registry: stitch the 1998 grain re-code at scale 0.2

[mspinola]

Summary

• ZC, ZS, ZW, ZO and KE gain a hist_codes predecessor at scale 0.2: 002601, 005601, 001601, 004601, 001611. The five CBOT/KCBT grains re-coded at the 1997/1998 year boundary and the registry carried only the new codes, so the store's grain history began 1998-01-06 although the CFTC publishes it from 1986-01-15.
• The old codes are the same contracts reported in thousands of bushels rather than contracts (5,000 bu each). Evidence, measured against the CFTC deacot1986_2016 bundle:
  • every position column is divisible by 5 on 99.3 to 100% of pre-1998 rows and on 0% after the switch
  • open interest drops 5.0x (ZC), 5.5x (ZS), 5.0x (ZW), 5.7x (KE) at the seam; oats 3.8x, which is a real OI move (its rows are still 100% divisible by 5)
  • reportable trader counts do not move: corn 366 to 345, wheat 164 to 169, oats 31 to 30
• No overlap week (old ends 1997-12-30, new starts 1998-01-06), so the overlap rule from get_cot: overlap weeks keep the larger scaled open interest #122 is not exercised.
• Registry comment carries the evidence inline; CHANGELOG Unreleased / Added.

Test plan

• tests/test_registry.py::test_grain_predecessors_scaled_to_contracts: the five mappings at 0.2, and ZM / ZL / ZR still single-code
• Full suite passes; ruff check src tests scripts clean
• End-to-end against the real producer into a scratch COTDATA_STORE: cftc.update(codes=[the ten codes]) then get_cot for each symbol gives 1986-01-15 to 2026-10-06, no duplicate dates, one code label, and a level-continuous seam (ZW OI 93,742 on 1997-12-30 to 97,105 on 1998-01-06; KE 45,071 to 45,252)

Deployment

The producer writes the five new per-code files on its next --cot run after the Windows box pulls this. Until then get_cot returns the post-1998 series exactly as before (a missing predecessor file is skipped). The 08:10 catch-up syncs both replicas unconditionally, so the files reach the Mac the morning after the pull.

🤖 Generated with Claude Code