run-series.cmd: sync on a partial build, stop only on a total failure

[mspinola]

The failure this fixes

The wrapper stopped on any non-zero code from the build. On 2026-09-25 that cost five sessions on both replicas: TradingView restated two put/call closes, the build refused those two by design, and the wrapper exited before its syncs while thirteen breadth series sat correct on the producer.

The change

It reads the build's new exit codes (marketdata EXIT_PARTIAL): sync on 0 and on 2, stop only on 1. The refusal stays visible, because the wrapper exits with the build's code rather than the push's, so a partial run still reports 2 to the operator and the verifier.

Mirroring on a partial is safe by construction: the store is a valid state whatever refused, since the build never writes a bad bar and writes parquet atomically, so a mirror can only make a replica fresher.

Two details worth the review:

• Testing a specific code needs equality, not if errorlevel, which is true for that code or any larger one and would read 2 as 1. Two chained if not string compares are the cmd idiom and avoid a numeric parse on an empty ERRORLEVEL.
• The push now captures its own ERRORLEVEL on its own line, so the final exit /b %RC% reports the build's outcome without masking a push failure.

Deployment

This is the versioned example. The live copy at C:\Users\matt\code\cotdata\scheduler\run-series.cmd has to be updated by hand, and it needs marketdata's side merged first or the wrapper will never see a 2.

Noticed, not fixed

run-series.cmd, run-prices.cmd and run-equities.cmd each carry one >= inside a REM line, which every one of these files' own headers warns against. They evidently run fine nightly, so it is latent rather than broken. I kept my additions clear of angle brackets and left the existing three alone.

Test plan

• Suite 225 passed, ruff clean (the examples are not linted or executed by CI; this is a documented wrapper).

Build side: mspinola/marketdata#35.

🤖 Generated with Claude Code