Report-only native intraday terminal quote

Summary

Use the explicitly declared native chart terminal quote through the existing
report hook, without feeding another bar into execution. The inclusive chart
range end can be one chart interval after the last completed magnified bucket.
No kernel, settlement implementation, runtime C export or codegen change is
part of this PR. The branch is rebased onto 288f189c; the final head is
7ec2f15f0a6bed976a2bb5f3d6117733b6cb6925.

Derived report metrics recomputed from the corrected terminal mark are presentation, provided trades, fills, book and state remain byte-identical.

• The report-only declaration pins quote time, close and native-feed SHA-256
  in runtime provenance; ON/OFF and any changed component have different
  fingerprints, while undeclared runs retain the old fingerprint shape.
• Apply only on the last mark's chart grid: the completed mark or exactly one
  chart interval later. Stale, off-grid, future, daily, nonmagnified and
  incomplete runs are inert. An applied/inert observable and harness detail
  distinguish application from declaration.
• Re-sum displayed trade rows for report.net_profit and last equity, matching
  the metrics calculator's accumulation order without a one-ulp discrepancy.
• Repeated report reads are pure and idempotent; stored report rows, fills,
  position, pending orders and broker-state hashes do not change. No new
  !stream_mode_ gate is introduced.

Review Field Table

The projected row deliberately retains its old max_runup, max_drawdown
and exit_bar_index; it is not a synthetic execution bar. Buy-and-hold still
uses the last executed close. The last equity point is relabelled, not
appended, so the preceding gap can span two chart intervals. At a month
boundary the corrected point enters a one-point trailing month for monthly
Sharpe/Sortino, and the preceding point supplies the closing month's final
sample. The trade-only parity grader does not validate these derived metrics
against TradingView. The harness line infers APPLIED from the terminal
timestamps, which is exact for a quote one interval after the mark but not for
a quote stamped at the mark itself. See docs/report-terminal-quote.md.

Review

An independent review returned merge-with-fixes; every fix was applied. A
focused re-review of the fix round returned merge, with no blocking finding.
It re-proved execution identity on real probes with per-bar hash recording on:
the NQ 15-minute range-end probe has 63,086 of 63,086 per-bar broker-state
hashes identical with the quote off and on, and an ETH magnifier probe
37,999 of 37,999. Trade rows and the pending-order book are identical in both.
The rebase onto 288f189c conflicted only in documentation citation digits;
both code commits are patch-identical.

Gate

• Full-population measurement of head 7ec2f15 with the current codegen:
  7,989 probes, no grade change, none missing, none erroring.
• Gate verdict: PASS (no improvement, no regression), verdict schema v3.
  Hard lane (1,009 probes): 0 moves, tier-rank sum 4027 → 4027. Target lanes
  (6,980 probes): 0 up, 0 down.
• This is the expected result: the quote is inert until the companion
  workflow PR's input pin is applied after its declared-case gate. The paired
  measurement with the quote declared (below) is where NQ and EURUSD lift.

Validation

• Focused final-head quote test: 1476 checks, zero failures; four related
  CTest rows pass. Magnifier/range-end Python suites: 23 + 38 pass.
• Runtime fingerprint self-test: 291 pass; documentation anchors: 1744/1744.
• Refreshed measurement uses the rebased engine and checked-settings codegen:
  19 probes/87 pairs plus an independent two-probe/19-pair OFF ladder. All
  actual trade-row and pending-order-book/live-position digests are identical.
  Only NQ and EURUSD in the ruled range-end pair lift strong to excellent.
• The five probes without per-bar hash tapes have explicit raw actual-row
  and book/position identity witnesses; empty hash arrays are not evidence.
  Unit tests also compare complete per-bar and final state hashes.
• Full C report comparisons normalize the terminal presentation fields and
  derived metric blocks, serializing floats as binary64 bits; the unit test
  validates the metrics against the existing pure report calculators.
• Full verification profiles on the exact final head (x86-64 Linux, GCC 13):
  preflight 44 stages; release 807 CTest rows; sanitizers 788; debug 788;
  kernel 303 with zero residual findings; documentation with zero Doxygen
  warnings. Full corpus trade identity is 312/312 and the 54-probe subset
  54/54, with zero moved, unrecorded, missing or schema-violating rows. The
  macOS build and native-live profiles do not run on these hosts.

Apply Gate

Merge does not enable the input. The companion workflow PR
(pineforge-4pass/pineforge-workflow#93) must first pass
one declared case end to end on its candidate immutable runner image before
the input pin is applied. Its receipt, APPLIED detail, new runtime quote
fingerprint and fresh OFF control must agree. Disable probe reuse and verifier
cache seeding, use fresh science/output keys, and reject an OFF cached result.
Only the supervisor applies compatible image/input pins in a paused window
and measures the reference pair under the new science identity. No production
migration, deployment, push or gate recording is performed by this PR's lane.

🤖 Generated with Claude Code