forward-projection-methodology.md

May 2, 2026 · View on GitHub

🔭 Forward-Projection Methodology

📈 Multi-Horizon Forecasting Protocol — 7d / 30d / 90d / 12m / term-end / EP-election
🎯 WEP Decay · Structural-Break Detection · Reference-Class Forecasting · Outside-View Enforcement

📋 Document Owner: CEO | 📄 Version: 1.1 | 📅 Last Updated: 2026-05-01 (UTC) 🔄 Review Cycle: Quarterly | ⏰ Next Review: 2026-08-01 🏢 Owner: Hack23 AB (Org.nr 5595347807) | 🏷️ Classification: Public

🔄 Tradecraft Anchors

Element	Value	Reference
F3EAD Stage	ANALYZE — forward projection synthesises analysis into horizon-conditional forecasts	Cross-cutting protocol used by every prospective horizon ≥ quarter
PIRs Served	Forward-looking PIRs: pipeline transit, coalition trajectory, presidency-cycle hand-overs, seat dynamics	See `political-style-guide.md` §PIR/EEI
Admiralty Floor	Procedural data: A1 (EP API); macro context: A2 (IMF/WB); polling: B2	See `political-style-guide.md` §Admiralty
WEP Requirement	Every probability claim carries WEP band + explicit horizon; bands tighten as horizon shortens (see §3)	See `political-style-guide.md` §WEP + ODNI
ICD 203 Gate	Standard 3 (judgement vs assumption), 4 (alternative analysis — at least 6 scenarios for long-horizon), 5 (information sources), 7 (explain changes from prior run)	See `political-style-guide.md` §ICD 203
SAT(s)	Reference-Class Forecasting (Tetlock), Outside-In Thinking, Indicators & Warnings, Pre-Mortem, Morphological Analysis, Devil's Advocate	See `political-style-guide.md` §SATs

🎯 Purpose

Replace ad-hoc "what's coming up" prose with a horizon-conditional, evidence-anchored forecasting protocol. Every prospective horizon ≥ 7 days produces a forward-projection.md artifact whose probability claims are calibrated against:

Reference-class base-rates (Tetlock outside view) drawn from historical-baseline.md.
WEP decay tables (§3) — bands tighten as the horizon shrinks.
Structural-break tripwires (§4) — explicit conditions that branch the projection into a "regime change" scenario.
Pipeline transit-time priors for legislative procedures (§5) — Monte-Carlo P10/P50/P90 transit times consumed by legislative-pipeline-forecast.md.

This methodology is the data-input contract for the eight new long-horizon and electoral artifacts catalogued in artifact-catalog.md.

1. Horizon Lattice

Horizon	Window	Anchor	Article types
T+7d	next week	run date	`breaking`, `week-ahead`
T+30d	next month	run date	`month-ahead`
T+90d	next quarter (≈ 3 plenary sessions)	run date	`quarter-ahead`
T+12m	next year	run date	`year-ahead`
T+term-end	run date → next EP-election week	next-election	`term-outlook`
T+EP-election ±6m	electoral window	next-election	`election-cycle`

Every horizon shares the same artifact schema; what changes is (a) the WEP decay band (§3), (b) the minimum scenario count (§4) and (c) the mandatory inclusion of forward-projection.md.

2. Reference-Class Forecasting (Tetlock Outside View)

Before producing any probability claim, list the reference class — the set of historical analogues from which a base rate is computed. Pull at least 6 such analogues from historical-baseline.md. Examples:

Forecast subject	Reference class	Lookup
Trilogue completion within 90 days	Last 24 months of trilogues on comparable file types	`data/voting-records.json` + `data/procedures-feed.json`
Coalition cohesion drop > 10pts	Past 3 EP terms — count of comparable drops per quarter	`historical-baseline.md` §Coalition baselines
Spitzenkandidaten arithmetic flip	EP9, EP10, EP11 lead-candidate dynamics	`historical-baseline.md` §Electoral baselines
Presidency-Trio rupture	Last 5 Trios — count of unilateral handover changes	`presidency-trio-context.md`

The reference-class table MUST appear in every forward-projection.md, with at least one row per primary forecast.

3. WEP Decay Table

Words-of-Estimative-Probability bands tighten as the forecast horizon shrinks. Use the band that corresponds to the horizon and the reference-class size; widen one band when the reference class has fewer than 6 analogues.

Canonical surface for authors. This table is the single source of truth for horizon-conditional WEP floors across the EU Parliament Monitor platform. The article-style cross-reference lives in political-style-guide.md §Confidence Labels Are Horizon-Conditional, which deliberately does not duplicate the numbers — any change to the bands below propagates everywhere via that link.

Horizon	Floor band (probability range)	Notes
T+7d	Almost certain (95–99%) / Highly likely (85–95%)	Pipeline data is dense — narrow bands required
T+30d	Highly likely (85–95%) / Likely (60–85%)	Roll-call publication delay tolerated
T+90d	Likely (60–85%) / About even (40–60%)	Three plenary cycles — multiple Council Trios may overlap
T+12m	About even (40–60%) / Unlikely (15–40%)	Commission Work Programme cycle complete; one Trio handover
T+term-end	Unlikely (15–40%) / Highly unlikely (5–15%)	Two EP elections may bracket the horizon
T+EP-election ±6m	Highly unlikely (5–15%) / Almost no chance (1–5%)	Outcome conditioned on election; widen bands accordingly

Calibration check (Pass-2): every forward-projection.md § 4 must contain at least one judgement at the horizon's floor band — if every band is "About even" the analyst has avoided commitment.

