The Series

R1: Jobs vs. market reaction PLAUSIBLE

−corr over the last 24 Employment Situation releases between the payroll surprise (m/m change minus its trailing 12-month mean, in trailing-σ units) and the S&P 500 return on release day. Surprise is proxied against the series' own trend (revised data, not real-time consensus), which is why R1 is PLAUSIBLE, not ROBUST.

Good news should lift equities. When strong payrolls sell stocks off, the market is trading the Fed, not the economy. Persistence for DECOUPLED: 2 consecutive readings. Surprise is proxied against the series' own trend (revised data, not real-time consensus); disclosed on Methodology.

R10: Energy prices vs. core inflation PLAUSIBLE

Rolling 12-month correlation of monthly changes in WTI crude (monthly average, lagged 2 months; oil leads core by ~1–2 months) and changes in core CPI YoY (CPILFESL). Core tracking energy is full pass-through: the shock spreading beyond the pump. The lag is a judgment call (1–4 months observed) and oil's correlation with core was higher pre-2000, so the 20-year framing matters; PLAUSIBLE. Added 2026-06-10 (Addendum A3).

An oil shock stays contained when core inflation holds steady while energy spikes. When core starts tracking energy, the shock has spread beyond the pump: into goods, services, and eventually wages. The pass-through detector. Persistence for DECOUPLED: 2 consecutive readings. Rolling 12-month correlation of monthly changes in WTI crude (monthly average, lagged 2 months) and changes in core CPI YoY. The oil-to-core lag is variable (1–4 months depending on the shock); 2 months is a central estimate, a judgment call, hence PLAUSIBLE. Oil's correlation with core was higher pre-2000 and lower post-2010 (shale, services-dominant economy), so the reference window matters; disclosed.

R11: PCE vs. CPI inflation ROBUST

A SPREAD, not a correlation: the signed gap CPI YoY minus headline PCE YoY. The normal methodological wedge is ~0.2–0.5pp (CPI above PCE). Pre-committed bands: ≤0.5pp BEHAVING · 0.5–0.7pp WATCH · 0.7–1.0pp STRAINED · >1.0pp EXTREME. A widening gap signals shelter (over-weighted in CPI) or healthcare (over-weighted in PCE) behaving unusually. Added 2026-06-10 (Addendum A3).

CPI and PCE normally track each other with a stable methodological wedge: CPI a few tenths above PCE. When they diverge, shelter (over-weighted in CPI) or healthcare (over-weighted in PCE) is behaving unusually, or the Fed's preferred gauge is parting ways with what consumers actually experience. Persistence for DECOUPLED: consecutive readings. The signed spread, CPI YoY minus headline PCE YoY, a level, not a correlation. The normal methodological wedge is ~0.2–0.5pp (CPI above PCE). Pre-committed bands: ≤0.5pp BEHAVING · 0.5–0.7pp WATCH · 0.7–1.0pp STRAINED · >1.0pp EXTREME. The bands gate the high (CPI-over-PCE) side per the addendum; an unusually low or negative spread is left BEHAVING and read in context, disclosed.

R12: Inflation vs. unemployment ROBUST

A SPREAD, not a correlation: the signed gap CPI YoY minus the unemployment rate (UNRATE) — which leg of the misery index dominates. Negative (unemployment above inflation) is the labor-led norm and reads BEHAVING; it holds ~83% of the 1948-present record, so the whole negative side is the calm band. Pre-committed bands gate only the inflation-led side: >0pp WATCH (inflation crosses above unemployment — the sign flip), >1pp STRAINED, >3pp EXTREME (≈ the historical 90th-95th percentile, the 1970s-80s and 2021-22 tier). The decisive threshold is zero, the crossover, and it is fixed, never tuned to a single episode. Added 2026-06-22.

