Session 18
Session 18 (2026-02-19)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_6.json.gzCapture family: post-master-rebase quest split set (session18_quest_gz_post_master_rebase)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_6.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix: tick 5384 (rng_stream_mismatch) with downstream one-tick projectile hit timing slip at 5386/5387
after this fix: ok (no divergence found with current settings)
Key Findings
The 5384 RNG drift was caused by rewrite-only RNG work on tick 5383, not by the 5384 fire path itself:
focus trace before fix: tick 5383 capture/rewrite calls = 100/102 with full value-prefix match,
rewrite consumed two extra values (14154, 19675), shifting all tick 5384 weapon-fire rolls.
The extra draws came from world-level death-SFX planning for plague-timer kills:
plague kill path in creatures.runtime already consumes native contact-SFX RNG,
world_state.step additionally ran plan_death_sfx_keys for those same deaths, adding two rewrite-only draws.After skipping world death-SFX planning for plague deaths:
focus-trace realigned at both tick 5383 and tick 5384 (prefix_match = capture_calls),quest_1_6 baseline report went fully clean.
Landed Changes
src/crimson/creatures/runtime.pyadded CreatureDeath.plan_death_sfx: bool = True,
threaded plan_death_sfx through handle_death,
set plan_death_sfx=False for plague timer kills.
src/crimson/sim/world_state.py_plan_death_sfx_now(...) now skips deaths with plan_death_sfx=False.Tests:
tests/test_plaguebearer_perk.py now asserts plague kills mark plan_death_sfx=False,tests/test_death_timing.py adds test_plague_kill_death_event_skips_world_death_sfx_planning.
Validation
Targeted tests:
uv run pytest tests/test_plaguebearer_perk.py::test_plaguebearer_infection_kill_increments_global_count tests/test_death_timing.py::test_plague_kill_death_event_skips_world_death_sfx_planning tests/test_death_timing.py::test_death_sfx_rand_consumes_past_cap tests/test_death_timing.py::test_detonation_followup_does_not_double_plan_death_sfxFocus traces:
uv run crimson original focus-trace artifacts/frida/share/gameplay_diff_capture.quest_1_6.json.gz --tick 5383 --near-miss-threshold 0.35 --no-cache --json-out analysis/frida/reports/session18_quest_gz_post_master_rebase/gameplay_diff_capture.quest_1_6_focus_5383_after_plague_sfx_fix_nocache.jsonuv run crimson original focus-trace artifacts/frida/share/gameplay_diff_capture.quest_1_6.json.gz --tick 5384 --near-miss-threshold 0.35 --no-cache --json-out analysis/frida/reports/session18_quest_gz_post_master_rebase/gameplay_diff_capture.quest_1_6_focus_5384_after_plague_sfx_fix_nocache.jsonBaseline recheck:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_6.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache → result=ok (no divergence found with current settings)Repository checks:
just check
Outcome / Next Probe
quest_1_6 in this post-rebase capture family is clean again.Continue the remaining quest split frontier from the earliest unresolved file/tick in the current sweep.
Session 18 (continued: strict projectile radius boundary)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gzCapture family: post-master-rebase quest split set (session18_quest_gz_followup)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix: tick 7655 (players[0].experience, score_xp, creature-count drift)
after this fix: tick 7722 (players[0].experience, score_xp, creature-count drift)
Key Findings
Focus tracing at the previous frontier showed a rewrite-only projectile hit accepted with positive margin (+0.000928) by _within_native_find_radius(...).
The acceptance was caused by a local positive epsilon (0.001) that allowed near-edge non-native hits.
That rewrite-only hit/death path created an immediate XP/score overcount and pushed subsequent RNG/collision timing off the native track.
Landed Changes
src/crimson/projectiles/runtime/collision.pyset _NATIVE_FIND_RADIUS_MARGIN_EPS to 0.0,
kept strict native boundary semantics for hit acceptance.
tests/test_projectiles.pyadded test_within_native_find_radius_uses_strict_boundary to lock the regression case where a tiny positive over-margin must not count as a hit.
Validation
Targeted tests:
uv run pytest tests/test_projectiles.pyRepository checks:
just checkDivergence rechecks:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache
result=diverged kind=state_mismatch tick=7722
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_3_2.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache
result=ok (no divergence found with current settings)
Outcome / Next Probe
This fix cleared quest_3_2 and moved the earliest unresolved frontier later in quest_1_8 (7655 -> 7722).
Next probe should focus quest_1_8 at tick 7722 and isolate the first rewrite-only kill/death branch now driving creature-count and XP drift.
