Add T0.95 late-order heat stage

Request:
- Reflect that near-final market heat can be used when the user can still place an IPO order at T0.95.

Changes:
- Added T0_95_final_heat as a separate analyst decision stage with executability and no-leakage rules.
- Added an experimental T0.95 rule overlay for late-order heat scoring and calibration discipline.
- Updated archivist guidance and the market-heat archiver so snapshots can be explicitly stored as T0_95_final_heat.
- Added market_heat_stage to the analysis dataset and refreshed the model report to show T0.95 coverage separately.

Verification:
- Ran py_compile for the modified scripts.
- Checked archive_t0_5_market_heat.py --help for the new --stage option.
- Rebuilt data/snapshots/analysis_model_v0_dataset.csv and reports/2026-06-15_analysis_model_v0.md.
- Ran git diff --check.

Next useful context:
- Current archived heat rows remain T0_5_market_heat only; there are no true T0.95 rows yet.
- external_ipo_history.public_oversubscription_times is still calibration-only unless a comparable value is archived before the executable order cutoff.

rules/ipo_score_v0_95_final_heat_trial.yaml

@@ -0,0 +1,73 @@
+version: ipo_score_v0_95_final_heat_trial
+effective_date: 2026-06-15
+status: experimental_overlay
+owner_skill: analyst
+base_model: ipo_score_v0
+
+purpose: >
+  Add a stage-safe T0.95 late-order overlay for Hong Kong IPO subscription
+  decisions. T0.95 is used only when the user can still place, amend, or
+  cancel an IPO order near the subscription cutoff and a reproducible
+  near-deadline market-heat snapshot has been archived.
+
+data_policy:
+  stage: T0_95_final_heat
+  required_condition: >
+    observed_at must be earlier than the user's actual executable order,
+    amend, or cancel cutoff. If this cannot be established, treat the data as
+    post-hoc research rather than as a live T0.95 input.
+  allowed_inputs:
+    - archived_near_deadline_margin_subscription_multiple
+    - provider
+    - observed_at
+    - subscription_deadline
+    - user_order_cutoff_if_known
+    - T0_prospectus_score
+  calibration_only_inputs:
+    - external_ipo_history.public_oversubscription_times
+    - external_ipo_history.one_hand_win_rate_pct
+    - external_ipo_history.grey_market_return_pct
+    - external_ipo_history.first_day_return_pct
+  forbidden_live_inputs:
+    - official_allotment_results_released_after_order_cutoff
+    - public_oversubscription_times_from_T1
+    - international_oversubscription_times_from_T1
+    - grey_market_result
+    - post_listing_price_performance
+  caveat: >
+    Historical final public oversubscription is close to what T0.95 tries to
+    approximate, so it may be used for post-hoc bucket calibration. It must not
+    be represented as data that was visible before the original order cutoff.
+
+score_components:
+  T0_95_final_heat:
+    margin_subscription_multiple:
+      source: archived_near_deadline_market_data_snapshot
+      points:
+        missing: 0
+        lt_10x: -4
+        gte_10_lt_100x: -2
+        gte_100_lt_1000x: 2
+        gte_1000_lt_5000x: 6
+        gte_5000x: 8
+    timing_quality:
+      source: observed_at_vs_order_cutoff
+      points:
+        observed_after_order_cutoff: disallowed
+        older_than_24h_before_cutoff: -2
+        same_day_before_cutoff: 0
+        within_3h_before_cutoff: 1
+
+decision_bands:
+  T0_plus_T0_95:
+    score_lt_5: no_late_order
+    score_5_to_8: optional_small_order
+    score_9_to_13: selective_late_order
+    score_gte_14: priority_late_order
+
+comparison_policy:
+  - Compare T0.95 snapshots only when each snapshot is still actionable for the user's order workflow.
+  - Prefer the latest archived snapshot before the executable order cutoff.
+  - If two IPOs have different close times, normalize by time remaining before cutoff rather than by calendar date alone.
+  - Always discuss allocation probability; very high heat may raise D1 payoff odds while lowering one-lot win rate.
+  - T0.95 can upgrade a T0 watchlist item into a late order candidate, but should not rescue a structurally weak T0 case without explicit PM judgment.

rules/rule_change_log.md

@@ -1,5 +1,28 @@
 # Rule Change Log
 
+## 2026-06-15 - Add T0.95 late-order heat stage
+
+Request:
+
+- Treat near-deadline heat as usable when the user can still place an IPO order at T0.95.
+
+Change:
+
+- Added `T0_95_final_heat` to the analyst skill as a separate actionable late-order stage.
+- Added `rules/ipo_score_v0_95_final_heat_trial.yaml` for stage-safe T0.95 heat scoring and timing discipline.
+- Updated the archivist skill and `scripts/archive_t0_5_market_heat.py` so market-heat snapshots can be explicitly archived as `T0_95_final_heat`.
+
+Rationale:
+
+- If the user can still place, amend, or cancel an order near the subscription cutoff, near-final heat is a legitimate live decision input.
+- Historical final public oversubscription can help calibrate near-final heat buckets, but it remains post-hoc calibration unless the value was visible before the user's executable order cutoff.
+
+Verification:
+
+- Kept the script default stage as `T0_5_market_heat` for backward compatibility.
+- Added explicit `--stage T0_95_final_heat` support for late-order snapshots.
+- Verified with py_compile, script help output, dataset rebuild, and `git diff --check`.
+
 ## 2026-06-15 - Use Chinese for analyst reports
 
 Request: