version: "1.0.0" name: decision-journal description: Record investment decisions and recommendations with their original evidence, falsification conditions and follow-up reviews; distinguish proposals from user-confirmed or executed actions.

Decision journal

Read investing/mandate.md, investing/decisions/index.md, relevant open decision entries, the current/prior distinct dated reports and investing/research-queue.md. Keep long-term decisions separate from supplementary swing plans. Paths resolve from the repository root.

Record when a decision exists

Create an entry for a material new recommendation, an explicit decision to wait/no change after evidence review, a user-confirmed strategy choice, or an executed action the user reports or a reliable transaction record confirms. Mere ownership, price movement, unanswered research, source failure or another unchanged weekly report is not a new decision. Recommendations are not proof the user acted. Do not reconstruct trades or cost basis from snapshots.

Use D-YYYY-MM-DD-NNN IDs and files investing/decisions/<ID>.md. Reuse the entry on same-date reruns for the same decision; do not allocate another ID simply because wording changes. Keep the index's status consistent with the entry.

Original decision record

Record: created date/cutoff; status (proposed, user-confirmed, executed, superseded or closed); strategy horizon; account/security scope if applicable; the precise decision and alternatives; evidence links and source periods; thesis and valuation basis; material uncertainty/confidence; falsification conditions; expected observations and review trigger; linked research IDs; execution details only when confirmed.

A recommendation may be recorded without complete portfolio data if its limits are explicit. Do not manufacture a price, expected return, risk budget, allocation or realized action to complete fields. Date and preserve the original reasoning. Capture a minimal original-evidence snapshot in the entry: material facts/assumptions/input values, source periods/access cutoff and the relevant report conclusion. A mutable same-date report link alone is insufficient. When a supporting report is already committed, also link its immutable Git commit version. Subsequent report revisions do not replace the entry's original snapshot. Correct factual errors with a dated correction that identifies the affected conclusion; do not rewrite history to match outcomes.

Review

Review open entries when their stated trigger occurs or material new evidence appears. Before adding a review, identify it by D-ID, trigger/event identity, review cutoff and evidence version. Reconcile an existing review with the same identity; repeating the run with identical evidence adds no review or journal-only commit. Add a dated review only for new trigger/evidence, with evidence known at that review, process assessment, thesis progress, outcome if measurable, and the next decision/research action. A factual correction to a published review is a dated correction, not a silent replacement. If a trigger cannot be evaluated, retain it as pending and identify the blocker.

Separate decision quality from subsequent price performance. Only evaluate realized investment/trade returns with confirmed execution, prices and appropriate benchmark/fees/corporate-action treatment; do not label an unexecuted recommendation a profitable trade. Do not treat a lucky outcome as proof of a sound process, or an adverse outcome alone as a thesis failure. Avoid irrelevant weekly updates to unchanged strategy entries.

If a materially different recommendation replaces the earlier one, link a new decision and mark the old one superseded; preserve both rationales. Only execution confirmation can set executed status. Skill-generated recommendations never mutate holdings or submit orders.

Output

Maintain investing/decisions/index.md, entry files and the stock report's 'Decision changes and follow-up' section. Summarize new/revised proposals, user-confirmed decisions, verified executions and due reviews separately. If no new material decision exists, say so without manufacturing an entry. Link to the frozen original evidence and unresolved R-IDs.

Household investment data may be public under the mandate. Scheduled commits include only the journal/index files generated by this run, following investing/publication.md. Ad hoc journaling is local unless publication is authorized.