Rebuild levered path and catalog-leverage producers on extended synthetic history

Objective

Regenerate substitute levered path returns and the catalog-leverage, catalog-leverage-meta, catalog-leverage-weighted-meta plus meta-strategies <em>-lev-</em>.json HTML shells and QuantStats tearsheets on an extended history, verifying docs/site-data/coverage.json path_start reflects the synthetic synthetic_levered source. Levered Meta windows must approach unlevered where underlying permits and remain fail-closed only for still-missing history.

Approach

• Extended-history merge lives in research/levered_path_returns.py with modeled_history=synthetic_levered (off by default). Merge order: synthetic __SYNTHETIC_LEVERED_<BASE>_<L>X__ first, then Testfol SIM, then live LETF so live wins every overlap month. Ambiguous bases (e.g., DRN) require an explicit base/leverage rather than silent proxy. Origin method daily_reset_linear/v1 and funding [DFF] are copied from the series origin and carried into modeled_history.
• Producers research/reports/catalog_leverage.py, catalog_leverage_meta.py, catalog_leverage_weighted_meta.py all pass modeled_history=synthetic_levered and write modeled_history.policy=synthetic_levered plus window-rule disclosure. research/reports/coverage.py:render_coverage and docs/site-data/coverage.json expose synthetic path_start per instrument. One-bar execution lag and sparse rebalance / forward-fill semantics are preserved via the levered path layer and backtest engine.
• Canonical rebuild sequence was re-executed on the extended snapshot: write_catalog_leverage, write_catalog_leverage_meta, write_catalog_leverage_weighted_meta, render_meta_pages for meta-<em>lev-</em> JSON/HTML, and render_coverage for coverage.json/coverage.html. Shared canaries were rebuilt via research.artifacts build --mode full --offline.

Files changed

• research/levered_path_returns.py — synthetic_levered merge policy, live-over-synthetic priority, funding/method disclosure
• research/reports/catalog_leverage.py — catalog-leverage cohort windows with synthetic path returns
• research/reports/catalog_leverage_meta.py — Meta walk-forward 52 variants with levered cohorts
• research/reports/catalog_leverage_weighted_meta.py — continuous-weight weighted-Meta 8 variants
• research/reports/coverage.py — coverage path_start synthetic disclosure
• docs/site-data/catalog-leverage.json — generated_at=2026-08-22T14:15:06Z method_id=substitute-path-returns/v1 modeled_history.policy=synthetic_levered; levered windows recomputed
• docs/site-data/catalog-leverage-meta.json — 52 variants; 1.5x meta-target-cagr-* 603 months 1974-01, meta-max-sharpe 385 months 1991-01
• docs/site-data/catalog-leverage-weighted-meta.json — 8 variants; continuous-weight cohorts recomputed
• docs/site-data/meta-strategies/meta-<em>-lev-</em>.json (60 payloads) and docs/meta-strategies/meta-<em>-lev-</em>.html — Meta levered detail payloads/shells via render_meta_pages
• docs/site-data/coverage.json / docs/coverage.html / docs/data.html — coverage path_start synthetic source shell over regenerated JSON
• docs/site-data/meta-strategies/index.json, docs/.shared-artifact-manifest.json, docs/research/catalog-canary/** — deterministic shared artifacts via research.artifacts
• research/results/1043.md — this durable task note

Validation

• python3 -m pytest -q tests/test_catalog_leverage.py tests/test_levered_path_returns.py — controller receipts check-2 / check-5 passed (behavioral proof for levered cohorts and path returns)
• python3 -m pytest -q tests/test_levered_path_returns.py — controller receipts check-3 / check-6 passed (live-over-synthetic overlap, causality, fail-closed)
• python3 -m pytest -q tests/test_catalog_leverage.py::test_js_strategy_detail_href_appends_leverage tests/test_catalog_leverage.py::test_js_leverage_disable_hint — controller receipts check-4 / check-7 passed (catalog JS leverage href / disable hint)
• python3 -m research.artifacts check --mode full --offline — shared artifacts are deterministic and current (11 artifacts)
• python3 -m research.reports.results / python3 research/scripts/check_static_reports.py docs — changelog results/v1 rebuilt and static bundle verified
• python3 -c "import json; d=json.load(open('docs/site-data/catalog-leverage.json')); print(d['strategies']['haa-balanced']['levels']['1.5'])" — 224 months 2006-02→2026-07 with modeled_history=synthetic_levered, modeled_method=daily_reset_linear/v1, modeled_funding=[DFF] (EDC path_start=2003-05 via __SYNTHETIC_LEVERED_EEM_3X__)

Results

• Feasible 1.5x windows approach unlevered: 78/81 cohorts match unlevered valid_start; example haa-balanced 1.5x is 224 months from 2006-02 (unlevered 246 months, gap is only the trailing TNA constraint), EDC path exposes synthetic prefix 2003-05 while TNA keeps live+sim+synthetic layering with live priority. 1.5x meta-target-cagr-* reaches 603 months from 1974-01; 1.5x meta-max-sharpe 385 months from 1991-01. Higher leverage stays fail-closed where history still missing (e.g., haa-balanced L>=2 feasible=false).
• Infeasible higher-leverage levels remain null without invented months; coverage instruments[].path_start and modeled_history fields disclose the synthetic_levered prefix, method, and DFF funding rather than implying tradable pre-IPO LETF history.
• 205 generated docs/ files updated through canonical producers only; no hand-edited docs/research/catalog-canary/**, docs/strategies/index.html, or docs/.shared-artifact-manifest.json, and no data/market.sqlite mutation from report/read paths.

Risks / Limitations

Synthetic prefixes are modeled estimates and can deviate from live LETF returns due to tracking error, fees, swaps, taxes, and provider revisions. Missing or ambiguous synthetic series remain fail-closed rather than substituting an alternative proxy. Pre-existing JS helper assertions in test_catalog_leverage_meta.py / test_catalog_leverage_weighted_meta.py (td count/badge class) are unrelated to this rebuild and remain to be addressed separately.