Lodestar · reference implementation
A week of Lodestar
Lodestar gives each household a standing mandate (“minimize lifetime tax”, “watch for a better plan”) and works it without being asked: when a deadline approaches, the household changes, the rules or the firm’s assumptions change, or a review falls due. It proposes; an advisor decides. Below is the first week of November 2026 on a fictional book of five households, as the advisor would find it on Monday morning.
Every household here is fictional, built for this demonstration. The figures are computed by the production engines under labeled illustrative assumptions; nothing on this page is investment, tax, or legal advice.
The inbox
What is waiting for a decision. Every figure comes from a stored decision record; nothing happens until someone with an approver key approves, and even then Lodestar only records the decision. It never moves money or contacts a client.
Fill the 12% bracket each year
The household was updated. Recommended: Fill the 12% bracket each year. Projected value $102,614 of lower lifetime tax (vs. the current plan, lifetime federal and state tax through 2055, nominal dollars (not discounted), under profile demo-cma@1.0.0). 1 other alternative considered.
- Convert $100,800 from traditional to Roth in 2026, to the top of the 12% bracket.
- Convert $100,800 from traditional to Roth in 2027, to the top of the 12% bracket.
- Convert $100,800 from traditional to Roth in 2028, to the top of the 12% bracket.
- Convert $100,800 from traditional to Roth in 2029, to the top of the 12% bracket.
- Convert $100,800 from traditional to Roth in 2030, to the top of the 12% bracket.
Alternatives considered, caveats and evidence
- No conversions: $0
- ending net worth: $0; lifetime tax: $420,200
- converts $504,000 over 5 years
- estimated IRMAA surcharges: $0; score = ending net worth − surcharges
Decision record sha256 9f09bba777c15623791e512d64fa621b16d796b1bff5f4934d96a8d84e58bcb8
What is the cost basis of account “brokerage”?
Account brokerage is a taxable account with no cost basis, so every gain on it is treated as unknown; harvesting, location and withdrawal-order results depend on it.
Answering with a figure files a change that an approver confirms before anything is written; confirming the data is right closes the question for good.
Move the book to demo-cma 1.1.0
Assumption profile demo-cma 1.1.0 was published (the book runs on 1.0.0). Moving to it changes the recommendations for 1 of 5 households; $127,726 of projected value moves. Nothing moves until this is approved.
| Household | Recommendation on 1.0.0 → on 1.1.0 | Change |
|---|---|---|
| Retired couple, unequal earnings records | Draw cash → taxable → roth → traditional → Draw cash → taxable → roth → traditional | −$127,726 |
Approving moves 6 mandates pinned to 1.0.0: the book and every household mandate listed. Until then, nothing moves. Each of those 5 household mandates also filed its own re-pin; approving the book's withdraws them.
What Lodestar did this week
Every cycle, including the ones that proposed nothing, with the reason. A quiet cycle is a result: the gate holds back what is too small, already approved, or still in its cooldown, so the inbox only holds what deserves a decision.
Sunday, November 1
- Tax Optimization, The conversion windowHeard: Tax year ends in 60 days
- Bracket-fill Roth conversions: no proposal; proposal 1 for bracket-fill was approved on 2026-07-04 and covers this tax year.
- Tax Optimization, Pre-retiree, traditional-heavyHeard: Tax year ends in 60 days
- Nothing to propose: the current plan is already best for this mandate's strategies.
- Tax Optimization, Retired donor taking RMDsHeard: Tax year ends in 60 days
- Charitable giving: bunching and QCDs: no proposal; proposal 2 already put this policy in place on 2026-07-04, and the recommendation has not changed.
- Strategy Watch, Retired couple, unequal earnings recordsHeard: Tax year ends in 60 days
- Social Security claiming: no proposal; proposal 3 for ss-claiming was approved on 2026-07-04 and covers this tax year.
- Withdrawal order: no proposal; proposal 4 already put this policy in place on 2026-07-04, and the recommendation has not changed.
- Strategy Watch, Brokerage with no cost basis on fileHeard: Tax year ends in 60 days
- Withdrawal order: no proposal; projected value $0 is below the $1,000 floor.
Tuesday, November 3
eventThe advisor records that the primary earner retired: the salary stops.
- Tax Optimization, Pre-retiree, traditional-heavyHeard: The household was updated
- Proposed: Fill the 12% bracket each year ($102,614 of lower lifetime tax).
Wednesday, November 4
eventThe investment committee publishes demo-cma 1.1.0 with a lower expected return (5% from 6%).
- Tax Optimization, The conversion window · Tax Optimization, Pre-retiree, traditional-heavy · Tax Optimization, Retired donor taking RMDs · Strategy Watch, Retired couple, unequal earnings records · Strategy Watch, Brokerage with no cost basis on file · Rules Change (the book)Heard: Assumption profile demo-cma 1.1.0 was published
- Each: filed a re-pin proposal (demo-cma 1.1.0).
The ledger
Moves the advisor approved between July and the end of October. Each entry records the projected value at approval, against the decision’s own status-quo baseline, in nominal dollars, with a signed record anyone can verify later.
| Approved | Household | Move | Projected value |
|---|---|---|---|
| Jul 4 | The conversion window | Fill the 22% bracket each year | $97,130 lower lifetime tax |
| Jul 4 | Retired donor taking RMDs | Give through qualified charitable distributions | $4,840 lower lifetime tax |
| Jul 4 | Retired couple, unequal earnings records | Claim: m1 at 70, m2 at 67 | $763,213 higher ending net worth |
| Jul 4 | Retired couple, unequal earnings records | Draw cash → taxable → roth → traditional | $33,143 higher ending net worth |
How this page is made
The book ran from July 1 to Nov 7, 2026, one simulated day at a time, through the same routing, deadline calendar, strategy sweep, gate and proposal code that runs a customer’s mandates, in deterministic mode (no AI model). Until November the advisor approved what came due; this week is left undecided. The same harness replays scripted scenarios against expectations written in advance, and every change to Lodestar has to pass them.