Market Pulse Index (MPI) — Changelog

Why we publish a changelog

An index that quietly changes its inputs is an index you cannot trust. Institutional research desks publish methodology revision histories for the same reason ratings agencies publish criteria updates: when the formula changes, the score is no longer directly comparable to its earlier self. A reader looking back at an MPI of 64 in March 2026 needs to know whether that 64 was produced by the same set of sub-indicators, the same rolling baseline, and the same confidence-band rule that today’s 79 was produced by.

This page is the public record of every methodology change to the Market Pulse Index. It is append-only. Earlier entries are never edited or removed — corrections are added as new dated entries on top of the prior text. If a sub-indicator is added, removed, reweighted, or has its data source changed, it shows up here. Daily score updates do not. Daily scores live in the Pulse posts.

How to read this

Entries are dated descending — newest at the top. Each entry is scoped to a methodology change: a new sub-indicator, a revised calculation, a baseline adjustment, or a documentation upgrade that affects how the score is interpreted. Entries are not generated for routine daily refreshes.

If an entry references a version (v1, v2.0), that version is the methodology spec in force from that date until the next versioned entry. Scores published under a prior version are not directly comparable to scores under the current version without re-baselining.

The log

2026-07-07

Public documentation reconciled line-by-line against the live payload (schema 2.0 — no scoring changes). Inputs now formally logged as retired from the composite and the public accounting: the weekly retail sentiment survey (its free feed went dark in June 2026; CNN Fear & Greed is now the primary sentiment input and is disclosed as such), the consolidated exchange put/call composite (endpoint began rejecting programmatic access, June 2026), SKEW, the VIX9D/VIX6M supplementary tenors, WTI crude spot, and TIPS breakevens. The US-dollar input is measured via the UUP dollar-index ETF as a disclosed proxy. The Data Sources and MPI Methodology pages were updated in lockstep, and a sentiment-concentration limitation was added to the methodology page. Restoring an independent survey-based sentiment input remains an open roadmap item.

2026-05-14 (logged 2026-07-07)

Post-close recompute moved from 16:30 ET to 18:00 ET with a 22:00 ET safety-net pass, after a staleness incident in which exchange end-of-day files had not yet rolled at compute time and the pipeline scored the prior session’s bar under the current label. The cron windows in the pipeline workflow are the operational source of truth. This entry is recorded late; the change itself took effect 2026-05-14.

2026-05-08 (PM)

Cron schedule corrected: post-close refresh moved from 4:15 PM ET to 4:30 PM ET to land after CBOE EOD CSV publish window. Holiday gating made explicit via [gate_holiday] log line. Window extended to 16:50 ET to absorb GitHub Actions cron jitter.

2026-05-08

Schema v2.0 published. Documented sub-indicators with category-level data sources. Confidence-interval calculation formalized as a dispersion-based band.

2026-05-07

Auto-update pipeline architected — daily refresh at 9:15 AM ET pre-market and 4:30 PM ET post-close (Mon–Fri, market days only — NYSE holidays excluded). Stale-threshold flag introduced.

2026-04-21 → 2026-05-05

MPI backfilled across 11 trading days following methodology refinement. Score range observed: 52–79.

2026-04-15

Confidence interval methodology reviewed. Switched from absolute fixed band to relative dispersion-based band.

2026-04-13

Multi-factor composite scored against a rolling historical baseline introduced. Sub-indicators span trend, breadth, volatility, yield curve, credit spreads, sentiment, sector rotation, currency/commodity, and liquidity.

2026-03-20

Methodology v1 published. Earlier composite covered trend, vol, sentiment, breadth, and regime.

2026-01-10

Initial MPI prototype on internal dashboard. Not publicly displayed.

Future scheduled changes

• Q3 2026 — regime-conditional adaptive weighting (evaluation). Currently sub-indicators are equally weighted within their categories. We are evaluating whether weights should adapt to the prevailing regime (e.g., breadth and credit spreads carry more signal in transition regimes than in stable Bull regimes).
• Q3 2026 — independent sentiment input restoration. Evaluating replacement survey-based or market-derived sentiment feeds to reduce single-source concentration in the sentiment sub-indicator.
• Q4 2026 — feature drift monitoring. Each sub-indicator’s distribution will be tracked against its rolling baseline.
• Q4 2026 — walk-forward validation publication. Once methodology is stable for 6+ months, walk-forward results will be published on the Performance Archive.