refactor(vintage): drop the flow decomposition, which was a less capable copy

[mspinola]

Removes vintage_flow.decompose and the cotdata-vintage flow subcommand. Requested after the duplication was measured from the consumer side.

The two were not alternatives, they were one function

vintage_flow.decompose (built 2026-07-30) and crowdmon.futures.flow.decompose (built 2026-08-01) were carried side by side as "defensible alternatives answering slightly different questions". Measured:

cotdata.vintage_flow.decompose  ==  crowdmon.futures.flow.decompose
                                    at tolerance=1.0, gap rule off

On 27 markets and 135,835 transitions from 2006 to 2026:

	result
row sets	identical, 135,835 both, 0 left-only, 0 right-only
label agreement	100.000000%, zero mismatches
d_long, d_short, d_net	identical on every row

Not a coincidence. Both take the dominant leg as argmax(|ΔLong|, |ΔShort|), both break exact ties to the long leg, and both treat a doubly-unmoved week as quiet unconditionally. Setting tolerance=1.0 makes nothing ever mixed, which is precisely this classifier.

So this copy was never a second opinion, it was the less capable one: it could not decline to label a genuinely two-sided week, and it differenced across a 294-day absence as though it were a week. The general implementation stays in crowdmon, where the module spec puts the positioning engine.

The dedup could not go the other way. crowdmon/tests/test_boundaries.py forbids cotdata from importing crowdmon, because a producer depending on its consumer inverts the dependency direction. The equivalence is asserted from that side, in crowdmon/tests/test_flow_equivalence.py, and written up as crowdmon/docs/design/amendments-2026-08-02.md §B29.

No deprecation path, and the reason is checkable

claim	evidence
not exported	absent from cotdata.__all__
never documented	no mention in CHANGELOG.md under any version
not in a published release	PyPI carries 0.1.0, which predates the whole vintage subsystem. The 0.2.0 note says in terms that "tagging it is not publishing it"
one caller	the CLI subcommand removed alongside it

Kept, dropped, renamed

Kept. zero_sum_check, from_vintage, from_current_store. The zero-sum identity is a claim about cotdata's own parse, and crowdmon.futures.cot_adapter runs it on every load.

Renamed. cotdata-vintage flow becomes cotdata-vintage zero-sum, keeping the half of that command that was cotdata's own and adding the rounding-tolerance count plus a report of non-weekly intervals (COT was fortnightly before 1992-10-13). A command named flow that no longer decomposes flow would be worse than a renamed one. Verified against the real store, and it reproduces §7 Finding 1 exactly: gold 2026-07-21, OI 383,368, both sides 351,385, gap 31,983, and 1926/1926 weeks balanced.

Dropped deliberately. min_frac_oi, the optional dead zone. It defaulted to 0.0, nothing set it, and the surviving implementation answers the same problem with a dominance tolerance that has a sensitivity sweep attached. docs/design/cot_vintage.md §7 records where that argument turned out half right: parameter-free was a real virtue, and it cost the ability to say a week was two-sided, which is the modal outcome at 60% of weeks.

Coverage did not vanish with the tests

Every behaviour the removed tests pinned is pinned on the consuming side, and one that was not already covered there was ported rather than dropped: oi_corroborates is live code in crowdmon but its only test in the workspace lived here. It moved in crowdmon#27, with a second case the original did not cover (the column must be null on mixed/quiet/gap). tests/test_vintage_flow.py carries the mapping table.

A new test_decompose_is_gone_and_stays_gone guards re-adding it out of habit. That is not hypothetical: this classifier was already re-derived once, by a session that did not know this module had held it since 2026-07-30.

docs/design/cot_vintage.md §7 is kept rather than deleted. What it records is a result about the canonical schema obtained by running a decomposition over it; the zero-sum sweep, the spreading finding and the validate() finding do not depend on where the classifier lives.

Merge order

Either order is safe, verified both ways rather than reasoned about:

against	result
crowdmon PR #27 vs this branch	433 passed, the 4 equivalence tests skip with a stated reason
crowdmon PR #27 vs current cotdata main	437 passed, equivalence enforced

Checks

• 262 tests pass, ruff clean
• Rebased onto a8a2477 (Release 0.2.0), which landed mid-work. pyproject.toml and __init__.py are untouched and stay at 0.2.0
• No em dashes in added prose

🤖 Generated with Claude Code