Live Telemetry

CASS Intelligence Board

History may not repeat, but it rhymes. If the past is cyclical, each turning shares a common anatomy:

elite overproduction
enclosure of what was once open
a vanguard that mistakes its moment for destiny

Most forecasting fails not from bad models but from unfalsifiable ones. So this board does the opposite: every claim is registered in advance, carries a dated falsifier, and stays on the board when it misses. We're not predicting the future. We're testing whether the rhyme is real.

The transitions genuinely in motion — with the evidence base shown alongside every number — and the honest record of how we've done.

⚡ Updated

Loading model state…

📖 Methodology — how this board is built and how it is tested ▶

The MGET framework. The board runs on a four-leg model of systemic failure: Leg 1 — complexity cap / material substrate (energy, logistics, money); Leg 2 — elite overproduction / class friction; Leg 3 — enclosure mutation (institutions becoming opaque, weaponized, or unaccountable); Leg 4 — vanguard alignment (coordinated counter-elites forming). Each leg is scored 0–15 from visible, public evidence only.

Forecasts are bets on mechanisms, not dates. Each entry names its mechanism, its evidence base, and its falsifier — the specific condition that would retire it. When a falsifier triggers, the forecast is retired publicly, on this page. A forecast that cannot be falsified is not a forecast; it is a mood.

Lead-time measurement. Every forecast shows the date it was logged and the date it resolved. The gap between the two is our lead time — measured, not claimed.

The Threshold Watch. The watch is the board's public storm signal. A watch does not say an event is happening; it says conditions are primed. It carries an expiry date, is re-checked weekly, and can only do three things: upgrade, hold, or expire. It never quietly lapses.

Blind backtesting. The model is validated by reconstructing past eras using only period-available information, then scoring the call against what actually happened. The protocol requires three passes: correct mechanism class, directional macro call, and nulls produced where nulls were warranted. To date, seven macro-instrument eras and one counter-case battery have been tested (the core engine's headline suite), plus a separate six-run policy/regulatory track reported apart from the core number:

• Nov 2019 (pre-corridor): board printed 22%, no watch — correct null.

• Sept 2001 (exogenous shock): macro silent at 12%; the vanguard leg caught the shock class at 35–40%.

• Sept 2008 (corridor detonation): board printed ~70%, watch issued; detonation landed inside the stated 30-day window.

• Full historical sweep (n=8): Reformation 1517, Glorious Revolution 1688, French Revolution 1787, 1848 revolutions, 1968 counterculture, plus three control eras — legs scored on period wire only. Score: 6 of 8 clean PASS, 1 PARTIAL, 1 MISSED.

• Counter-case battery: Soviet 1965, Singapore 1990, China 2000 — raw opacity shows ~zero conversion to decay timing; Leg 3's mutation signal, not opacity itself, carries the timing power.

Full protocol: CASS Blind Backtest Protocol (Standing Methodology), Samuel Adams, 2026, updated 2026-09-22. Headline sample: n=3 original + n=8 historical sweep + 3 counter-cases (14 eras; 4 PASS, 1 STRAIN, 1 MISS, 1 counter-case rejection class). A secondary track of 6 policy/regulatory blind runs (Basel III, GDPR, Medicaid expansion, No Surprises Act, Oregon Measure 110, PPACA subsidies) is reported separately and never mixed into the headline number. Known limits are stated in the protocol, including the risk of outcome contamination in self-graded runs and the provenance caveat that sweep eras are reconstructions rather than true blind runs. One falsifier-triggered failure (PPACA enrollment) is honestly logged on the secondary track. The Falsifiability Collapse Protocol — governing claims structured so that falsifying evidence is systematically unobservable — is a standing clause. Per-era leg scores and grades are recorded in the MGET control ledger.

🧪 Confirmation Standard — the 2026-09-22 ratification ▶

Every card on this board carries a registered confirmer: a pre-registered observable that, if seen, counts as confirmation — with base rate, tier, window, and dedup rule declared before the fact. Registered after the prediction, before the evidence.