The misery index adds inflation and unemployment; this reads which one is in charge. For most of the record unemployment sits above inflation — labor slack is the bigger problem, and that is the calm norm. When inflation crosses ABOVE the unemployment rate the misery index flips to inflation-led: the 1970s-80s regime, and 2021-22, the configuration that has historically pulled the Fed toward hikes. Persistence for DECOUPLED: consecutive readings. The signed gap, CPI YoY minus the unemployment rate (UNRATE), a level, not a correlation. Across 1948-present the gap is below zero ~83% of the time — unemployment normally exceeds inflation — so the BEHAVING band is the whole negative side, and the instrument reads green in the labor-led norm. Pre-committed bands gate only the inflation-led side: >0pp WATCH (the sign flip), >1pp STRAINED, >3pp EXTREME (≈ the 90th-95th percentile of the historical gap, the 1970s-80s and 2021-22 tier). The threshold that matters is zero — the crossover — and it is not adjusted to fit any single episode.

R2: Stocks vs. bonds ROBUST

60-trading-day Pearson correlation of daily S&P 500 log returns and the Treasury return proxy (−Δ 10-year CMT yield). Positive correlation = the hedge failing. Regime caveat: the negative stock-bond correlation is a ~25-year, low-inflation-era norm; before ~2000 the two were often positively correlated. FRED redistributes only ~10 years of S&P 500 data, so the reference window cannot see the inflationary regime: a positive reading may be a regime change, not a failing stabilizer (A5, 2026-06-07).

Treasuries should hedge equities (negative correlation). When both sell off together, the classic portfolio stabilizer is failing. Persistence for DECOUPLED: 10 consecutive readings. The arithmetic is mechanical: a plain rolling correlation of daily moves, on arctanh-transformed values. But the NORM is regime-dependent: Treasuries hedging equities (negative correlation) is a ~25-year, low-inflation-era phenomenon. Before roughly 2000, stocks and bonds were often POSITIVELY correlated, both driven by inflation. FRED redistributes only ~10 years of S&P 500 data (licensing), so this statistic's reference window sits entirely inside the negative-correlation regime and cannot see the alternative. A positive reading today may be a regime change, not a failing stabilizer; read the state with that caveat.

R3: Inflation vs. expectations ROBUST

Rolling 24-month correlation of monthly changes in the 5y5y breakeven with monthly changes in published CPI YoY (CPI for month m aligned to m+1, when it is published). Positive = expectations moving WITH realized inflation: de-anchoring measured directly as co-movement. Reconstructed 2026-06-06; see changelog.

Long-run expectations should stay near 2% no matter what realized CPI does. When they start moving WITH CPI, the anchor is dragging. Persistence for DECOUPLED: 3 consecutive readings. Rolling 24-month correlation of monthly changes in the 5y5y breakeven with changes in published CPI: co-movement measured directly. Reconstructed 2026-06-06; see the Methodology changelog.

R4: Growth vs. yields ROBUST

−corr over 26 weeks of the growth proxy (−Δ 4-week-average initial claims) and the weekly Δ 10-year yield. Positive = yields moving against growth.

Yields should track growth. When yields rise as growth weakens, the debt is being repriced on something other than the economy. Persistence for DECOUPLED: 4 consecutive readings. Initial claims serve as the weekly growth proxy; disclosed on Methodology.

R5: Gold vs. real yields ROBUST

60-trading-day correlation of daily gold log returns (official LBMA price) and changes in the 10y TIPS real yield (DFII10). Gold should fall as real yields rise; positive co-movement is the debasement/sovereign-risk side. Added 2026-06-07.

Gold should fall as real yields rise: holding it costs more when safe real returns improve. When gold climbs WITH real yields, the market is pricing debasement or sovereign risk, not opportunity cost. Persistence for DECOUPLED: 5 consecutive readings. Plain rolling correlation of daily moves, on arctanh-transformed values; gold is the official LBMA price, the real yield is the 10y TIPS (DFII10, 2003+). Positive co-movement is the debasement/sovereign-risk side, the tie to the Crisis-Ratio thesis.

R6: Dollar vs. auction demand PLAUSIBLE

Rolling 24-auction correlation of changes in the broad dollar and coupon bid-to-cover, aligned on auction dates. A falling dollar alongside weak auction demand is the loss-of-confidence signature; correlation cannot separate both-weak from both-strong, so the partner level tiles (F10, F12) corroborate. PLAUSIBLE. Added 2026-06-07.

