Models rebuild, 16 October 2026

Owner request: act on the models review, improve the library and presentation, and investigate international total-return inputs. This is a local rebuild; no deployment or change to registered portfolio rules is included.

Comparison contract

• Today, Explore, Compare and Evidence are separate reading tasks.
• Core groups are editorial classifications, selected without a performance screen. Existing variants and failed experiments remain accessible.
• Comparisons use the intersection of actual observation dates for every selected model and optional benchmark. No filling missing observations. Rebase on the first shared observation after the requested start date.
• Compare growth, annualised return and fall from peak after model costs, before tax. Weekly curves can understate daily maximum drawdowns. Full-record after-tax results are shown separately with their own dates. Never truncate a published full-record after-tax CAGR and present it as a window return.
• A benchmark is described by its actual holdings. Same-assets comparisons, broad-market references and other static alternatives answer different questions. There is no automatic claim that every gap isolates timing.
• These are descriptive comparisons, not new tests of predictive hypotheses. No tuned parameters, new strategy, new trial or claim of significance.

Recording contract

• Returns since entry calculated from current wealth are a recomputed backtest.
• Frozen log returns and their coverage are displayed separately. Sessions recorded on a later India calendar date are reconstructed, even if they happen to be the latest session available to that pipeline run.
• Preserve every existing log row and hash. Derive the late classification from both the stored flag and recording timestamp, so earlier incorrect flags do not create apparent contemporaneous evidence.
• Show only a contiguous suffix recorded on the session date as the latest contemporaneous record; never chain scattered on-time sessions into an apparently continuous return. This is paper tracking, not actual execution.

International data decision

MSCI private inputs do not automatically grant rights to publish derived products. Retain the existing MSCI publication gate. ETF NAV and distribution downloads are technically useful, but website availability is not a grant of public analytical reuse. See docs/reference/international-etf-sources.md. Do not silently replace a registered price-index rule with an ETF total-return rule: the signal, costs, dividend withholding and tax lots would change. Such a version needs a dated specification, source permission, reconciliation, and its own entry date. Historical NAVs already include fund expenses; do not deduct those expenses again. Keep dividend income and capital gains distinct.

Further research, requiring separate specifications

Risk-comparable static alternatives; fixed factor exposures before factor rotation; debt duration; threshold and cash-flow rebalancing; accumulation and withdrawal outcomes. These are not silently added to the model library.