Tier convention (ratified 2026-09-22, provenance: Iran naval card): Tier 1 confirms the mechanism is live; Tier 2 confirms the claim. Only Tier 2 counts as skill evidence. A Tier 1 hit upgrades attention, not the score.

Counterfactual control: conditional claims ("X refuses regardless of outcome") register which branches are unconfirmable this cycle, so a routine outcome can't be spun into confirmation. Example: the Regime-Loss card's win-branch is recorded as unconfirmable; its Tier 2 requires loss AND refusal.

Dedup rule: the confirmer binds to fork-specific observables, not to shared-theater events already serving as tripwires on adjacent cards. One event, one confirmation — no double-counting across the board.

Movement honesty: ⚡ badges key to last_moved, stamped the moment a card's status, evidence, or adjudication changes. Retroactive stamps carry provenance; movement is never fabricated to light a badge.

📋 Data Dictionary — every field on this board, defined ▶

Definitions for every term used in the Forecast Register, Macro Forecast, and MGET Ledger below. Written so a first-time reader can interpret any card without prior context.

Confidence %	The board's probability estimate that the statement resolves as written. Rendered on the card's badge; the sparkline beside it shows how that number has moved over time. A flat line at a fixed value means no new evidence moved the estimate — itself information. Known defect (flagged 2026-09-22): historical entries mix 0–1 and 0–100 scales in the same field; normalization to 0–100 is planned at export. Compare only within a card's own history.
⚡ Movement	Badge on a card or theme group showing recorded activity within the last 7 days (from the `last_moved` timestamp). Movement stamps are progressively enriched — only a handful of cards carry dated stamps so far, so absence of a badge is not absence of change.
Theme group	One of five master categories (Gulf & War Escalation, Domestic Consolidation & Repression, Economic & Material Substrate, AI & Substrate Emergence, Method & Framework) that every forecast card belongs to — exactly one. The theme header is a collapsible master card; its status sub-headers (confirmed / emergent / pending / failed) list that group's own cards.
Confirmer	Evidence registered against a prediction's falsifier criteria, scored to a tier: Tier 1 = direct on-record event matching the falsifier as written; Tier 2 = partial or corroborating; Tier 3 = interpretive/amplifying. Registration tiers are kept separate so a Tier 3 interpretation never upgrades a Tier 1 confirmation (Gauntlet Case 003).
Counterfactual control	A comparison case logged alongside a prediction — a matched scenario where the forecast mechanism should NOT fire. Its purpose is dedup: distinguishing the mechanism's signal from events that would have happened anyway.
Canonical source	`data.json` is the board's single source of truth. The shard files (`pred_shard_.json`, `ledger_shard_.json`) are build inputs, not the live layer — deploys render from `data.json`.
Status: pending	Statement is live; no falsifier has triggered and no resolution date has arrived.
Status: emergent	Partial confirmations are accumulating but the statement's full resolution criteria are not yet met — the mechanism is visibly underway.
Status: confirmed	The falsifier's symmetric criteria have been met — the event happened as specified, on record. Note: confirmation validates the honesty of the registration process, not the skill of the forecast (disclosed limit, Gauntlet Case 003).
Status: failed	The falsifier triggered against the prediction. Logged publicly — misses stay on the board.
Falsifier	The specific, dated, observable condition that would kill or resolve the prediction. Written before the outcome is known. If the falsifier is ambiguous, the prediction is not instrument-grade.
Evidence confirmed	Facts on the record that support the statement. Listed itemized so each can be independently checked.
Evidence contradicted	Facts on the record that cut against the statement. Logged with equal prominence — the board's honesty test.
Watch (forward)	Named signals to monitor next — the specific events that would move confidence up or down.
⏱ lead time	Interval from the log date to the resolution date, shown on resolved entries. The board's calibration record.
Macro Forecast	The board's single highest-conviction systemic call. One at a time; falsifiers armed; retired publicly if triggered.
MGET Ledger	Running log of mechanism tests: each entry records a real-world event scored against the model's legs (complexity cap, elite overproduction, enclosure mutation, vanguard alignment). Verdicts: confirms / strains / undermines / neutral.
MGET legs	The four structural legs of the model being tested by each ledger entry. An entry may test one or several.
Forward signal	What the ledger entry implies to watch next — feeding the next day's scan.
Sparkline	Per-card confidence trend line from history.json — daily snapshots, oldest to newest, red if rising, green if falling. Hover for dated values. (History note: sparklines shipped v2.4, removed v2.6 for signal-to-noise, re-added with the share-card engine — earlier change-log entries reflect that back-and-forth.)
Watch (term)	A watch is NOT a prediction. It means conditions are primed — if a spark lands, it will travel fast.