Session 18 (continued: Fire Cough pre-move spawn origin parity)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_9.json.gzCapture family: post-master-rebase quest split set (session18_quest_gz_after_unlink_sweep)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_9.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix: tick 9259 (rng.value_stream_mismatch)
after this fix: tick 9482 (rng.value_stream_mismatch)
Key Findings
quest_1_9 showed a rewrite-only RNG burst at tick 9257 (+68 draws) with one extra projectile hit (capture=6, rewrite=7), then first value-stream mismatch at tick 9259.Focus traces localized the first projectile trajectory split to projectile idx=2 at tick 9254, with rewrite projectile position offset by about (+1.923, +1.923).
That offset exactly matched the player movement delta in the same tick, proving rewrite spawned the Fire Cough projectile from post-move player position while native capture spawned from pre-move position.
Landed Changes
src/crimson/perks/impl/fire_cough.pyswitched Fire Cough projectile origin math to ctx.player_pos_before_move for:
muzzle position,
spread-distance basis,
jitter-to-angle vector.
tests/test_player_update.pyadded test_player_update_fire_cough_uses_pre_move_position_for_spawn regression to lock moving-frame pre-move spawn semantics.
Validation
Targeted tests:
uv run pytest tests/test_player_update.py -k "fire_cough"Divergence recheck:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_9.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads/quest_1_9_after_fire_cough_pre_move_fix.json
result=diverged kind=rng_stream_mismatch tick=9482 (frontier moved later from 9259)
Spot checks on unrelated fronts remained stable:
quest_1_10 remains at tick 10018 (rng.projectile_hit_resolution_shortfall)quest_2_1 remains at tick 11417 (rng.value_stream_mismatch)quest_4_5 remains at tick 46214 (rng.value_stream_mismatch)
Outcome / Next Probe
Fire Cough pre-move spawn parity removed the early quest_1_9 branch split and pushed the first sustained mismatch significantly later (9259 -> 9482).
Next probe should focus quest_1_9 at tick 9482, where first mismatch now maps to native player_update RNG callers (0x00415a6d cluster), to identify the next branch-order/source-of-draw divergence.
Session 18 (continued: ranged-shock lethal death branch parity)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gzCapture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix set: tick 10018 (rng.projectile_hit_resolution_shortfall, missing tail 20)
after lethal burst branch only: tick 10018 (state mismatch with RNG overrun +1)
after death-SFX suppression for this path: tick 10018 (state mismatch; focus RNG stream aligned 328/328)
Key Findings
Native creature_apply_damage has a lethal branch for flags & 0x10 that consumes 20 RNG draws (5 iterations x 4 draws) and spawns burst effects instead of random death-SFX selection.
Rewrite was missing that lethal burst branch, then after adding it still consumed one extra RNG draw due world-level death-SFX planning on the same death.
Suppressing death-SFX planning for lethal kills of RANGED_ATTACK_SHOCK creatures removed the extra RNG draw and restored focus-tick RNG alignment.
Landed Changes
src/crimson/creatures/damage.pyadded lethal RANGED_ATTACK_SHOCK burst path to mirror native RNG/effect semantics,
threaded optional effect context through damage helpers.
src/crimson/sim/world_state.pypass effect context into projectile damage apply path,
suppress death-SFX planning on lethal RANGED_ATTACK_SHOCK damage follow-up,
forward plan_death_sfx into handle_death(...).
src/crimson/creatures/runtime.pypass effect context into lethal-followup callsites,
suppress death-SFX planning for lethal RANGED_ATTACK_SHOCK callback deaths.
src/crimson/perks/impl/final_revenge.pypass effect context into lethal-followup callsite,
suppress death-SFX planning for lethal RANGED_ATTACK_SHOCK callback deaths.
tests/test_death_timing.pyadded regression test_ranged_shock_lethal_skips_world_death_sfx_planning.
Validation
Targeted checks:
uv run pytest tests/test_creature_damage.py tests/test_creature_damage_callsite_guardrails.py tests/test_projectile_hit_effects.py tests/test_death_timing.py -qRepository checks:
just checkDivergence rechecks:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads/quest_1_10_after_ranged_shock_lethal_burst_fix_nocache.json
result=diverged kind=state_mismatch tick=10018 with focus RNG overrun +1
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads/quest_1_10_after_ranged_shock_death_sfx_suppression_nocache.json
result=diverged kind=state_mismatch tick=10018 with focus RNG stream fully aligned (328/328, missing_tail=0)
Outcome / Next Probe
The quest_1_10 focus RNG shortfall at tick 10018 is resolved; the remaining mismatch at that tick is now state-only (perk.pending_count, players[0].level) with aligned RNG stream.
