What a company is doing, read from its own filings.

What is shown today

An axis is shown as a score only after it passes a back-test whose rules were written down and committed before any result was computed. An axis that has not passed shows its components only, with the reason.

Latest run 2026-10-22: 503 companies read; readings with enough evidence to score: expanding 378 · confident 499 · pressure 503. Back-test last run 2026-10-21.

The three axes

Each axis runs from 0 to 100. Each component is weighted as below; the weights were fixed in the pre-registration.

Item 2.01 filings cover completed acquisitions and dispositions alike; the filing does not say which in structured form, so both count. Fails-to-deliver are settlement failures, not short interest. Product-safety and enforcement events, consumer complaints, private-credit marks on a company’s loans and trademark filings are not in this version; each enters only with its own pre-registered test. News tone is not used: Palanor has no per-company news source it is licensed to use this way.

How a score is made

Each component is placed twice: against the company’s own readings over the previous 36 month-ends, and against the other S&P 500 companies in its GICS sector on the same date. The two percentiles are blended 50% / 50%. A company with fewer than 12 months of its own history is placed against its sector only; a sector with fewer than 8 readable companies falls back to all members. The axis is the weighted mean of its readable components.

Coverage gate. If less than 60% of an axis’s weight is readable on a date, the axis is withheld for that company and date, with the reason. It never falls back to a middle value.

Stages. At or below 40 and at or above 60 the axis reads Retrenching / Expanding, Hedging / Confident, Low pressure / High pressure; in between, Steady, Mixed, Moderate pressure. A stage changes only when the score moves 5 points past the boundary, so readings near a line do not flip back and forth; a stage held that way is marked as held.

Drivers. Each reading names the three components that moved it most, each linked to its evidence: the filing, the release, the guidance statement or the source page.

Point in time

A reading dated on a day uses only what was public that day. Cash-flow figures come from the 10-Q or 10-K that first reported them, and a later restatement never changes an earlier reading. 8-K items count from their SEC acceptance time, guidance from its 8-K, calls from their date, insider trades from their Form 4. Hiring filings count 120 days after the decision quarter, because the Labor Department publishes them quarterly; fails-to-deliver 30 days after settlement, because the SEC publishes them twice a month. Patent applications count from publication, since they are secret until then.

Readings run at month-ends from January 2022 and nightly for the current day. Stored month-ends are never rewritten. A component counts only when its whole window lies inside the stored history of its source, so early months read fewer components and some axes are withheld there.

The back-test

The question: does each axis tell you something about what happens to a company over the next 182 days that a naive reader would not already guess? The events, all from primary sources (no credit ratings, no analyst estimates):

• guidance lowered or withdrawn
• revenue below the company’s own guided range
• exit or restructuring filings (8-K 2.05)
• impairment filings (8-K 2.06)
• impairment, debt acceleration or bankruptcy filings (8-K 2.06, 2.04, 1.03)

The naive reader knows whether the company had the same event in the previous 182 days and how often it happened in the company’s sector. Each axis is added to that reader in a logistic model fitted only on dates whose outcomes were already known (walk-forward), and the test dates are split into two halves. An outcome passes when the axis improves the forecast in both halves, its direction holds in most fits, and each half has at least 20 events. An axis passes when its main outcome passes and it does not point the wrong way on the others.

• Expanding ↔ Retrenching: main outcome exit or restructuring filings (8-K 2.05); also checked against impairment filings (8-K 2.06) and guidance lowered or withdrawn.
• Confident ↔ Hedging: main outcome guidance lowered or withdrawn; also checked against revenue below the company’s own guided range.
• Pressure: main outcome impairment, debt acceleration or bankruptcy filings (8-K 2.06, 2.04, 1.03); also checked against guidance lowered or withdrawn and revenue below the company’s own guided range.

Skill is the improvement in log-loss over the naive reader (positive is better). The test is re-run every 90 days on new data; the rules do not change. The quarter-over-quarter change in each axis is tested the same way for information only, and each axis is also submitted as an unstaked Numerai Signals probe; neither result affects what is shown.

Limits

The universe is the S&P 500 by dated membership. Before 26 September 2026 Palanor’s stored membership is the current list carried back, so the history and the back-test leave out companies that have since left the index. Several sources start recently (8-K items from late 2023, the Guidance Ledger from October 2024, insider filings from mid-2024, fails-to-deliver from late 2024), so the test periods are short. A failed test can mean too little history rather than no signal; the gate says which.

Universe-wide series (the share of S&P 500 companies expanding or retrenching, the median Confident ↔ Hedging reading) are published only for an axis that passes, and only on month-ends when at least 70% of members have a reading.

Company pages show Company Posture under each company. See also the corporate events ledger.