📜 Change Log — how this board evolved ▶

2026-09-27 · v3.1.0	New "Today's Read" collapsible panel above the Forecast Register: hand-authored plain-English daily trend summary (10th-grade level) with a Watching-next list, rendered from Daily Read.md at build time (new daily_read field in data.json). Corpus-share scoring replaces raw-count velocity in the wire sweep (fixes instrument ceiling); DW added as ninth source; Eurasia Hemisphere Pivot card 68→70 (Argentina peg-defense entry).
2026-09-22 · v3.0.2	Movers card persistence fix (export now preserves movers across deploys; canonical hand-maintained overlay); timestamp rendering fixed; stateful-bug class closed. Methodology section updated: headline suite n=7 macro eras, 6-run policy track reported separately, Falsifiability Collapse Protocol promoted to standing clause.
2026-09-21 · v3.0.0	v3.0.0 — collapsible MGET Ledger (215 entries, newest-first, verdict filter chips); header explainer defining the four legs and verdict vocabulary; fix: prediction card field mismatch (confidence_score → confidence) restoring badge/sort/share-card fidelity.
2026-09-21 · v3.0.1	Staging pipeline rebuilt on MCP push protocol (staging-first verify before prod); data rebuilt from canonical /space at 215 ledger entries; VERSION bumped.
2026-09-21 · v3.0.0	Shard-aware loader (renderer expands _sharded arrays, killing the predictions.forEach error); movers card rebuilt: collapsed by default, descending order, 24h window keyed to newest ledger entry, one-line format (date · title · confidence shift + evidence); macro_forecast restored as build-time derivation from canonical prediction card; deploy pipeline hardened: 40KB transport cap, stdin payloads, API-based verification.
2026-09-17 · v2.8.1–v2.8.7	Share-card engine rebuilt after a deploy incident: shell/data separation hardened, card renderer now pulls live sections (plain-English, evidence, falsifier) at click time; plain-English auto-draft persisted in the export layer so it survives deploys.
2026-09-17 · v2.7	📸 Share Card on every prediction card and the macro forecast header. Card exports the claim, plain-English translation, evidence chain, and armed falsifier as one shareable image.
2026-09-17 · v2.6	Sparklines removed (low signal-to-noise); share-card engine v1 shipped.
2026-09-15 · v2.4	Added Data Dictionary section (15 field definitions) and this Change Log. Data layer rebuilt from live source-of-truth export after a stale-data bug was caught: Hormuz entry flipped to confirmed per its own falsifier criteria.
2026-09-15 · v2.4	Confidence-history sparklines added to every prediction card — daily snapshots auto-merged on each deploy via persistent history.json.
2026-09-14 · v2.3	Methodology explainer section: MGET legs, falsifier discipline, lead-time, Threshold Watch, blind backtest results.
2026-09-14 · v2.2	Professional layout; log/resolve dates with lead-time display; Threshold Watch banner indicator with expandable issuance text.
2026-09-14 · v2.1	Forecast language, confidence normalization, public header.
2026-09-14 · v2.0	Data-driven rebuild: board renders live from data.json; live 94% macro forecast; credentials stripped from public files.
2026-09-11 · v1.x	Early auto-sync builds — static board, first public deployment.

Full commit history: github.com/pilot2969/cass-dashboards