Next probe should move earlier in this capture to the still-reported projectile-hit-resolution lead (tick 9757) and isolate the missing rewrite hit-resolve path.
Session 18 (continued: quest_1_10 progression pacing and idle-complete reflex parity)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gzCapture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix set: tick 10018 (state mismatch: perk.pending_count, players[0].level)
after single-step survival level-up pacing: tick 10477 (bonus_timers.9: expected 0, actual 210)
after quest idle-complete active-slot parity fix: result=ok (no divergence found with current settings)
Key Findings
Native survival progression advances at most one level threshold per update tick; rewrite looped multiple thresholds in one tick, over-incrementing quest perk pending state.
Native creatures_none_active checks creature slot active flags, not hp > 0; corpses remain "active" until lifecycle finalization.
Quest-mode idle-complete reflex clear (bonus_reflex_boost_timer = 0) is correct only when native-style active-slot emptiness and empty spawn table are both true.
Landed Changes
src/crimson/gameplay.pychanged survival_check_level_up to advance at most one threshold per tick.
tests/test_gameplay_progression.pyadded test_survival_level_up_advances_one_threshold_per_tick.
src/crimson/sim/sessions.pyswitched quest creatures_none_active computation to native active-slot semantics (creature.active),
preserved quest idle-complete reflex clear and completion-transition gating using the same spawn_table_empty_now evaluation.
tests/test_quest_deterministic_session.pyadded test_quest_session_clears_reflex_boost_when_quest_is_idle_complete.
Validation
Targeted checks:
uv run pytest tests/test_quest_deterministic_session.py tests/test_reflex_boosted_perk.py tests/test_gameplay_progression.py -qDivergence recheck:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_10.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads/quest_1_10_after_idle_complete_reflex_clear_fix_nocache.json
result=ok (no divergence found with current settings)
Repository checks:
just check
Outcome / Next Probe
quest_1_10 is now cleared and can be unlinked from the active divergence set.Remaining unresolved capture fronts should continue from currently outstanding quest files (quest_1_9, quest_2_1, quest_4_5 in this family).
Session 18 (continued: quest_1_9 Fire Cough spread cooldown ordering parity)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_9.json.gz (now unlinked after verification) Capture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command (pre-unlink):
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_9.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix: tick 9482 (rng_stream_mismatch)
after this fix: result=ok (no divergence found with current settings) (pre-unlink no-cache run)
Key Findings
Fire Cough samples player.spread_heat inside perk tick execution.
Rewrite was decaying spread heat before apply_player_perk_ticks(...); native ordering cools spread after perk ticks/movement while still before weapon fire.
On the trigger tick in this capture, the sampled spread differed by exactly dt * 0.4, which shifted Fire Cough jitter angle enough to flip downstream projectile hit acceptance and RNG tail alignment.
Landed Changes
src/crimson/gameplay.pymoved the spread-heat cooldown block to after perk ticks/aim update and before fire-gate + _player_fire_weapon(...):
Sharpshooter: clamp to 0.02,otherwise: max(0.01, spread_heat - dt * 0.4).
Validation
Repository checks:
just checkDivergence checks:
quest_1_9 (pre-unlink no-cache): result=ok.quest_1_8 remains diverged at tick 7722 (see analysis/frida/reports/session18_alt_leads/quest_1_8_after_spread_decay_reorder_nocache.json).quest_2_1 remains diverged at tick 11417 (see analysis/frida/reports/session18_alt_leads/quest_2_1_after_spread_decay_reorder_nocache.json).
Outcome / Next Probe
quest_1_9 is cleared and unlinked from artifacts/frida/share/.Earliest unresolved frontier in this set remains quest_1_8 at tick 7722; continue by isolating the first RNG/value stream fork there.
Session 18 (continued: angle_approach decompile-order parity)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gzCapture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix: tick 7722 (state_mismatch, progression-timing drift)
after this fix: tick 7756 (rng_stream_mismatch, projectile-hit-resolution shortfall)
Key Findings
Native angle_approach (0x0041f430) updates the wrapped heading arc in decompile order and does not enforce extra float32 truncation on intermediate turn-delta arithmetic.
The rewrite helper had over-quantized turn-step intermediates; this amplified slot-level heading drift before the frontier window and flipped kill timing.
Moving _angle_approach(...) to the decompile-faithful arithmetic order moved two active frontiers later with no earlier regressions in the linked no-cache quest sweep:
quest_1_8: 7722 -> 7756 (+34 ticks),quest_4_5: 46214 -> 46308 (+94 ticks; category moved from rng.value_stream_mismatch to rng.tail_shortfall).