The dollar and Treasury auction demand are both confidence gauges. When the dollar falls AND auctions are bid weakly, the world is stepping back from US assets at once, the loss-of-confidence signature. Persistence for DECOUPLED: 3 consecutive readings. Bid-to-cover is sparse (auction-dated) and the correlation cannot separate both-weak from both-strong, disclosed; the strain reading is corroborated by the partner level tiles (the dollar, F12; auction demand, F10), not taken alone.

R7: Home prices vs. mortgage rates PLAUSIBLE

Rolling 24-month correlation of monthly Case-Shiller home-price changes and changes in the 30y fixed mortgage rate. Prices rising while rates rise is affordability decoupling. Added 2026-06-07.

When mortgage rates rise, home prices should soften: financing is the binding constraint on what buyers can pay. Prices climbing WHILE rates climb is affordability decoupling from the cost of money. Persistence for DECOUPLED: 3 consecutive readings. Rolling 24-month correlation of monthly Case-Shiller price changes and changes in the 30y fixed rate; the lag from rate to price is real, so read the persistence, not a single month.

R8: High-yield spreads vs. equities ROBUST

60-trading-day correlation of daily changes in the HY OAS and S&P 500 log returns. Spreads should tighten as equities rise; spreads widening while equities rise is the complacency tell. Added 2026-06-07.

Credit spreads should tighten as equities rise: both should price improving conditions the same way. Spreads widening WHILE stocks rise is complacency: the equity market and the credit market disagree about risk. Persistence for DECOUPLED: 10 consecutive readings. Plain rolling correlation of daily moves, on arctanh-transformed values; HY OAS (F1) and the S&P 500. Positive co-movement of rising spreads and rising equity is the divergence tell.

R9: Inflation vs. wage growth PLAUSIBLE

A GATED RULE, not a bare z: STRAINED only when the rolling 12-month correlation of monthly changes in CPI YoY and nominal wage growth (S20) exceeds 0.6 AND both CPI YoY and nominal wage YoY exceed 4% (the pre-committed wage-price-spiral condition); otherwise BEHAVING. The gate suppresses the known false positive: in supply shocks inflation leads wages by 3–6 months, so the correlation runs high without a spiral. The purchasing-power-erosion side is read on the real-wage level tiles (S2, S21). PLAUSIBLE. Added 2026-06-10 (Addendum A3).

Prices and wages should not chase each other. When CPI and nominal wages co-move AND both run hot, the wage-price spiral the Fed fears is forming, and tightening becomes the only lever left. Persistence for DECOUPLED: consecutive readings. Pre-committed rule (Addendum A3): STRAINED only when the rolling 12-month correlation of monthly changes in CPI YoY and nominal wage growth (S20) exceeds 0.6 AND both CPI and wage growth exceed 4%: co-movement alone is not a spiral. The gate matters: in supply shocks inflation leads wages by 3–6 months, so the correlation runs temporarily high without a spiral (the false positive the gate suppresses). The other side of the rule (purchasing-power erosion, real wages persistently negative) is read directly on the real-wage level tiles (S2, S21). Causation is contested (cost-push vs demand-pull); PLAUSIBLE.

The Cost-to-Exist Ratio (Claimed Share), flagship

ClaimedShare = (Taxes + Obligations + Shelter + Food + Fuel + Health + RetirementGap) / Income

Decency floor, not comfort: shelter defaults to the HUD FMR national unweighted median. Conservative where it lowers the share: state/local tax and out-of-pocket health are omitted (both read LOW). Refundable credits are not. Omitting the EITC and Child Tax Credit would overstate tax and read HIGH, so they are modeled and netted against the tax line (B1, 2026-06-07; zero at these incomes without children). Median income: $83,730 (Census via FRED, 2024-01-01). Every default is adjustable in the basket editor on the board.

