On the record

Track Record

Every Score Call we have published, graded against the rescore that followed. Only calls that were actually published count — a draft that was never published is never scored, however it would have turned out.

0
completed calls
0
correct on direction
0
incorrect
—
direction accuracy
Figures as of 11 November 2026. Source: the Score Calls Ledger (0 published calls) and Earnings calendar (no call published for any listed earnings date). Direction accuracy is undefined until the first published call is graded — we show a dash rather than 0% or 100%.

What Score Calls are

Before a company's earnings, we publish item-level predictions for its next rescore: the A1–D4 section items and the total score, each with a direction (up, down, or flat) and a condition. After earnings, we rescore within 24 hours and grade every call as hit, miss, or void. Published calls are immutable: each carries a SHA-256 hash and an archive snapshot, and is never edited or deleted. Calls cover business scores only — never stock prices, price targets, or buy/sell judgments. Full rules: Score Calls Ledger.

How accuracy is measured

Direction. A call is correct when the rescore moved the way the call said it would: “up” means the item or total score rose, “down” means it fell, “flat” means it was unchanged. A call is incorrect when the rescore moved another way. A call is void when the condition it depended on cannot be judged from the disclosed results.

Two rates, not one. Directional calls (up/down) and flat calls are reported separately, so a flood of safe “flat” predictions cannot inflate the headline rate. The headline figure on this page is the direction accuracy of TOTAL-score calls only: the number of TOTAL calls whose predicted direction matched the actual rescore, divided by the number of completed TOTAL calls.

Only published calls count. A call counts only if it was published before the earnings date, with a recorded publication time and hash. Drafts, internal notes, and post-earnings reconstructions are excluded and listed below as not counted. Rescores are taken from the published post-earnings rescore record (Earnings) and the quarterly score history.

Completed calls

CompanyEarnings datePredicted TOTALActual rescore TOTALResult
No completed calls yet. No Score Call has been published and rescored as of 11 November 2026 — see the Score Calls Ledger, which also shows 0 published calls.

Not counted — drafts and unpublished calls

These were never published, so they are not scored here — even where the earnings outcome is already known. Showing them without a grade is the point: a track record that graded its own drafts would not be a track record.

CompanyEarnings dateDraft TOTAL prediction (unpublished)Actual rescoreWhy not counted
Palantir (PLTR)2026-11-02Up — total rises by at least +2 points (score before: 65/B). The draft also held 4 item calls: A1 up, B4 flat, B1 flat, D4 flat.65 → 65, unchanged (B). Q3 2026 earnings report §④: “유지 — 65/B 유지” (no item changes); quarterly history confirms 2026Q4 at 65/B.Never published. Source file data/calls/2026q3/palantir.json has status: "draft", published_at: null, sha256: null, and no results recorded. The 2026-10-24 build report states it was a draft only, with publication awaiting approval that was never recorded.

Note: had this draft been published, its TOTAL call would have been graded incorrect (predicted up ≥ +2; actual unchanged). It is not graded, because it was not published.

What this page does not claim

This page does not predict stock returns, and it says nothing about whether any company's shares will rise or fall. Score Calls predict changes in our own Blitz Index scores — a research judgment about the business — not market prices. A correct score call is not a profitable trade, and an incorrect one is not a losing trade.

With zero completed calls, there is no accuracy rate to interpret yet. When calls accumulate, a small sample still proves little: a handful of correct calls is not evidence of skill, and a handful of incorrect ones is not evidence of failure. The sample size is shown next to every rate on this page for that reason.

Scores follow model v1.0 (frozen). Past scores are never rewritten; each rescore is a new quarterly point.

Sources

Completed-call count and publication status: Score Calls Ledger (0 published calls) and data/calls/ (one file only, 2026q3/palantir.json, status draft). Rescore record: data/rescores/ (empty — no published rescore file) and scores/SCORE_DELTAS.md (no Palantir row, consistent with no score change). Palantir draft prediction: data/calls/2026q3/palantir.json. Palantir actual rescore: research/earnings_reports/PLTR_Q3_2026_report.md (§④, unchanged 65/B) and scores/history.json (PLTR 2026Q4: 65/B). System rules and publication procedure: reports/2026-10-24-calls.md. Page compiled 11 November 2026; figures change only when a new call is published and graded.