Landed Changes
src/crimson/creatures/runtime.pyrewrote _angle_approach(...) to mirror native wrap/direct-vs-wrapped arc selection and step-direction logic in decompile order.
Validation
Targeted no-cache rechecks:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads_recheck/quest_1_8_nocache_with_angle_probe_v3.jsonuv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_2_1.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads_recheck/quest_2_1_nocache_with_angle_probe_v3.jsonFull linked-capture no-cache canary sweep summary:
analysis/frida/reports/session18_alt_leads_recheck_angle_v3_nocache/_summary.tsvRepository checks:
just check
Outcome / Next Probe
No linked captures are clean in this pass, so there are no new capture unlinks.
Earliest unresolved frontier moved to quest_1_8 tick 7756; next probe remains projectile-hit/kill-resolution parity at the first missing native tail branch in that window.
Session 18 (continued: man_bomb bootstrap interval inference false-wrap fix)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_2_1.json.gz (now unlinked after verification) Capture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command (pre-fix):
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_2_1.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this fix: tick 11417 (rng_stream_mismatch)
after this fix: result=ok (no divergence found with current settings) (no-cache run)
Key Findings
Replay bootstrap interval inference treated the first observed active man_bomb timer drop as a trigger wrap.
In this capture, the first drop (11380 -> 11381, 1.225 -> 0.0) was a movement-gated reset, not a trigger wrap; inferred interval became ~1.301s.
With that wrong interval, rewrite incorrectly triggered Man Bomb at tick 11416 (8 extra RNG calls/projectiles), creating the tick 11417 RNG stream fork.
True man-bomb trigger wraps later in the same capture infer ~4.0s, consistent with native post-trigger behavior.
Landed Changes
src/crimson/original/capture.pystopped inferring bootstrap man_bomb interval from timer drops in _infer_bootstrap_perk_intervals(...);
kept inference for hot_tempered and fire_cough.
tests/test_original_capture_conversion.pyadded test_convert_capture_to_replay_bootstrap_payload_does_not_infer_man_bomb_from_timer_drop.
Validation
Targeted tests:
uv run pytest tests/test_original_capture_conversion.py -k "bootstrap_payload_infer or man_bomb_from_timer_drop" -qDivergence recheck:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_2_1.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_alt_leads_continue_pass/quest_2_1_after_bootstrap_interval_fix_nocache.json
result=ok (no divergence found with current settings)
Outcome / Next Probe
quest_2_1 is cleared and unlinked from artifacts/frida/share/.Remaining earliest unresolved frontier in the active set remains quest_1_8 tick 7756.
Session 18 (continued: projectile-hit shortfall owner-collision normalization)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gzCapture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: before this tooling fix: tick 7756 (rng_stream_mismatch, categorized as rng.projectile_hit_resolution_shortfall via pre-focus lead at tick 7753)
after this tooling fix: tick 7756 (rng_stream_mismatch, categorized as rng.tail_shortfall; projectile-hit shortfall lead removed)
Key Findings
In this capture family, projectile_find_hit_count can include owner-collision query rows; at the old lead tick (7753), capture reported hit_count=3 with owner_collision=2 while rewrite reported actual_hits=1, so effective capture hits matched rewrite hits.
Using raw projectile_find_hit_count for lead/category detection produced false projectile-hit shortfall signatures and misdirected follow-up probes.
Landed Changes
src/crimson/original/divergence_report.pyadded _effective_capture_projectile_hit_count(...) helper,
updated _find_first_projectile_hit_shortfall(...) to compare rewrite hit count against effective capture hits (raw_hits - owner_collision_queries, clamped at >= 0),
updated _classify_divergence_category(...) to use the same effective-hit metric,
updated focus diagnostics printout to show both raw and effective capture projectile-hit counts.
tests/test_original_capture_divergence_report_rng_calls.pyadded owner-collision guardrail tests for shortfall detection and category classification,
updated existing projectile-hit shortfall fixtures to assert both raw and effective hit semantics.
Validation
Targeted tests:
uv run pytest tests/test_original_capture_divergence_report_rng_calls.py -qFocus recheck:
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cache --json-out analysis/frida/reports/session18_unlink_continue_pass6/quest_1_8_after_owner_collision_shortfall_fix_nocache.json (expected non-zero exit while diverged)
result=diverged kind=rng_stream_mismatch tick=7756 focus_tick=7756divergence_category.id = rng.tail_shortfallno Native projectile hit resolves exceed rewrite hit events lead in investigation_leads
Full linked-capture no-cache sweep summary:
analysis/frida/reports/session18_unlink_continue_pass6/_summary.tsv (21 total, 0 ok, 21 diverged)Repository checks:
just check
Outcome / Next Probe
No linked captures are clear in this pass, so there are no new capture unlinks.