Pedigree: the credentialed neighbors

The components and the inverted form of this arithmetic are long established; the ratio presentation with live data and an adjustable assembly is this project's contribution. The lineage:

The Cumulative Real Wage Gap

RealWageIndex(t) = (AHE(t)/CPI(t)) / (AHE(base)/CPI(base)) × 100

The computation is mechanical; the base period is a disclosed assumption, which is why you can change it. Two wage series, shown separately and never spliced: all-employees AHE begins 2006-03; production-and-nonsupervisory AHE carries the history to 1964. They differ in level and composition.

The Sentiment Gap

OLS: sentiment ~ const + unemployment + cpi_yoy + real_wage_growth, monthly, full available joint sample; Newey-West (HAC, 12 lags) standard errors

Sample

1965-02-01 to 2026-04-01 (631 months)

R²

0.2769: the fundamentals never fully explained the mood; that is the finding, not a flaw to hide

Coefficients (std. err.)

const: 107.1739 (5.4807) · cpi_yoy: -2.4403 (0.5521) · real_wage_growth: -0.745 (0.9739) · unemployment: -2.3268 (0.8969)

Stationarity (ADF / KPSS)

cpi_yoy: non-stationary (both tests agree) (ADF p=0.0691, KPSS p=0.01) · real_wage_growth: inconclusive (the two tests disagree) (ADF p=0.0006, KPSS p=0.01) · sentiment: inconclusive (the two tests disagree) (ADF p=0.1449, KPSS p=0.0957) · unemployment: stationary (both tests agree) (ADF p=0.0213, KPSS p=0.0711)

Cointegration (Engle-Granger)

Residual ADF p = 0.5491: not clearly cointegrated at 5%; treat the gap as descriptive only.

The gap as a band

-37.9, 95% CI [-42.5, -33.4], 95% CI = gap ± 1.96 × the Newey-West (HAC, 12-lag) standard error of the fitted prediction. residual_std is the typical scatter of the gap around the model: the envelope a gap must clear to be unusual rather than ordinary model error.

Stated limitations

The regression specification is a choice; it is published in full so it can be challenged. The residual is not proof of any particular cause: the Cost-to-Exist story is ONE candidate explanation; documented partisan-composition shifts in the Michigan survey and omitted variables (rates, wealth, gas prices) are others. The residual is also highly persistent, so the gap's size carries less statistical certainty than a naive reading suggests; standard errors are Newey-West for this reason. Real wage growth uses production & nonsupervisory AHE (AHETPI, 1964+) so the sample reaches 1978; disclosed.

2026-06-22

Long View gains a sixth analog: real 10-year total return, stocks vs. Treasuries (1881-2023, vintage). Two immutable Shiller archives were ingested - SHILLER_STOCK_TR (real total-return price, dividends reinvested then deflated) and SHILLER_BOND_TR (real total return on a constant-maturity 10-year Treasury) - both the same 2023-09 vintage as the CAPE and long-rate archives, both scholarly provenance, era-banded and DRS-discounted like every reconstruction. The chart plots each asset's trailing 120-month annualized real return, (index_t / index_{t-120})^(12/120) - 1, on one frame so the crossing is visible: sustained negative real Treasury returns cluster only on the inflations of the 1910s-20s, the 1940s-50s peg, and the 1970s-80s - and the window now opening. It extends the existing CAPE precedent exactly: the scholarly splice stops at its 2023-09 vintage and the live leg is the board's job, not this page's. An analog is not an argument.

Why:

2026-06-21

C41, WTI crude oil, promoted from CONTEXT to a SCORED fast-tell. The board's only energy-strain signal was energy CPI (C35), which is monthly and lags the pump by ~1-2 months, so a supply shock did not register in near-real-time - the gap was visible when oil swung hard in mid-June 2026 while the board still read the prior month. Oil is a daily series with a meaningful absolute scale, so it now takes its STATE from fixed cutpoints (rising is the household strain): <90 CALM, 90-110 WATCH, >110 EXTREME, in $/bbl. A z-score would be poisoned by oil's own spikes (2008 to $145, the 2020 collapse to a negative print, 2022 to $120). The cutpoints are anchored on the modern 2010+ regime (mean ~$72, p75 ~$89, p95 ~$105) and the oil-to-gas pain points: $90 ~ the top quartile, gas pushing past ~$4.25; $110 ~ the 2008/2022 shock tier, gas toward $5. Only a SPIKE is strain - cheap oil is household relief, not a recession tell (the labor tiles carry that), so it is scored one-sided (up). At $84.65 (2026-06-15) it reads CALM, consistent with the week's pump relief; a Strait-of-Hormuz-style spike would trip it to WATCH/EXTREME the same day. The z and percentile remain for context but do not gate the state. Cutpoints are pre-committed and tunable only via a future dated entry.

