* docs: split market sources by asset class

Closes #68.

* docs: add economic and specialized source references

Closes #69.

* test: enforce source-reference contracts

Closes #70.

* ci: refresh ecosystem qualification snapshot

---------

Co-authored-by: Stefan Jansen <[email protected]>

…(Kalshi, ForecastEx, Polymarket, Polymarket US) (#76)

* feat(kalshi): enumerate resolved markets and trades across live and archive tiers

KalshiProvider gains iter_markets/fetch_markets and iter_trades/fetch_trades.
Both page through the live endpoints (/markets, /markets/trades) and the
archive (/historical/markets, /historical/trades), using /historical/cutoff
to skip a tier that cannot match and dropping duplicates. Filters the
archive does not accept (status, close-time window, multivariate exclusion
combined with a ticker filter) are applied on the client.

fetch_markets returns a typed frame with a normalized resolution outcome
(result in yes/no/void/other or null, settlement_value, settlement_ts), and
fetch_trades returns trade_id, ticker, timestamp, price, count, taker_side,
is_block_trade. Both the *_dollars/*_fp fields and legacy unsuffixed fields
are read. list_markets and the other existing methods are unchanged.

BaseProvider gets a shared rate-limited _request/_request_json helper that
maps 429/404/5xx/transport failures to the shared exceptions and retries
transient ones under the existing provider retry policy, now exposed as
provider_retry. prediction_markets.py holds the outcome vocabulary and trade
schema shared by the prediction-market adapters.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* feat(providers): add ForecastEx and Polymarket US event-contract providers

ForecastExProvider reads the free daily CSVs in the public
forecastex-public-data bucket: fetch_prices (daily OHLC, marks, settlement
price), fetch_trades (every matched YES/NO pair) and fetch_products. Its
fetch_markets derives the resolution outcome: a contract absent from the
next day's prices file has expired, and a final YES settlement price of
1.00/0.00 is the payout. Intraday contracts that drop out with a daily mark
instead of the payout are reported as expired with a null result rather
than guessed.

PolymarketUSProvider reads the public gateway of the CFTC-regulated
Polymarket US exchange (separate from the global PolymarketProvider, which
is unchanged): offset-paginated markets with a normalized outcome taken
from the long market side, /settlement, and price history in fixed or
24-hour windows. The gateway's outcomes labels are not always in
outcomePrices order, so labels are never zipped with prices.

Both are registered as "events" providers outside DataManager routing.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* docs(providers): document prediction-market history and outcome loaders

Add provider pages for ForecastEx and Polymarket US, extend the Kalshi page
with resolved-market enumeration, the normalized outcome columns and trade
history, and list both new venues in the provider index, the
prediction-market source reference, provider selection and the navigation.

Rename the new offline test modules so their basenames do not collide with
the integration modules of the same venue.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* build(deps): lock patched multidict, urllib3 and virtualenv

The dependency audit now flags multidict 6.7.1 (CVE-2026-104874),
urllib3 2.7.0 (PYSEC-2026-4175..4177) and virtualenv 21.7.1
(PYSEC-2026-4011..4014). Upgrade the three transitive pins to 6.9.1,
2.8.0 and 21.14.6; python-discovery moves with virtualenv.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* feat(kalshi): expose mve_filter and keep series, previous price and taker sides

Replace the exclude_multivariate flag with Kalshi's mve_filter parameter
(None, "exclude", "only") on iter_markets/fetch_markets, so combo (KXMVE*)
markets can be included, excluded or selected. /markets receives the value
directly; /historical/markets rejects "only" and takes "exclude" only
without a ticker filter, so the selection is completed on the client.

The normalized market frame gains series (series filter or event-ticker
prefix) and previous_price; the trade frame keeps taker_outcome_side and
taker_book_side next to taker_side.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* feat(kalshi): tier-aware candlesticks, streaming market frames, live contracts

Price history now follows the market's tier. fetch_candles (and
fetch_ohlcv, which shares the routing) reads archived markets from
/historical/markets/{ticker}/candlesticks and live ones from
/series/{series}/markets/{ticker}/candlesticks. An explicit tier or
series_ticker skips detection; otherwise a live 404 falls back to the
archive, and a second 404 looks the series up from the market's event,
which fixes old tickers whose prefix is not their series (FED-23DEC is in
KXFED). Ranges beyond Kalshi's 5000-candle limit are split.

Prices are converted per field by encoding: *_dollars values and
unsuffixed strings are dollars, unsuffixed integers are legacy cents. The
previous blanket /100 on unsuffixed fields would have scaled archive
candles down 100x. fetch_candles returns trade OHLC, mean and previous,
YES bid/ask close, volume, open interest and the serving tier.

iter_market_frames streams fetch_markets' normalized rows in frames of
chunk_size rows for listings too large to materialize; fetch_markets now
delegates to it.

The provider-contracts public job runs the new Kalshi, ForecastEx and
Polymarket US live smoke tests, and the workflow contract test checks it.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* feat(polymarket): bulk resolved-market listing, windowed price history and trades

iter_markets / iter_market_frames / fetch_markets page through Gamma's
GET /markets/keyset (GET /markets refuses offsets beyond 2000), passing
next_cursor back as after_cursor; any other parameter name silently
returns page 1, so a repeated cursor raises instead of looping. The
server honours closed, volume_num_min, end_date_min and end_date_max
(verified 2026-10-08) and caps pages at 100; the filters are re-applied
on the client. include_tag=true brings the market tags, the reliable
topic field now that category is mostly null.