Earliest unresolved frontier remains quest_1_8 tick 7756; next probe should continue from movement-drift ancestry and the remaining RNG tail shortfall branch in the 7753-7756 window.
Session 18 (continued: quest_1_8 follow-link threshold dead-end probe pack)
Capture: artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gzCapture family: post-master-rebase quest split set (session18_alt_leads)Baseline verifier command (start/end of pass):
uv run crimson original divergence-report artifacts/frida/share/gameplay_diff_capture.quest_1_8.json.gz --float-abs-tol 1e-3 --window 24 --lead-lookback 1024 --run-summary-short --run-summary-focus-context --run-summary-focus-before 8 --run-summary-focus-after 4 --run-summary-short-max-rows 30 --no-cacheFirst mismatch progression: pass start baseline: tick 7756 (rng_stream_mismatch, rng.tail_shortfall, missing tail 2)
pass end baseline: unchanged at tick 7756 (same category/signature)
Key Findings
The first decisive branch split in this ancestry is earlier than focus:
at tick 7647, capture keeps slot 3 on player-force (force_target(after)=1), while rewrite flips to link-target (force_target(after)=0).
Extracted raw creature_update_micro rows from capture (event_heads) and compared against rewrite trajectory at the same window.
In the edge window (7638..7650), rewrite link-distance-before is consistently biased relative to capture:
rewrite minus capture bias stays around +0.126 .. +0.133 through tick 7647,
crossing behavior at tick 7647 maps directly to the < 40.0 force-target threshold branch.
This indicates cumulative upstream movement ancestry drift; no local constant tweak below isolated the root cause without collateral regressions.
Dead-End Probes (all reverted)
src/crimson/creatures/runtime.py: _angle_approach(...) float-store-boundary variant.result: no 7647 force-target correction; slot drift worsened on control path.
src/crimson/creatures/ai.py: target-heading delta path without pre-rounding dx/dy to f32.result: no change in first force-target mismatch tick (7647 remained).
src/crimson/creatures/runtime.py: turn-rate expression variant (move_speed * 0.33333334 * 4.0 form).result: no material change in ancestry/frontier.
src/crimson/math_parity.py: heading_from_delta_f32 using +1.5707964 literal.result: slight drift changes but no branch/focus improvement.
src/crimson/math_parity.py: disable left-axis tie-break branch.result: no measurable impact in this window.
src/crimson/creatures/runtime.py: movement direction from heading_to_direction_f32(...) pre-rounded components.result: no measurable impact in this window.
src/crimson/creatures/runtime.py: movement radians with -1.5707964 literal.result: severe run-level regression (run_score_xp=9833, extra RNG-state-change ticks at 7759/7763/7767) while still diverged at 7756.
Validation
Baseline no-cache (start): analysis/frida/reports/session18_continue/quest_1_8_baseline_resume_nocache.json
Baseline no-cache (end): analysis/frida/reports/session18_continue/quest_1_8_baseline_end_of_pass_nocache.json
Regression canary (movement half-pi literal probe): analysis/frida/reports/session18_continue/quest_1_8_after_move_halfpi_constant_probe_nocache.json
Edge-window evidence artifact:
analysis/frida/reports/session18_continue/quest_1_8_slot3_force_target_edge_7638_7650.tsvSupplementary trajectory probes generated in this pass:
analysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_angle_store_probe.jsonanalysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_target_heading_probe.jsonanalysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_turnrate_probe.jsonanalysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_halfpi_probe.jsonanalysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_no_left_axis_probe.jsonanalysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_f32_direction_probe.jsonanalysis/frida/reports/session18_continue/_tmp_cre3_7506_7708_move_halfpi_probe.json
Outcome / Next Probe
No gameplay/runtime patch landed from this pass; all probe edits were reverted.
To avoid repeating dead-end arithmetic tweaks, next run should switch to a different axis:
verify replay tick-timing application in the 7638..7650 window (dt_frame, dt_frame_ms_i32, and apply_world_dt_steps behavior) against capture micro dt_frame rows,
instrument rewrite world-step per-tick/per-slot (slot 0 and slot 3) around pre-AI and post-move boundaries to confirm whether the persistent ~+0.13 link-distance bias is timing-path driven rather than local movement constants.
frida
differential-sessions