Why:

2026-06-21

S7, the top-1% wealth share, moved to ABSOLUTE thresholds - the one concentration tile the 2026-06-12 false-negative sweep missed. On 2026-06-19 the tile relaxed from EXTREME to WATCH when the Fed Distributional Financial Accounts published Q4 2025 at 31.6% (down 0.2 from 31.8%), the second-highest reading in the entire 1989-2026 record. It read WATCH, not EXTREME, only because its decade reference window is poisoned by the series' own secular rise: the last ten years all sit at 30-31%, so the window mean is ~30.5 with a 0.54 SD, and a fresh near-record scores just z 1.9. A structural-strain tile that eases precisely as wealth concentration sets records is reading the trend, not the level. It now takes its STATE from fixed cutpoints anchored on the full record: 28.5 (~ the 1989-2026 median of 28.1, the pre-financialization norm) and 30.5 (~ the 80th percentile, the floor of the post-2010 record regime). So 31.6% reads EXTREME; the late-1990s/2000s broadening (27-28%) correctly reads CALM. The z and percentile remain on the tile for context but no longer gate the state, exactly as for the inflation and S2/F10 tiles. Cutpoints are pre-committed and tunable only via a future dated entry.

Why:

2026-06-12

False-negative sweep: six more tiles moved to absolute thresholds. A full audit of every level tile (after the inflation fix) found the same window-poisoning had left a cluster reading CALM at genuinely elevated levels - the board erred toward false negatives, never false positives. The clearest proof: all-worker real wage growth (S2) read CALM at -0.82% while its production-worker sibling (S21) read WATCH at -0.71%, the only difference being S2's decade window includes the 2022 real-wage collapse. Converted to absolute thresholds: S2 real wages (negative is strain; <-1.5 EXTREME, -1.5-0 WATCH, >0 CALM) -> WATCH; F10 Treasury auction demand (weak bid-to-cover is strain; <2.25 EXTREME, 2.25-2.45 WATCH) -> WATCH, consistent with the STRAINED R6 relationship that uses it; S1 Misery Index (8/11) -> WATCH; F2 VIX (20/30) -> WATCH; C24 credit-card delinquency (2.75/4.5) -> WATCH; C16 unemployment duration (15/22) stays CALM but is now un-poisoned. The band engine gained a low-is-strain direction for S2/F10. 2017 still reads calm and the Q4-2018 VIX gradient still reads EXTREME, so the symmetry guards hold; census moved 49/13/3 calm/watch/extreme to 38/21/6 across the two fixes.

Why:

2026-06-12

Inflation tiles moved to ABSOLUTE thresholds. The realized-inflation levels (C2 CPI YoY, C3 core PCE, C40 headline PCE, C39 CPI MoM, C35 energy, C36 shelter, C37 core goods, C38 core services) were scoring CALM at multi-year-high, accelerating inflation (CPI 4.27% in May 2026, the highest since 2023) because their z-score reference window is poisoned by the 2021-23 spike: a decade window with a 9% peak has a wide SD, so 4.27% reads as only z 0.49 even at the 79th percentile of its own decade. A board that cannot read red when inflation is genuinely high fails the same way one that can only read green does. These tiles now take their STATE from fixed cutpoints anchored on the 2% target (CPI/core/PCE/shelter/services 2.5 and 4; energy 8 and 20; CPI MoM 0.25 and 0.35), the same absolute-threshold approach the A4 household tiles use. The z and percentile are still shown for context but no longer gate the state. Core goods at 1.03% correctly stays CALM (genuinely below target), which a percentile or re-baselined-z rule would wrongly flag. The 2017 calm-window symmetry guard still passes (2% inflation reads CALM); 2022's 9% reads EXTREME. Thresholds are pre-committed and tunable only via a future dated entry.

