# Verification · 8 September 2026

This is an implemented analysis tool with a shared causal Pine engine. Evidence below distinguishes source compilation, actual engine execution, market rendering, and performance validation.

## Source checks

All three final product sources passed TradingView's `translate_light` server compiler with **0 errors, 0 warnings**. Exact source hashes are in `src/manifest.json`; the final responses and source hashes are retained in [product-server-compile-final.json](evidence/product-server-compile-final.json).

Four Node integrity checks pass: generated source/hash/engine parity, exact fixture engine/manifest parity, footprint entitlement isolation, and bounded fixture function sizes. These are artifact checks, not substitutes for Pine execution.

## Native engine execution

The engine is copied exactly into three standalone Pine probes. Across the three suites there are **25 synthetic scenarios, 129 input frames and 889 assertions**. They check the full A→E→BU path in both normalized directions, very small prices, heavy/medium/unknown-volume tests, expiry, frozen boundaries, confirmed-bar commitment, duplicate/out-of-order frames, failure priority, late reclaims, both-edge ambiguity and the no-spring route.

The original combined probe passed the server checker but the full chart compiler rejected its 161,723 compiled tokens against the 100,256-token limit. It was replaced with three smaller probes; this is why compiler success alone is insufficient. See [split server checks](evidence/probe-split-server-compile.json).

Native results: **889 assertions executed, 0 failures** on TradingView's actual chart runtime. Suite 1: 359/0; suite 2: 262/0; suite 3: 268/0. Records are in [suite 1](evidence/native-suite-1.json), [suite 2](evidence/native-suite-2.json), and [suite 3](evidence/native-suite-3.json). Each contains the expected suite title and executed assertion count; these are not inferred from an empty error list.

The fixtures execute `WEngine.step` on synthetic normalized WBar inputs. They do **not** independently prove market feature extraction, exchange feeds, broker fills, alert delivery, profit, or true Wyckoff labels.

## Market and platform scope

The final base indicator was attached and rendered on TradingView Desktop with no runtime errors. The micro Nasdaq 5-minute snapshot includes 405 loaded bars and a populated hypothesis/event dashboard: [native state](evidence/native-base-final.json), [actual chart screenshot](evidence/native-base-final.png). The screenshot was visually inspected for range, event labels, AVWAP, historical trail and dashboard readability. This is an actual chart render, not a mockup.

The same final indicator also ran on OANDA EURUSD 5 minutes with 400 loaded bars, no runtime error and populated opposing-hypothesis, context and event tables: [second-symbol snapshot](evidence/native-eurusd-final.json).

The available account is Basic. The Footprint variant has server compile coverage; native footprint execution cannot be verified on that plan. The base source and Lab contain no footprint request, so they do not inherit that entitlement requirement.

The final Lab also executed on the native chart with no runtime errors and a populated dashboard: [Lab snapshot](evidence/native-lab-final.json). A development smoke run on the loaded micro Nasdaq 5-minute data (10 August–8 September 2026) produced 8 closed trades, 2 winners and approximately −446.03 USD with the unchanged generic 0.05% commission and 2-tick slippage settings: [captured strategy report](evidence/native-lab-smoke-report.json). This small sample was negative and was not used to optimize the settings. It establishes that entries/exits execute; it does not establish a trading advantage. Market-specific costs and held-out validation remain necessary.

Closed-bar state processing and previous confirmed HTF requests are implementation properties. Full native Bar Replay / frozen-history prefix comparisons and held-out multi-market profitability validation are not claimed. Footprint data may change with underlying historical granularity. No win rate or expected return is claimed.

## Development recovery

A connector's `pine_new` replaced editor text without creating a new saved script identity. The original Fractal script was recovered from its saved history, and the original chart entity was restored from its exact prior source snapshot. Original indicator inputs matched; unrelated chart studies were unchanged. Verification records stay in the local `docs/recovery` directory and are deliberately excluded from the distributable because they contain unrelated user source.

Subsequent QA used a specifically identified isolated chart tab and an actual newly saved script. No indicators were published and no account orders were placed.