docs: release notes and version scoping for v1.2.0 - #340

What and why

The release workflow writes only VERSION; this docs-only change gives v1.2.0 its CHANGELOG section and scopes the
version-bearing pages to the 1.2.0 pair (engine v1.2.0 with pineforge-codegen 1.2.0, hub image
ghcr.io/pineforge-4pass/pineforge-release:1.2.0).

1.2.0 is a minor release: <pineforge/pineforge.h> gains the compiled execution-capabilities extension
(PF_CAPABILITIES_API_VERSION 1, strategy_capabilities_api_version, strategy_capabilities_receipt; #332), which a
strategy generated by codegen 1.2.0 defines. PF_ABI_VERSION stays 4, the script ABI epoch stays
engine_script_run_v19, the domain tags pineforge-broker-state/v19 and pineforge-source-adapter/v4 are unchanged,
and native_c_api.h, the native C++ host headers and the frozen ABI fixtures are unchanged since v1.1.0.

• CHANGELOG.md: ## Unreleased becomes ## 1.2.0 — 2026-10-05: the lead paragraph, one bullet per change since
  v1.1.0 with its PR (Pine adapter: TradingView's exit-child order at one point and the callback lifecycle (trail tolerance, revived exits, tick-built limit reach) #328, Pine adapter: TradingView's margin, sizing and admission rules for one-position books #330, Live runner: service operation, cumulative report export and live redelivery #331, Live runner: refuse compiled strategies a close-only stream cannot honour (capability receipt v1) #332, Make the native-live routing E2E deterministic (no wall-clock races) #333, Pine adapter: explicit entry quantity floor, script equity mark, process-orders-on-close bar fills and the session clock #334, Pine adapter: admission fill half, close-first entries, point order, exit tombstones and pyramiding records #335, Pine adapter: recover the per-bar cost of the margin, sizing and admission rules #336, docs: render the active scoreboard from the facts tokens at pineforge-release 03b8dcc #337, Native stream: fewer temporary strings and lookups on the confirmed-bar path #338, and the session-clock change,
  Session clock: read each day's occurrences once per bar; time() arguments resolved without copies #339), State hashes, Report keys and Migration. The two Unreleased runner
  sections (Live runner: refuse compiled strategies a close-only stream cannot honour (capability receipt v1) #332, Live runner: service operation, cumulative report export and live redelivery #331) are kept, shortened and given PR links; the runner's graceful signal exit (130 → 0), the
  report replay for older ledgers and the legacy-library warning are kept as stated in runner/README.md.
• Version scoping: README (the codegen install line, the prebuilt v1.2.0 tarballs and release link, the pairing
  sentence, "its releases 1.0.0, 1.0.1, 1.1.0 and 1.2.0", the benchmark refresh line, the "This is the runtime"
  paragraph, a v1.2.0 line in Releases), docs/pages/install.md (the v1.2.0 tarballs, hub tag
  engine1.2.0-codegen1.2.0), docs/pages/public-contract.md (the 1.2.0 release line, engine v1.2.0 with codegen
  1.2.0 among the supported pairs, the four C-SURFACE-2 rows not in 1.2.0 either), and Live runner: refuse compiled strategies a close-only stream cannot honour (capability receipt v1) #332's "since the next
  release" / "fixed on main" in docs/strategy-capabilities.md and runner/README.md (now 1.2.0, and 1.1.0 (Confirmed-bar streams compute what the batch computes #325)
  for the stream fix). Lines that state what an earlier release shipped stay.
• Facts: the README's release sentence names 1.2.0 and renders releases[1.2.0] (registry baseline
  pineforge-parity-baseline-20261005-engine-52292db9, Session clock: read each day's occurrences once per bar; time() arguments resolved without copies #339's merge, which v1.2.0 equals in behaviour: 7,970 excellent
  / 19 strong of 7,989, none below strong), and the main scoreboard renders the same baseline, from pineforge-release
  facts/facts.json at 0c2f6838 (pineforge-release Production-readiness probes + engine state-reset / finite-fill fixes #28). lab facts check against that pinned file passes, with
  --github-render.
• License: README, LEGAL.md and docs/pages/index.md said pineforge-codegen is PolyForm Noncommercial; from its
  release 1.2.0 it is under the PineForge Source License 1.1 (pineforge-codegen ci: add a build gate check the branch ruleset can match #168, fix: key calendar aux routing by the covered session #171), and releases up to 1.1.0
  keep the terms they shipped with. The three lines now say so and point to its LICENSE.

