualized) — “How fast they burn cash” | > 10% | 3 | minfin.gov.ru |
| 4 | Consolidated regional budget balance — “Are the regions staying afloat?” | < −₽2.5 trn | 2 | roskazna.gov.ru |
| 5 | Overdue receivables (unpaid invoices) — “Are firms paying each other?” | > 2× | 2 | rosstat.gov.ru |
| 6 | Annual inflation — “Prices in shops” | > 40% | 3 | rosstat.gov.ru |
| 7 | M2 money supply growth — “How fast they print money” | > 30% | 3 | cbr.ru |
| 8 | RZD rail loading — “Stuff moving by rail” | < 80% | 2 | rollingstockworld.ru |
Physical confirmation now rests on a single indicator, rail loading: electricity generation was retired on 2026-08-21 (see §5). One physical check is weaker than two — said plainly rather than hidden.
A threshold is a judgment, so each one carries its anchor. Where the anchor is literature or arithmetic, we cite it; where it is a registered judgment, we say so plainly. None of them changes in response to a current reading.
- Oil & gas below 50% of its pre-war real level — the same calibration the retired rate line used, restated on the level: 50% sits below the deepest point the war has produced (54% in April 2026), so only something worse than anything seen so far fires it. Rolling 12 months, CPI-deflated, against 2021.
- Liquid NWF < $20 bn — arithmetic: at the registered deficit range (₽6.5–7.5 trn/yr), $20 bn ≈ ₽1.6 trn covers roughly 2.5–2.9 months of it — the point where the buffer stops being a buffer.
- Deficit > 10% GDP (rolling 12-mo) — set well above the entire realized wartime range on our own series (~2–5%): crossing it means the financing need has left the corridor Russia has managed so far. Class 3 regardless — it can only confirm.
- Inflation > 40% y/y — the classic inflation-crisis threshold in the growth literature (Bruno–Easterly, 1998).
- M2 money growth > +30% y/y — the printing-press detector: war-years growth ran +17 to +25% on the CBR's own series (peak mid-2023); +30% marks a regime shift toward monetized deficits that healthy-looking bond auctions cannot hide.
- Rail loading below 80% of its pre-war tonnage — 80% sits below the war's own floor (87%), and roughly halfway to the 2009-crisis collapse. The weakest-anchored line here — registered judgment, said plainly. Rolling 12 months against 2021.
- Overdue receivables > 2× pre-war, in real terms — arithmetic on the index scale: the pile of invoices Russian firms fail to pay each other, deflated by CPI and measured against January 2022. A doubling in real terms is the line; it currently stands at 1.96×. Because the pile is measured in 2021 prices, inflation alone can never carry it over.
- Regional budgets: rolling-12m balance below −₽2.5 trn — registered judgment anchored on the pre-war level: the consolidated regional balance was a +₽0.66 trn surplus in 2021 and is −₽1.6 trn now. −₽2.5 trn is roughly 1.1% of GDP — the scale at which regional gaps stop being absorbable locally and land on the federal budget.
2 · Indicator classes & voting rights
Series formed outside the government's pen — a foreign customs office or a court registry, not a Russian ministry. This class is currently EMPTY: the bankruptcy registry was our only entry and it was retired on 2026-08-22 (see §5). Until we register a replacement, no single indicator can trigger a verdict on its own, and the bar is higher than it was. Said plainly rather than quietly dropped. One firing class-1 indicator can trigger alone.
Oil & gas revenues, the reserve fund, regional budgets, overdue receivables, rail loading. Published by a state with an incentive to manage them, but the quantity is physical or the failure is self-evident. Triggers only in a pair — two class-2 indicators firing simultaneously.
Inflation, money growth, deficit-to-GDP. They reflect expectations and policy choices and can be painted — Russian CPI baskets and administered prices are managed, and the sign can invert exactly under stress. Never trigger.
