Python Market-Structure Detection Module

We're a private research group. We need one well-tested Python module that detects market structure on OHLCV candle data from an exact 14-rule specification we provide (annex below). Two layers: (1) candidate pivot points from a candle-counting rule, and (2) confirmed structure highs/lows, where a level is only confirmed when price revisits the prior extreme. Candles in, two labeled event streams out as JSON. No trading logic, no exchange connections, no UI — the rules are already defined; your job is implementing them exactly. Requirements Python 3.11+, standard library only. Type hints and docstrings. Two entry points with identical results: run(bars) (batch) and on_bar(bar) (streaming, one candle at a time). Streaming must never revise an already-emitted event — an event is emitted once, at the bar where it becomes knowable, and never changed. We test this by comparing the two modes on data you haven't seen. Output as JSON, two event types: candidate pivots (side, price, bar index/time, forming bar index/time) and confirmed structure highs/lows (side, price, bar index/time, revisit bar index/time, label HH/HL/LH/LL). pytest tests covering each numbered rule in the spec, plus the fixture files we provide passing. Clean, commented code — each rule number cited at the line implementing it. Our team maintains this after handover; write for that reader. Checkpoints (required): we use the module on three timeframe levels, which we call L1, L2, and L3, and we provide a candle dataset for each. Work level by level. When your L1 detection is working, stop and send us 3 chart examples (simple images are fine — matplotlib is allowed for these images; the module itself stays stdlib-only) showing your detected pivots and confirmed structure highs/lows drawn on the L1 data, and wait for our review before moving on. Then the same for L2, and again for L3: 3 chart examples each, reviewed before you continue. These checkpoints are how we catch misreadings of the spec early instead of at delivery. Acceptance & payment: $200 fixed via escrow, released when all three checkpoints have been reviewed, the provided fixtures pass, our own additional test data passes (batch ≡ streaming included), and the code reads clean. One round of fixes for spec divergences is included. Delivery expected within ~1 week. Confidentiality: work-for-hire; code is ours on payment; no portfolio use, no public repos, no reuse. What the module is used for downstream is out of scope and not discussed. To apply: one short paragraph on how you'd guarantee streaming and batch modes stay identical, plus a Python sample of yours (link is fine). Applications without both are skipped.