Why:

2026-06-12

Addendum A4: the bottom-up household cluster. Eleven new tiles (HD1–HD4 captive demand, HM1–HM5 household margin, HB1–HB2 breaking points), scored on a separate engine from the macro board: PRE-COMMITTED ABSOLUTE BANDS (a fixed spread, gap, index, or rate) rather than z-scores, on a BEHAVING/WATCH/STRAINED/EXTREME scale with a 2-period persistence gate on upgrades, plus a combination rule (HM4) and a CONTESTED-as-status path (HB2). They live in their own board section, apart from the z-scored level census and noise floor. Two threshold calibrations are disclosed here rather than shipped silently: (1) HM1, the constructed margin index, is CONTESTED: its weights (captive gap ×2.0, rent overrun ×0.5) are judgment calls, with the formula and a sensitivity table on this page. (2) HB1's severe-renter-burden bands were RECALIBRATED from the A4 spec's 9–14% scale, which was a denominator error: the real ACS B25070 ratio (renters paying 50%+ of income on rent ÷ all renters) runs ~22–25%, so the spec's scale would have pegged the tile permanently EXTREME. The shipped bands (23.5 / 25 / 27) are anchored to the 2015–19 pre-pandemic range, keeping the pre-2020 baseline discipline. New series verified against live FRED/BLS/Census/USDA before wiring; corrections logged in docs/A4-series-verification.md.

Why:

2026-06-10

Addendum A3: the inflation decomposition. Eight new instruments make 'what is driving inflation, and is it spreading?' scorable, where the board previously carried only headline CPI (C2) and core PCE (C3). New level tiles, all mechanical reads of BLS/BEA/EIA data: energy CPI YoY (C35, scored against the full record because energy cycles run longer than a decade), shelter CPI YoY (C36), core goods YoY (C37, the tariff channel), core services YoY (C38, the sticky last mile), CPI month-over-month (C39, the base-effects input), headline PCE YoY (C40, the Fed's 2% target measure, beside core C3), and real wage growth for production & nonsupervisory workers (S21, AHETPI YoY − CPI YoY, the less mix-distorted cousin of S2). WTI crude (C41) ships as CONTEXT, not scored: the supply-shock INPUT behind energy CPI. Three new relationships: R9 inflation vs wage growth (a GATED rule: co-movement counts as strain only when both CPI and wages also run above 4%, which suppresses the supply-shock false positive), R10 energy vs core inflation (the pass-through detector, oil lagged 2 months), and R11 PCE vs CPI (a SPREAD with pre-committed bands, not a correlation). Codes follow the house scheme, not the addendum's spec codes (P/E/W collide with the board regex and The Wedge). Strain sides follow the symmetric engine: PCE is two-sided like CPI, the addendum's one-sided 'above target' framing notwithstanding.

Why: The May 2026 CPI field note exposed the gap: the board could read THAT inflation was elevated but not WHAT was driving it or whether it had spread from energy into core. The decomposition cluster makes the drivers legible; R10 and R11 read spread and pass-through directly. Causal attribution (tariffs, war) stays in Field Notes, never in a tile's normative sentence: the tiles score the signal, Field Notes interpret the source (Addendum A3 §9).