Rows are normalized to the shared resolution vocabulary. Gamma markets
are binary and ordered: the first outcome is the YES side whatever its
label, final prices [1,0]/[0,1] are yes/no, [0.5,0.5] is void, a UMA
"resolved" market with other prices is "other", the rest is null.
settlement_ts is closedTime (else umaEndDate), and the frame carries
token ids, event slug/title, neg-risk and the raw fee fields.

fetch_candles always sends startTs/endTs with fidelity, split into
windows of at most 15 days: interval=max returns nothing for resolved
markets below daily fidelity, and longer spans are rejected with 400.

fetch_trades pages data-api /trades newest first. Past the offset cap of
10000 it moves the inclusive end bound to the oldest second seen and
drops the boundary trades already yielded; a repeated page or a second
too dense to page past raises. Prices are converted to the YES price
and the taker side to yes/no from outcome index and BUY/SELL.

Offline tests use responses recorded on 2026-10-08; each was checked to
fail with its defect planted.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* fix(polymarket): fetch_ohlcv returned nothing for past date ranges

The CLOB treats interval as a range ending now that overrides startTs
and endTs, so fetch_ohlcv's interval=1h chunks came back empty for any
past window (reproduced live on 2026-10-08 for a market resolved in
2025). Chunks now send their bounds with the interval's step as
fidelity. YYYY-MM-DD bounds were also converted in the machine's local
zone; they are UTC days now, and "max" no longer builds a single
multi-year window the CLOB rejects.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* docs(polymarket): document resolved history and run its live contracts

The Polymarket provider page covers fetch_markets / iter_market_frames
(keyset paging, normalized columns, first-outcome-is-YES mapping),
fetch_candles (15-day windows, the empty interval=max history of
resolved markets) and fetch_trades (offset cap and end-bound paging),
and replaces the outdated 7-day range advice. The prediction-market
reference lists Polymarket among the venues with normalized outcomes.

TestPolymarketResolvedHistory smoke-tests the three calls against a
market resolved on 2026-01-01; the provider-contracts public job runs
it and the workflow contract test checks that it does.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

* test(polymarket): skip the local-zone test where time.tzset is missing

Windows has no time.tzset, so the ecosystem qualification job errored in
the fixture that switches the local zone to America/New_York.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

---------

Co-authored-by: Stefan Jansen <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>

)

Bumps the github-actions group with 3 updates in the / directory: [astral-sh/setup-uv](https://github.com/astral-sh/setup-uv), [actions/upload-artifact](https://github.com/actions/upload-artifact) and [actions/download-artifact](https://github.com/actions/download-artifact).


Updates `astral-sh/setup-uv` from 10.1.0 to 10.3.0
- [Release notes](https://github.com/astral-sh/setup-uv/releases)
- [Commits](astral-sh/setup-uv@bec219d...1c37ad0)

Updates `actions/upload-artifact` from 7.0.1 to 7.0.2
- [Release notes](https://github.com/actions/upload-artifact/releases)
- [Commits](actions/upload-artifact@043fb46...cf430e0)

Updates `actions/download-artifact` from 8.0.1 to 8.0.2
- [Release notes](https://github.com/actions/download-artifact/releases)
- [Commits](actions/download-artifact@3e5f45b...9000827)

---
updated-dependencies:
- dependency-name: astral-sh/setup-uv
  dependency-version: 10.3.0
  dependency-type: direct:production
  update-type: version-update:semver-minor
  dependency-group: github-actions
- dependency-name: actions/upload-artifact
  dependency-version: 7.0.2
  dependency-type: direct:production
  update-type: version-update:semver-patch
  dependency-group: github-actions
- dependency-name: actions/download-artifact
  dependency-version: 8.0.2
  dependency-type: direct:production
  update-type: version-update:semver-patch
  dependency-group: github-actions
...

Signed-off-by: dependabot[bot] <[email protected]>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* docs: enforce published link integrity

* docs: replace stale contributor guidance

* fix: retry deployed documentation probes

* docs: remove stale legacy guidance

* fix: tolerate transient external link failures

* docs: describe supported Massive financial statements

* docs: make session validation examples executable

---------

Co-authored-by: Stefan Jansen <[email protected]>

* docs: enforce published link integrity

* docs: replace stale contributor guidance

* fix: retry deployed documentation probes

* docs: remove stale legacy guidance

* fix: tolerate transient external link failures

* ci: enforce minimum package coverage

* docs: align agent test command with coverage gate

* docs: document enforced coverage lane

* ci: retain current setup-uv pin in coverage job

---------

Co-authored-by: Stefan Jansen <[email protected]>

* docs: adopt shared contribution policy

* docs: align contribution checks with coverage gate

---------

Co-authored-by: Stefan Jansen <[email protected]>

Include the real month end in Fama-French filters, include the stored end-date bar in gap detection, request the inclusive macro end date from Yahoo, interpret Finnhub date bounds in UTC, and follow Massive aggregate pagination.

Closes #73

Preserves the five applicable fixes contributed by quantum-like in #74. The Kalshi and Polymarket UTC fixes from that contribution are already supplied by #76.

The regressions failed against the original base and passed with the fixes. The replacement retains the original contribution commits for review and passed the supported CI checks in #85.

Co-authored-by: quantum-like <[email protected]>

Addresses #84 with the documented free annual-history restriction and point-in-time caveat.

Co-authored-by: Stefan Jansen <[email protected]>