3 · What the index does not count — the other pockets
The eight indicators measure what Russia reports every month: the money coming in, the buffer it has left, and the strain showing up in prices and unpaid bills. Russia has other coping resources, and pretending otherwise would be the fastest way to lose your trust. Each is excluded deliberately — it is either not a regularly published series, or a change of regime rather than a buffer:
- Central-bank gold and yuan — the CBR sold ~44 t of gold in H1 2026 and its yuan composition is classified. These are the central bank's reserves, not the budget's; spending them is itself an escalation we track through the news, not a quiet buffer.
- Coerced bank purchases / CBR repo — makes auctions look healthy while monetizing the deficit (the 2022 scheme). Not a buffer: it is the printing press wearing a market costume — which is exactly what the M2 indicator watches.
- Tax hikes (VAT 22% from 2026, ~₽1.7 trn/yr) and sequestration — they shrink the deficit itself, so the deficit indicator absorbs them automatically once they show up in the data.
- Devaluation — mechanically flatters the reserve reading (the fund is counted in dollars, the deficit it must cover in roubles), and the model shows that honestly when it happens.
- Treasury balances, delayed payments, asset sales — real but opaque and one-off; weeks-to-low-months of coping, not a published series we could track without guessing.
Full stress-test of the model against the strongest resilience arguments — sources and scenario runs — in the project's model-critique document.
4 · Data rules
- Primary sources, and where a value had to be read by hand it carries a ◆ mark. Every value stores the URL, the verbatim fragment it was parsed from, the date of the data, and the fetch timestamp. The raw page of every fetch is archived.
- A value that can't be parsed is a logged failure, never a silent carry-forward and never an interpolation.
- A source that stops publishing sets its indicator to ○ UNKNOWN with a ◆ DATA DARK flag. Unknown is never treated as zero, and never as fine — 900+ official Russian datasets have been hidden since 2022, and going dark is itself a signal.
- An official announcement that a data-generating activity is suspended (e.g. "auctions will not be held") is recorded as an observation of the suspension itself, and puts the indicator into firing regardless of its last ratio.
- "Auction failed" and "auction not held" are different events and are logged as such.
5 · Threshold changelog
Thresholds may be amended only with a written rationale and a dated entry here — never retroactively, and never to make the current reading look better.
| DATE | INDICATOR | WAS → NOW | RATIONALE |
|---|---|---|---|
| 2026-08-14 | All | — → v0.1.0 | Initial public registration of the threshold set. |
| 2026-08-19 | Oil & gas budget revenues | index input: y/y rate → index input: real level vs 2021 | The year-over-year rate answered “versus last year”, so a rebound off a bad year read as health. The index input became a level measured against the pre-war year. The trigger threshold was NOT changed. |
| 2026-08-21 | Power generation | < −5% y/y, class 2 → retired | Physical confirmation duplicated by rail loading, no fetchable monthly archive, and never temperature-adjusted. Physical activity now rests on a single indicator — weaker, and said so on this page. |
| 2026-08-21 | Overdue receivables (unpaid invoices) | — → > ₽12 trn, class 2 | Bankruptcies had sat at the healthy end for every month they existed, so that reading carried no information in the index. Overdue receivables measure the same distress in money, from monthly Rosstat reporting. |
| 2026-08-21 | Consolidated regional budget balance | — → < −₽2.5 trn per 12m, class 2 | Regional deficits end up on the federal doorstep, and the consolidated balance swung from a +₽0.66 trn surplus in 2021 to −₽1.6 trn. Nothing we tracked measured that. |
| 2026-08-21 | RZD rail loading | index input: y/y rate → index input: 12m tonnage vs 2021 | The year-over-year rate answered “versus last year”, so a rebound off a bad year read as health. The index input became a level measured against the pre-war year. The trigger threshold was NOT changed. |
| 2026-08-22 | OFZ placement | < 20% of the quarterly plan, class 2 → retired | Bond placement measures how the deficit is financed, not whether Russia can pay for the war. When auctions fail, the deficit lands on the liquid reserve fund — which we already measure directly, so the two indicators moved as one. The ratio also depends on a borrowing plan the finance ministry writes for itself, and it rewrote that plan mid-year. The trigger verdict moves from “one class-2 signal” to “no trigger”; nothing else changes. |
