Apply syminfo.mincontract as the engine's lot grid in the Docker harness

[luisleo526]

What

docker/run_json.py (the harness docker/entrypoint.sh runs) now applies mincontract from the --syminfo file (PINEFORGE_SYMINFO in the entrypoint) as the engine's lot grid. Until now it applied only mintick, pointvalue, timezone and session, so a percent-of-equity strategy traded quantities below the instrument's lot size.

• A positive finite mincontract is set through the existing strategy_set_syminfo_metadata export, as the key qty_step (the engine floors order quantities to that grid) and as mincontract (what a script's syminfo.mincontract reads), before the other four setters.
• Absent or null: nothing is applied. The report and its fingerprint are byte-for-byte what they were, apart from elapsed_seconds.
• Any other value (0, -1, "0.001", true, NaN, Infinity, a list, an object), or a strategy library without strategy_set_syminfo_metadata: the run fails before it starts with one {"engine":"pineforge","error":"syminfo.mincontract must be a positive finite number, got <value>"} line on stdout (harness exit 1, entrypoint exit 4). It never runs without the grid it was given.
• An applied grid is recorded as applied_runtime.syminfo ({"qty_step": v, "mincontract": v}), so it is in fingerprint.provenance.runtime and a gridded run has its own digest.

The engine library and the C ABI are unchanged. docker/README.md, the docker/entrypoint.sh header comment and CHANGELOG.md document it.

Why here

pineforge-release vendors docker/run_json.py from the engine tag at every release (scripts/sync-harness.sh and the sync step of its handle-upstream workflow), which overwrites its own copy. The same change is open there as pineforge-4pass/pineforge-release#21 (head 857096f), and it only reaches a published image if the file also lands here. docker/run_json.py on this branch is byte-identical to that PR's file.

The case

One frozen case: BTCUSDT 4h, EMA 20/50 crossover, long only, 100 % of equity, 0.1 % commission, initial capital 10,000, 2,188 bars of Binance public spot klines, engine 1.0.1 and codegen 1.0.1 inside the release image.

	mincontract null or absent	mincontract 0.00001
trades (rows)	27	22
rows below one lot	5	0
first trade quantity	0.081668384426108	0.08166 (the engine reports 0.08166000000000001)
every quantity a whole number of lots	no	yes
fingerprint digest	sha256:e4315ec2…	sha256:b9e00cb9…

Tests

• docker/run_json_syminfo_test.py (pytest) covers the setter order and values, absent and null, every rejected value with its message, a library without the setter, the declared C signature, main() with the report and the fingerprint, --bench (every state gets the grid, none leaks), and the structured failure in the body run and in both bench loops. On main's harness 32 of its 37 tests fail; on this branch all 44 tests in docker/ pass. Like the older docker/*_test.py files, it is not registered with ctest or CI (it runs with python3 -m pytest docker/); wiring those tests in is a separate change.
• Image run: the release image built twice from pineforge-release's docker/Dockerfile (engine 1.0.1 static library, codegen 1.0.1), once with main's run_json.py and entrypoint.sh and once with this branch's. On the case, main returns 27 rows and ignores mincontract. This branch returns 22 rows with mincontract and 27 rows without it or with null, row by row equal to the recorded rows (1e-9). -1, 0, "0.001" and true exit 4 with the error line above.
• Reports without mincontract are byte-identical between main and this branch, with elapsed_seconds set to 0, for six runs: no syminfo file; the case's mintick, pointvalue, timezone and session only; the full instrument record with mincontract null; the same record without the key; and a second instrument (mintick 0.5, pointvalue 2) without the key and with null.
• scripts/fingerprint_self_test.py, scripts/test_report_schema_keys.py and tests/test_entrypoint_indir.sh pass.

🤖 Generated with Claude Code