2026-06-07

Flagship: The Wedge (Cost-to-Exist, indexed since 2000). A headline Long-View chart, not a board tile: it carries no z-score state. Every line is a primary-official series rebased to 100 at a user-selectable base period (default 2000; options 2000/2008/2013/2019/2020) by Indexᵢ(t) = 100 × Seriesᵢ(t) / Seriesᵢ(base). The rebasing runs client-side on the baked raw values, so the base selector is live and auditable. Lines: median household income, NOMINAL (MEHOINUSA646N, the benchmark, drawn heaviest); all-items CPI (CPIAUCSL, the dashed reference); shelter as rent (CUUR0000SEHA) OR Case-Shiller home prices (CSUSHPINSA), toggled, never summed; medical care (CPIMEDSL); tuition, fees & childcare (CUSR0000SEEB, begins 1978); food (CPIUFDSL); energy (CPIENGSL); new vehicles (CUSR0000SETA01); and an optional average-hourly-earnings wage line (AHETPI). Lines are NOMINAL (we do not deflate : deflating hides the wedge); the canonical cadence is ANNUAL (monthly CPI components resampled to calendar-year averages and joined on the year; income is annual and never interpolated, with an optional monthly view drawing income as un-interpolated step markers). CPI components are seasonally-adjusted except NSA rent. Disclosed caveats travel WITH their lines and cannot be dismissed: the hedonic quality-adjustment on new vehicles (which under-states the wedge) and the per-unit vs total-burden gap on medical care; average-transaction-price (vehicles) and NHE-per-capita (health) are declared, not shipped. The 2013 CPS income-redesign break is marked on the income line and never silently bridged. Confidence: chart PLAUSIBLE (a disclosed base-period choice), each line ROBUST (mechanical rebasing of an official series), the weighted composite CONTESTED (the basket weighting is a judgment). The per-category multiples are the unimpeachable lead; the composite (Σ wᵢ·Indexᵢ seeded from the live Cost-to-Exist basket and reweightable) is the companion, never a replacement.

Why: The present-tense Cost-to-Exist ratio reads what life costs today; nothing on the board read the TRAJECTORY. The Wedge is the sourced, honest version of the viral '$100k-in-1995' argument: since 2000 all-items CPI roughly doubled (×1.8) while housing, tuition and healthcare tripled-plus and median income did neither (×2.0), and every claim traces to an official FRED series, so the wedge cannot be waved away as cherry-picking. Reproducible from the public repo + FRED pulls.

2026-06-07

v0.3 Anchors & Deltas (E3 + E5): usability, no new claims. E3: every active level tile now carries a collapsed 'In context' table showing what it read at four reference moments: Lehman (2008-09-15), the Q4-2018 selloff (2018-12-24), the Aug-2019 curve inversion (2019-08-28), and the COVID crash (2020-03-23), using only data available on or before each date (step-function value, no interpolation) and the SAME state engine the live board uses (states.level_state), baked into data/derived/anchors.json by pipeline/anchors.py. Where a series had not begun (FRED carries only ~3y of HY OAS, ~10y of S&P), the anchor reads 'no data', never a blank that implies calm. E5a: each level tile shows its move vs the prior observation (value − previous from history.json), baked in board.py with cadence-driven phrasing ('since last week/month/…') and direction by shape AND colour (▲/▼); trivial moves (under 0.1% of the level) are suppressed to keep the board uncluttered. E5b: a 'what changed since…' diff over the board-state history: pick a date, or compare against your last visit (localStorage), listing transitions in the Wire's factual register. CAVEAT, stated on the control: board_history begins at launch (2026-06-06) and accrues daily; there is NO pre-launch history, and the picker is capped to the recorded span. The pipeline emits the array-form data/derived/board_history.json as the single source for the island.

Why: A z-score is a wall to a non-analyst; anchoring it to Lehman and COVID gives the number a felt scale, and an honest one: a tile that reads calmer today than in 2008 is a feature, not a hidden green. The deltas and the diff answer 'what changed since I last looked' without anyone editorializing. Anchor states are a pure function of the pipeline (never recomputed in the site), and every number traces to a primary source already in the repo.

2026-06-07

v0.2 Labor-composition cluster (Addendum A1): seven tiles the headline unemployment rate hides, in Standing Context beside the existing labor tells. C17 white-collar payroll share (USPBS/USPRIV, down, decade-relative), C18 temp-help share (TEMPHELPS/USPRIV, down, decade-relative), C19 JOLTS hires rate (down) and C20 layoffs rate (up), completing the churn triplet with quits (C10), C21 vacancy-to-unemployment (JTSJOL/UNEMPLOY, down, decade-relative), C22 long-term-unemployed share (UEMP27OV/UNEMPLOY, up), C23 new-entrant share (LNS13023569/UNEMPLOY, up). All ROBUST on primary-official BLS series (integrity-events factor applies); the narrower Professional/Scientific/Technical cut (CES6054000001) is carried as a disclosed secondary lens behind C17's supersector. ADP sector detail is deliberately NOT used (it would drop C17 to PLAUSIBLE and has been a poor predictor of the BLS series); ADP belongs in a Field Note.

