Investment Partner — Household Portfolio Review

01

A recurring household decision

Extra cash creates a recurring household decision: where can it go while keeping the portfolio inside its own rules? A generic stock pick cannot answer that. The useful comparison begins with current allocation against target, contribution priorities, concentration, and the exposure already carried both directly and through pooled funds.

I conceived Investment Partner to make that comparison repeatable and inspectable: where could incremental household cash go while staying inside the portfolio’s own rules? It applies the household policy to a validated view of holdings and cash, keeps overlap and unresolved uncertainty visible, and explains why a possible destination appears. The tool supports the review; it does not make the allocation. I remain the final decision-maker.

02

Make the policy executable and inspectable

The portfolio policy begins with decisions I make: targets, classifications, contribution priorities, concentration rules, and the basis for applying them. Investment Partner turns that authored policy into an executable contract. Given the same validated snapshot and configuration, it should produce the same ordered review result.

That result must remain inspectable. The system names its denominators, keeps cash explicit, shows current allocation beside target gaps, and carries the governing rule into each review candidate or flag. It does not quietly fill missing target coverage or hide a reconciliation difference. Deterministic here means that I can see why an output exists and which rule produced it—not that the software has developed an investment opinion.

I own the household objective, product and portfolio strategy, requirements, policy, classifications, scope, acceptance decisions, evaluation, and final portfolio judgment. AI tools implement code and tests under those specifications and acceptance decisions. The implementation applies policy I authored; it does not originate that policy or decide whether its output should govern a household action.

03

Holdings are not exposures

I expected fund look-through to be straightforward: expand each fund into constituents, apply the weights, and add the results. That view failed. A household can look diversified at the holding level while several pooled funds carry the same underlying companies. The visible holding is a wrapper; it is not the full economic exposure.

The difficulty was keeping that second view truthful. Fund and security identities have to resolve without ticker guesswork. Constituent coverage and normalization have to preserve what a source actually accounts for. A nested fund can stop before a terminal security. Overlap can use only known composition, while every result carries source, freshness, and provenance. Direct and indirect known exposure must share one denominator without double counting; any residual or unresolved exposure has to remain visible.

This is why look-through became the product’s stickiest correctness problem. More detail is not automatically more truth. Expanding a partial composition and making its remainder disappear would produce a cleaner report and a worse household decision. The useful result is a bounded account of known exposure, overlap, and what the system still cannot resolve.

04

Refuse false precision

Investment Partner therefore treats uncertainty as part of the result rather than a note beneath it. Some exposure is exact-known; other composition is partial or stale; residual, unresolved, and unknown exposure remains separate. Unknown is not zero, partial is not complete, and a source accepted for a bounded review does not silently become current.

Those distinctions change what a concentration result can say. Known exposure may be a lower bound: enough to establish that a threshold has already been crossed, but not enough to prove that an apparently lower concentration is safely below it. The same discipline applies to overlap and allocation review. Known constituents can support a bounded comparison; the unresolved remainder stays beside that comparison instead of being redistributed into a more reassuring answer.

Household trust depends on that restraint. The product shows which part of a result is exact, which part is qualified, and which part remains outside its knowledge. Refusing false precision is how the software keeps incomplete data from becoming a confident portfolio claim.

05

August 19, 2026 — cross the dogfood threshold

I date the meaningful private dogfood review to August 19, 2026. In testimony given August 25, I described it as “pretty promising” and said I “would have taken the guidance.” That is operator testimony, not a repository-proven date or outcome. I am not claiming that I executed a trade, that the guidance was objectively optimal, or that it improved performance.

The repository preserves sanitized evidence dated August 20. It shows deterministic issuer ranking and one bounded hypothetical scenario end to end, alongside qualitative findings and material presentation friction. It contains no private holdings or recommendation, and it does not independently establish the August 19 date. The one-day difference remains unresolved; this Record neither merges the dates into one source fact nor assumes they represent two separate runs.

Each source supports a different part of the milestone: the repository establishes the deterministic ranking and bounded scenario; my testimony supplies the acceptance judgment. Investment Partner crossed a practical trust and use threshold when a real private review produced rule-derived guidance I considered credible enough to follow. That is the milestone—not a launch, a performance claim, or proof that the product was finished.

06

Useful, and still unfinished

As of August 25, 2026, Investment Partner was Operational: it worked in real private household use and remained under active development. Constituent coverage and some source or provenance contracts remain incomplete. Preparing, selecting, and inspecting snapshots and composition sources still requires deliberate operator work. Audit-grade output can also be dense enough to slow scanning—the same detail that supports trust can create presentation friction.

Work continues on partial composition and residual or unknown treatment. There is no live review-time market-data feed, brokerage connection, or order execution.

That unfinished state locates the August milestone honestly. The tool became useful enough to support a consequential household review while retaining the limits that made its guidance trustworthy. Investment Partner can order facts, apply my rules, and show where knowledge stops. Every final portfolio decision remains mine.