| 2026-08-22 | Bankruptcies | > 1.3× the pre-war rate, class 1 → retired | Court bankruptcy counts are gated by moratoria and filing costs, and in every month we ever recorded they sat at the healthy end of the scale, so the indicator carried no information. Overdue receivables, registered a day earlier, measure the same unpaid-bills stress in money and monthly. Keeping both would have counted one stress twice. |
| 2026-08-22 | Liquid NWF | data label: month reported → data label: month covered | The first reserve-fund reading, hand-entered at registration, was labelled with the month it was REPORTED (2026-08) while every fetched reading since is labelled with the month it COVERS. The later label made a stale figure look like the freshest one: the index scored $46.2 bn while every page displayed the fetched $46.4 bn, and the chart ended below its own headline. The value and its source fragment are unchanged; only the month label was corrected, to “unknown”. |
| 2026-08-22 | All | index level: plain mean → index level: chain-linked | The index moved when a component appeared or disappeared, not only when Russia did. Money-supply data returning in June 2026 lifted the plain mean 6.1 points while every component present in both months had moved 0.4. The level is now chain-linked — each step measured only over the components present in both months — so it answers “better or worse than last month” honestly. Nine of fifty-six months carried such a step; the largest distortion was 5.7 points. No threshold and no anchor changed. |
| 2026-08-22 | Annual inflation | pre-war anchor: CBR's 4% target → pre-war anchor: 4.6% measured (2017-2021) | Every other indicator is anchored to what happened before the war; inflation was anchored to what the central bank aimed at. Rosstat's own December-to-December series gives the fact: 2.5, 4.3, 3.0, 4.9 and 8.4 per cent over 2017-2021, averaging 4.6. A rate has a pre-war level even though the price level never returns to one. Today's 6.0% scores 96 instead of 94; the index moves 0.25 of a point, which is why this could be corrected without it looking like tuning. |
| 2026-08-22 | All | scale capped at 0 and 100 → scale open past both anchors | Scores were capped at 0 and 100. The cap only ever bit at the top: in 2022 oil and gas revenue scored up to 141 (a real level of 120.5% of pre-war) and the reserve fund up to 135 (128.9% of its pre-war level), and both were recorded as 100. That flattened the start of the war and made the fall since look gentler than it was. Scores now extrapolate past both anchors. Zone names still saturate, because they are interpretation bands rather than measurements. |
| 2026-08-22 | Budget deficit, % GDP (annualized) | pre-war anchor: balanced budget → pre-war anchor: the 2021 surplus, 0.34% of GDP | The deficit was the last indicator anchored to a judgment rather than to what happened: its 100 meant a balanced budget, while 2021 actually closed with a surplus of 524.3 bn RUB, or 0.34% of GDP, in the ministry's own monthly execution file. Anchoring to the fact scores today's 3.3% deficit at 65 instead of 67 and moves the index a quarter of a point. The trigger line, 10% of GDP, is unchanged. |
| 2026-08-22 | All | trigger line: a year-over-year rate → trigger line: the level against pre-war | Zero on the index is defined as the registered trouble line, so “past the line” and “score at zero” must be the same event. For three indicators they were not: after the index moved to levels on 2026-08-19, the trigger kept testing the old number. Oil and gas was tested on a year-over-year rate against a −25% line while the index scored its real level against 2021; rail loading the same; overdue receivables was tested on a nominal rouble line while the index scored a real multiple. That is why the oil siren fell silent in May 2026, when the April figures took the year-over-year rate back above its −25% line, and stayed silent through July's +18.6% while the level sat at 57%, near its record low — a rate compares to last year, and last year was already collapsed. Each indicator is now one number with one line, and its line is the zero of its own scale. The lines: oil and gas below 50% of pre-war real revenue, rail loading below 80% of pre-war tonnage, overdue receivables above 2× pre-war in real terms — all three already registered as the index's zero, all three below the worst the war has produced. Replayed over the war, this makes the model quieter, not louder: the 18 months in which the oil rate line fired alone and never escalated now read as quiet. Nothing fires today and nothing fired differently in any month where a verdict was reached. |
The strength index — the formula
