Usage

A whole file: validate

validate(rows) is the check to run before training. It is check_item on every row, then check_batch on the finished set, so it covers both the checks that need no other row and the ones that need the whole set (shares, floors, pairs, phrase and formula limits).

One item: check_item

check_item(row) runs the checks that need no other row and raises DataError on the first refusal. It also refuses an apply item (split: apply, or an id with -AP-), because apply items are never trained.

An apply item has its own check, check_apply_item(row). It applies the same content and shape checks without the never-trained refusal, plus the apply rules, and it refuses any row that is not an apply item.

One item against the accepted rows: check_against

A generator that wants to hand a refusal back to its writer uses check_against(row, accepted). It returns a list of (code, message) pairs, one example per code, in a fixed order. An empty list means the row may be accepted. Run check_item on the row as well.

from aspire_si_gates import check_against, check_item

check_item(row)
for code, message in check_against(row, accepted_rows):
    print(code, message)   # hand the message back to the writer as it is

The codes never change. A message changes only on purpose, and a test pins the table.

The figures behind the checks

batch_report(rows) and freeze_report(rows) return the figures as dicts, so a generator or a reviewer can read the shares and counts a gate used.

Trace leaks

check_trace_leaks_label(row) refuses a kind C row whose checker trace names its own planted flaw, or any flaw stem outside the two-step carve-out. It is wired into check_batch.