Why: The board read aggregate labor well but nothing read its composition: the white-collar contraction and the low-hire/low-fire freeze. Every value traces to FRED and is reproducible from the repo.

2026-06-07

v0.2 Coverage: new relationships, level tiles, and counterweights. Relationships: R5 gold vs. real yields (LBMA official price × DFII10), R6 dollar vs. auction demand (PLAUSIBLE), R7 home prices vs. mortgage rates, R8 high-yield spreads vs. equities, each Fisher-transformed with effective-N reporting like R1–R4. Fast tells: continued claims, Treasury realized-vol (a keyless MOVE proxy), the 10y–3m curve, bank deposits. Context: unemployment duration, credit-card and CRE delinquency, banks tightening (SLOOS), the Black–white jobless ratio, real policy rate, M2 and bank-credit growth, federal outlays, oil & gas extraction. Counterweights: household net worth to disposable income (B3), refundable EITC/CTC in the basket (B1), childcare on by default with children (B2). Source corrections, all disclosed: the LBMA gold fixing and EIA crude series MCRFPUS2 were discontinued on FRED, so gold reads from the LBMA's own published price and crude from the live IP index IPG211S; COMREPUSQ159N was dropped for a ~14-month lag (CRE risk stays on the live delinquency tile). Eleven indicators with no in-policy automated feed are declared, not faked (MOVE, regional-bank strength, funding stress, near-term forward spread, r*, auto delinquency, renter cost-burden, uninsured, food insecurity, CEO-pay ratio, aggregate markups).

Why: The asset/distribution counterweights make the board even-handed; the new spreads and tiles extend coverage without diluting rigor: every shipped value is primary-sourced, every gap is declared.

2026-06-07

v0.2 Calibration: six corrections, no thresholds bent. (A1) Correlations are now Fisher-transformed before z-scoring (arctanh), because a raw correlation's sampling distribution is not normal; the displayed r is unchanged, the state is computed on the transform. (A2) Overlapping rolling windows make consecutive readings near-duplicates and bias the spread low; the spread is now corrected for autocorrelation and each relationship reports its EFFECTIVE N (independent windows) instead of the raw count. (A3) The status line now reports observed vs. chance-expected WATCH/EXTREME counts, gated per tile's own direction, and names the genuine outliers, excluding chronic trend features. (A4) The Sentiment Gap now ships ADF/KPSS stationarity, an Engle-Granger cointegration test on its residual, and a HAC confidence band rather than a bare point. (A5) R2 (stocks vs. bonds) carries a regime disclosure: its negative norm is a ~25-year, low-inflation-era phenomenon the licensed ~10-year window cannot see past. (B1) The Cost-to-Exist tax line now models the EITC and the refundable Child Tax Credit; the 'conservative, reads low' claim is scoped to the lines where it is actually true (omitting refundable credits read HIGH, not low).

Why: All six were raised in external review; all touch numbers already published. Recomputing R1–R4 moved no relationship's STATE (the before/after z-table is in the git log); the corrections refine the arithmetic and the disclosure, they do not manufacture drama. Stated here under the alpha daily-summary policy; the granular history is in the public git log.

2026-06-06

Launch Day.

Why: Built, externally reviewed, and calibrated in the open, all in one day. Every step is in the public git log.

The institutional-integrity factor: the full public table

Keys only on documented, specific, sourced events, never on partisanship, never on whether anyone dislikes the numbers. Small, capped (−8 max per agency), symmetric across administrations, and it reverses when conditions normalize. It measures elevated uncertainty and reduced precision, not manipulation. Challenge any entry by merge request with a primary source.

Counterweight, applied honestly: Applied honestly: there is no evidence of manipulation. BLS has continued to publish unfavorable readings (job losses in some months, an inflation jump tied to the Iran war), which indicates the core function persists, and funding strain predates this administration and spans both parties.