Mover v3 data files: XNYS session calendar and sourced SEC Section 31 / FINRA TAF fees (freeze precondition 2) - #186

Satisfies the data part of protocol-core-draft.json freeze_preconditions[2]: the XNYS session calendar from exchange_calendars 4.13.2, fees-v3.json with sourced SEC Section 31 and FINRA TAF rows from 2017-01-01 through retrieval (2026-09-24), and both amendment files (empty). It adds new files only. The protocol, README and other mover-v3 files are unchanged; the resolution/freeze PR copies the pins below into the protocol.

Pins (sha256, bytes)

• blueprints/us-equities/mover-v3/data/session-calendar.json fba93303b31f68ba9fceddcaa43e09c33649852313bc948a8748abf648abe965 502703 bytes
• blueprints/us-equities/mover-v3/data/fees-v3.json f45bed8b5b826a927716d3122ef9f0d07ce7adcb00028a9a343577f5a2677e3a 59794 bytes
• blueprints/us-equities/mover-v3/data/session-calendar-amendments.jsonl e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 0 bytes
• blueprints/us-equities/mover-v3/data/fees-v3-amendments.jsonl e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855 0 bytes

The same values are in blueprints/us-equities/mover-v3/data-pins.json, with the generator inputs' hashes.

Session calendar

• XNYS, exchange_calendars 4.13.2, tzdata 2026.4 (IANA 2026d); requested 2016-01-01 to 2030-12-31; sessions 2016-01-04 to 2030-12-31: 3769 sessions, 32 early closes (13:00 ET), 0 late opens. Each row: session, open/close in America/New_York (with offset) and UTC, early-close flag.
• 2016-2020 early closes: 2016-11-25, 2017-07-03, 2017-11-24, 2018-07-03, 2018-11-23, 2018-12-24, 2019-07-03, 2019-11-29, 2019-12-24, 2020-11-27, 2020-12-24 (includes the five the protocol names). 2021-2026 early closes equal the list mover-early-entry/rules.py took from the same version. The 2018-12-05 and 2025-01-09 days of mourning are absent; yearly counts 2016-2025 are 252, 251, 251, 252, 253, 252, 251, 250, 252, 250.
• Generator data-tools/generate_session_calendar.py with a hash-locked requirements.lock (uv pip compile --universal --generate-hashes, Python >= 3.12). Its pins equal the versions and hashes of blueprints/us-equities/data/requirements.lock, so the CI venv built from that lock (PROMOTION_GATE_PYTHON) runs the regeneration test. It uses only the pinned tzdata package and refuses any other package versions. Output was byte-identical under CPython 3.13 and 3.14 with this lock and under 3.13 with the data lock.
• Holdout reach: the calendar must extend 432 sessions past the freeze session (N0 = +40, 252 sessions, two 63-session blocks, read due within 15 more). The end of 2030-12-31 covers any freeze session through 2029-04-12.

Fees (fees-v3.json, retrieved 2026-09-24, no unverified periods)

SEC Section 31, $ per million of covered sales, keyed by charge date. Each row cites the SEC order that set it and the next order's statement of the rate it replaced and the last date at that rate (Federal Register full text):

FINRA TAF, covered equity sales, $ per share with a per-trade cap, keyed by transaction date. Each row cites FINRA's versioned rule text (By-Laws Schedule A, Section 1, ?version= links) and the SEC notices of the FINRA filings:

Every quote was checked as a verbatim substring of the retrieved text. data-tools/verify_fee_sources.py re-fetches the Federal Register, eCFR and FINRA sources and checks them again. The exception is one sec.gov sentence (the $20.60 rate stays until 60 days after the FY2027 appropriation): sec.gov answers scripted requests with HTTP 403, so it was read through a web-fetch tool and is labelled that way.

Findings the freeze PR has to decide

1. FINRA paused the TAF for transactions from 2026-10-01 through 2026-12-31 (SR-FINRA-2026-021, filed 2026-09-15, published 91 FR 60435 on 2026-09-23). It also postponed its scheduled increases by two years (SR-FINRA-2026-020): $0.000195/$9.79 holds through 2028, then $0.000232 (2029), $0.000240 (2030) and $0.000249 (2031). Both are in the base file as scheduled rows. FINRA's fee-adjustment-schedule page still shows the old schedule.
2. SEC Section 31 rates apply by charge date, which is the settlement date for cleared sales (17 CFR 240.31(a)(3)(i), quoted in the file), not the trade date. The protocol does not say which key the fee lookup uses. A trade-date lookup is off by the settlement cycle near each rate change.
3. No FY2027 Section 31 rate existed at retrieval. $20.60 stays until 60 days after the FY2027 appropriation. If the SEC announces the new rate before the freeze, it is added to the base file (and the pins change); after the freeze it goes to fees-v3-amendments.jsonl. Scheduled base rows that change later also need a precedence rule for amendment lines.
4. The amendment line format is not defined here. The amendment files are deliberately not hash-listed in manifests/evidence.json, so appending a line needs no manifest edit. Their integrity rule is the protocol's byte-prefix check, and the test enforces it against the pinned empty content. Because .gitignore ignores *.jsonl, both files were force-added, as the repo's other tracked .jsonl files were. Appends to them are ordinary modifications, but any new .jsonl log (run or access log) will also need git add -f.

Checks

Limits

• These are repository integration checks and synthetic assertions. They are not exchange_calendars upstream tests, which were not run. The calendar is exchange_calendars' scheduled calendar; unscheduled future closures are handled only by amendments.
• No market-data or broker calls and no credentials. Network use: PyPI (lock), federalregister.gov, ecfr.gov, finra.org and one sec.gov page read through a web-fetch tool.
• Mover studies: preregistered v1 and v2 (no pass in any family), data corrections D1-D6 #162's mover-early-entry/fees.json was not used as a source.

🤖 Generated with Claude Code