Each indicator is stretched linearly between two anchors registered in advance: its pre-war level (worth 100, recorded per indicator in the data registry with a note on where the number comes from) and its trouble line (worth 0, registered 2026-08-14). The index is the plain mean of the components available that month — no weights exist, and none can be tuned. “We do not know” drops out of the mean and coverage is shown next to the reading. Because a component appearing or disappearing would otherwise move the level by itself, the level is chain-linked: each month's step is measured only over the components present in BOTH that month and the one before it, and the steps are added up from January 2022. When money-supply data reappeared in June 2026 the plain mean jumped 6.1 points, while every component present in both months had moved 0.4 between them — the published line moves by the 0.4. The plain mean and the coverage are still shown, so you can see when the two differ. The scale is not capped at either end: a reading better than its pre-war anchor scores above 100 and one past its trouble line scores below 0, because both are real states. Capping only ever bit at the top here — in 2022 oil revenue scored up to 141 (a real level of 120.5% of pre-war) and the reserve fund up to 135 (128.9% of its pre-war level) — so it made the fall from 2022 look gentler than it was. Zone names still saturate: above 100 reads “peacetime normal”, below 0 reads “emergency financing”. Zone names are fixed interpretation bands: 80–100 peacetime normal, 60–80 elevated pressure, 40–60 stress, 20–40 crisis, 0–20 emergency financing.
The registry and the index are the same eight indicators: every number on an indicator page is a number the index scored. Two indicators were retired on 2026-08-22 and the reasons are in §5 — OFZ placement, which measured how the deficit is financed rather than whether Russia can pay for the war, and duplicated the reserve fund we already measure directly; and bankruptcies, replaced by overdue receivables, which measure the same stress in money and monthly. The index is a thermometer, the trigger is a siren, and neither overrides the other.
Levels, not rates. A year-over-year rate answers “versus last year”, and after a bad year that reads as health — a +38% bounce off the 2025 slump would have scored oil revenue at 100 while real revenue sat at 57% of pre-war. So the level-based components measure the pre-war level directly: oil & gas revenue as rolling-12m real revenue (CPI-deflated) against the 2021 total; overdue receivables as a real multiple of January 2022; rail loading as rolling-12m tonnage against 2021; regional budgets against the 2021 surplus. Inflation and money growth stay rates on purpose — they are flows, and the price level never returns to pre-war. Liquid NWF is a level in nominal dollars: the 2022 baseline and today's reading are not the same dollars (~18% of purchasing power apart), which is a known open flaw, not a hidden one.
The projection is not a trend line. We backtested the obvious rule — extend the last three months out by six — against our own history: it covered the actual outcome 43% of the time, and its midpoint was no better than assuming nothing changes (7.4 vs 7.6 points of mean error). Momentum in this index does not survive six months. What we publish instead is calibrated on the index's own record: the centre is today's level nudged by a damped slope (0.35 of it), and the band is the observed 10th–90th percentile of every six-month change this index has ever made. Walk-forward, that band contains the outcome 85% of the time and its centre errs by 5.5 points. It is a band, never a date, and it is recomputed from the series on every run.
Time to zero is a separate calculation from the six-month band, and a weaker one. We resample the index's own monthly steps in blocks of six consecutive real months — blocks, so whatever autocorrelation the series has survives the resampling — and walk each path forward until it crosses zero. The seed is fixed, so any figure we publish can be reproduced exactly. Checked walk-forward against our own history, the simulated 10-90 band covered the realised level 96% of the time at six months and 100% at twelve, eighteen and twenty-four, against a target of 80%: the band is too WIDE, so the spread errs toward “we don't know” rather than toward false precision. What it cannot check is the crossing itself, which has never happened — the index has lived between 46 and 100. Resample only the last two or three years instead of the whole war and the centre moves by about two years; that is the honest size of the uncertainty, and it is larger than the width of any single simulation.