4. Structural-Break Detection

Branch the projection into a "regime-change" alternative scenario when any of the following is true:

Tripwire	Threshold	Source
Coalition-cohesion drop	≥ 10 pts in any major group over the analysis window	`coalition-dynamics.md` §Cohesion trends
National-government changes	≥ 3 EU governments change ruling party within the horizon	`historical-baseline.md` + open-source national tickers
Presidency-Trio rupture	Unilateral handover change or programme suspension	`presidency-trio-context.md`
Treaty-revision IGC announcement	European Council announces an IGC	`external-documents` feed
Article-7 / sanctions escalation	Trigger or formal step taken against a Member State	`external-documents` feed + `parliamentary-questions`
EP president vacancy / removal	Resignation, censure, or mid-term replacement	`corporate-bodies-feed`

When any tripwire fires, scenario-forecast.md MUST contain a dedicated regime-change branch with its own WEP band and reference class.

For long-horizon (≥ 12m) and electoral runs, at least 6 scenarios are required (mainline + 2 adjacent + at least 1 regime-change branch + 2 wildcards/black-swans).

5. Legislative-Pipeline Transit Priors

legislative-pipeline-forecast.md consumes a Monte-Carlo prior produced from the trailing 24 months of stage-transition timings (rapporteur → committee → plenary → trilogue → adoption).

For each active procedure produce:

P10/P50/P90 transit time per remaining stage.
Probability of completion within the horizon, expressed as a WEP band.
Critical-path bottleneck: name the stage with the highest variance.

Planned implementation: a future aggregator component, scripts/aggregator/pipeline-transit-model.js, is intended to generate this prior in a later slicing-plan PR. Until that component lands, treat the Monte-Carlo prior as a methodological requirement and have the Stage B agent reason directly over the trailing 24-month stage-transition timings — not a currently-available Stage A/B dependency in this PR.

6. Carry-Forward & Forward-Statement Quality Gate

Every prospective run carries forward open forward statements from prior runs via the JSONL forward-statements-registry. With long horizons (≥ 12 months), the registry windows extend to 1825 days (~5 years).

A carried forward statement that has expired (now > horizon-end of original prediction) MUST receive a status update in Stage B:

Status	Trigger	Action
`resolved`	Prediction confirmed by EP data	Cite roll-call vote / adopted text / procedure event
`stale`	Horizon passed without resolution	Mark as withdrawn; do not re-emit
`extended`	Horizon passed; analyst extends with fresh evidence	New WEP band + reference-class refresh required

Stage C raises a 🔴 RED when the run carries > 2 unresolved expired statements without status updates. Enforced by scripts/validate-analysis-completeness.js via readExpiredUnresolved() (see #1563).

7. Outputs

Artifact	Owner	Required by
`intelligence/forward-projection.md`	This methodology	`week-ahead`, `month-ahead`, `quarter-ahead`, `year-ahead`, `term-outlook`, `election-cycle`, `quarter-in-review` (optional)
`intelligence/legislative-pipeline-forecast.md`	This methodology + transit priors	`quarter-ahead`, `year-ahead`, `quarter-in-review`, `year-in-review`
`intelligence/parliamentary-calendar-projection.md`	This methodology	`quarter-ahead`, `year-ahead`, `term-outlook`
`extended/forward-indicators.md`	Extended layer (per `strategic-extensions-methodology.md`)	`quarter-ahead+`, `term-outlook`, `election-cycle`

8. Quality Gates (Stage C)

Gate	Threshold	Pass-2 trigger
Reference-class table	≥ 6 entries with explicit base-rate	Re-author with deeper historical lookup
WEP decay compliance	Bands match horizon (§3)	Tighten or widen to match horizon
Structural-break section	Non-empty when any tripwire active	Add regime-change branch in `scenario-forecast.md`
Carry-forward hygiene	≤ 2 expired-without-status statements	Set `status` per §6
Pipeline forecast (where required)	≥ P10/P50/P90 per active procedure	Re-run transit model

9. Cross-References

ai-driven-analysis-guide.md — 10-step protocol; this methodology slots into Step 6 (Forecasting) for prospective horizons.
electoral-cycle-methodology.md — extends this protocol with electoral-specific reference classes.
strategic-extensions-methodology.md — long-horizon appendix raises scenario-forecast.md to 60 months when electoralOverlay=true.
historical-baseline.md — reference-class store.

📅 Document History

Version	Date	Summary
1.0	2026-05-01	Initial release — multi-horizon forecasting protocol, WEP decay table, structural-break tripwires, carry-forward quality gate.
1.1	2026-05-01	Marked §3 WEP Decay Table as the canonical surface; cross-linked from `political-style-guide.md` §Confidence Labels Are Horizon-Conditional so the article style guide references rather than duplicates the table.