Guard static TA precalculation so user series aliases that are not replayed in precalculate() use the live per-bar TA path instead of producing all-na results. Also emit strategy.exit profit/loss as relative tick offsets for the engine instead of freezing prices from position_entry_price_ before pending entries fill.

Teach codegen type inference to remember inline local declaration types outside helper functions and to carry array element TypeSpecs for for-in loop variables. This prevents string concatenation from wrapping existing strings, including UDT string fields and local string labels, in std::to_string; fixes standard compile failures in generic dashboard/label code.

Generated hour/day/time-field lambdas and timestamp(timezone, ...) used POSIX TZ directly. Emit calls to the engine timezone normalizer so TradingView-style GMT+/- and UTC+/- strings resolve with the same sign convention as Pine.\n\nUpdates timezone and timestamp regression tests to pin the normalized emission.

Preserve comma-separated simple statements so side-effecting assignments and array calls after the first comma are emitted.

Lower namespaced array.fill and array value arguments with typed na/default values instead of skipping or emitting incompatible na<double>() values.

Lower Pine v6 bool(x) casts so numeric na maps to false instead of C++ treating NaN as truthy.

Refresh the matrix PCA golden after the prior TA-precalc safety change and cover the new validation cases.

Reset replayed input.source Series members and native source backing series around static TA precompute so the cache pass cannot leak full-feed history into the real run.

Covered by a regression using input.source with static pivot precompute; validated with full codegen pytest, engine corpus, and all data/standard strategies.

Lower Pine for-loops through the same runtime direction path whether or not a by clause is present. This fixes descending loops such as array.size(x) - 1 to 0 by 1, where codegen previously emitted an ascending <=/+= loop and skipped array-removal bodies. Add regression coverage for explicit-by direction inference and compile-smoke coverage for descending array.remove loops.

Promote mixed int/double ternary branches to double in codegen type inference. This fixes ratio expressions such as range == 0 ? 0 : (close - low) / range being stored as int, which zeroed fractional values and suppressed valid entries. Adds regression coverage and passes codegen, engine corpus, engine ctest, and data/standard gates.

Lower timeframe.* reads inside request.security evaluators against the requested timeframe instead of the chart script timeframe. This fixes weekly/monthly security expressions such as timeframe.isintraday, timeframe.isweekly, and timeframe.ismonthly while preserving timeframe.main_period as the main chart/declaration timeframe.

Validation: focused security-timeframe TradingView probe; pytest -q; engine ctest 78/78; pineforge-engine scripts/run_corpus.sh 252 clean; scraper data/standard 77/77 ok.

request.security expressions like helper()[1] were lowered through a chart-context history lambda, so helper bodies could evaluate timeframe.*, OHLC, and nested math in the wrong context before applying the offset.

Add security-context history for non-TA helper call results and lower math.* calls while inlining helpers inside request.security. Verified with the TradingView-backed pf-probe-security-helper-history-context probe, full codegen tests, engine ctest, engine corpus, and data/standard.

math.min/math.max were lowered to std::min/std::max, which can turn NaN operands into numeric clamp results. Pine propagates na for these math calls, so expressions like math.max(-1, math.min(1, na)) must remain na.

Emit a shared is_na guard for math.min/math.max in both normal and request.security helper contexts. This fixes the Mayur weekly HTF score class where a weekly vwap component should be na and then nz(...)=0, improving the weak-tier probe from 0/4 to 3/4 and the original strategy from 0/289 to 193/304 matches.

Verified: focused tests, full pytest, engine ctest 78/78, data/standard 77/77 ok, engine corpus 252/252 with 251 excellent + 1 documented anomaly.

Register request.security evaluators with expanded runtime timeframe expressions when the timeframe comes from input-backed aliases, instead of default-initialized global members.

Add generic support for time/time[n] inside request.security bar-field history, including dynamic offsets and int64 history storage.

Validated against Nico weak mismatch, isolated security probes, standard data, engine corpus, engine ctest, and full codegen pytest.

Map Pine bar_index and drawing coordinates through pine_bar_index() so chart-history offsets do not leak into generated strategy logic.

Fix history-referenced bar_index[n] by feeding its Series from the offset-aware helper; this prevents unfed history buffers and inflated loop bounds.

Treat array<UDT>.get/first/last results as UDT lvalues when a local mutates their fields, so generated C++ emits references instead of value copies.

Preserve active function parameter TypeSpecs so typed array<UDT> parameters participate in lvalue detection; fixes Fluxchart-style zone state updates generically.