Report keys

No report key changes. docker/run_json.py is byte-identical between v1.1.0 and the release's code
(git diff --quiet v1.1.0 52292db9 -- docker/run_json.py), so no key is added, removed or renamed and
EQUITY_REPORT_KEYS (metrics.equity.sharpe_tv, sortino_tv) stays; pineforge.h's diff since v1.1.0 is the
trail-rounding comment plus the capability declarations, with pf_report_t and pf_trade_t unchanged. The runner's
new outputs (#331: pineforge-live report documents, pineforge-native-report-delta/v1 ledger rows, the
pineforge-live-status/v1 status file, pineforge-redelivery-request/v1 requests) are versioned runner formats, not
run_json keys.

State hashes

Every recipe change since v1.1.0 is an addition, all in the Pine adapter's part of the broker-state hash
(src/source/pine_state_hash.cpp, +64/-0; src/engine_state_hash.cpp and src/engine_stream.cpp unchanged):

• Pine adapter: TradingView's exit-child order at one point and the callback lifecycle (trail tolerance, revived exits, tick-built limit reach) #328: TradingView's entry-id table (key count, running digest, two peaks), folded once it holds two ids or a peak
  passed 13 keys.
• Pine adapter: TradingView's margin, sizing and admission rules for one-position books #330: the G + L realized-money sums with the closed-trade count, folded only when their sum differs bit for bit from
  the engine's net profit.
• Pine adapter: admission fill half, close-first entries, point order, exit tombstones and pyramiding records #335: two placement flags, the pyramiding records, the exit tombstones and the close-first fact, each only while set.
• Native stream: fewer temporary strings and lookups on the confirmed-bar path #338 adds a WordSink::s(const char*) overload with the same encoding as the std::string one (no recipe change).

Nothing is removed, renamed or re-encoded; domain tags and epoch unchanged. scripts/check_aggregate_cpp_versions.py
("aggregate v19 ownership verified") and scripts/check_broker_state_hash_coverage.py (OK) pass at v1.1.0, 44eab7b
and #339's tree (52292db).

Verified

• Tutorial numbers re-measured on a disposable cloud box (terminated) at 44eab7b with GCC 13.3: bash tutorial/run.sh
  prints 50 trades (17W / 33L), +569.97, best/worst +1149.00 / -1111.97, max dd -4045.15; run_advanced.py prints the
  documented rows; regenerating tutorial/macd/strategy.pine with pineforge-codegen main (48e7a13b) gives the same
  numbers, and the README's front-door commands (install, transpile, compile, scripts/run_strategy.py) print 50
  trades, net 569.97. The documented numbers stand; the session-clock change is value-identical by its own evidence,
  so 44eab7b holds for 1.2.0.
• scripts/check_doc_anchors.py, scripts/check_doc_lint.py, scripts/check_doc_reverts.py --base origin/main and
  git diff --check origin/main pass on the final head. The merge gate's docs basis (preflight + docs) passed on this
  change before its rebase onto Session clock: read each day's occurrences once per bar; time() arguments resolved without copies #339 (it is re-run on this exact head for the statuses): ci_preflight: passed (exit 0, 44 stages, 0 advisory reported) and Doxygen warnings: 0.
• An independent read-only fact-check of every claim against the tree and the squash bodies; its corrections (scope of
  the decimal sizing floor and of the long residual call, runner-only "unchanged" statements, the re-issue after a
  declined reversal) are in.

🤖 Generated with Claude Code