qvm: complete bounded aim function
Browse files- docs/ASSET_PIPELINE_TODO.md +3 -0
- docs/FINDINGS.md +3 -0
- docs/PROVISIONAL_TODO.md +3 -0
- docs/RESEARCH_LEDGER.md +5 -0
- docs/REVERSE_ENGINEERING_TODO.md +3 -0
- docs/research/QVM_COMPLETE_AIM.md +7 -0
- docs/research/README.md +3 -0
- docs/research/qvm-complete-aim.json +0 -0
- scripts/analyze_qvm_complete_aim.py +42 -0
- scripts/test_qvm_complete_aim.py +18 -0
docs/ASSET_PIPELINE_TODO.md
CHANGED
|
@@ -532,3 +532,6 @@ Research progress: RE-616 [report](research/QVM_AIM_EXTRA_OFFSETS.md); remaining
|
|
| 532 |
|
| 533 |
|
| 534 |
Research progress: RE-617 [report](research/QVM_AIM_ENDPOINT.md); remaining native/content/runtime/platform scope stays open.
|
|
|
|
|
|
|
|
|
|
|
|
| 532 |
|
| 533 |
|
| 534 |
Research progress: RE-617 [report](research/QVM_AIM_ENDPOINT.md); remaining native/content/runtime/platform scope stays open.
|
| 535 |
+
|
| 536 |
+
|
| 537 |
+
Research progress: RE-618 [report](research/QVM_COMPLETE_AIM.md); remaining native/content/runtime/platform scope stays open.
|
docs/FINDINGS.md
CHANGED
|
@@ -887,3 +887,6 @@ Bounded qagame paths recover bomb-plant/inactivity correction requests, direct f
|
|
| 887 |
|
| 888 |
|
| 889 |
**RE-617 — Aim endpoint construction and selected alias behavior:** The endpoint adds origin+16384×forward, then horizontal×global right and vertical×global up in distinct f32 write phases. [Evidence](research/QVM_AIM_ENDPOINT.md).
|
|
|
|
|
|
|
|
|
|
|
|
| 887 |
|
| 888 |
|
| 889 |
**RE-617 — Aim endpoint construction and selected alias behavior:** The endpoint adds origin+16384×forward, then horizontal×global right and vertical×global up in distinct f32 write phases. [Evidence](research/QVM_AIM_ENDPOINT.md).
|
| 890 |
+
|
| 891 |
+
|
| 892 |
+
**RE-618 — Complete bounded shared aim function control flow:** A single bounded source interpreter now follows every branch and call through290493, including rejection loops, added jitter and final endpoint writes. [Evidence](research/QVM_COMPLETE_AIM.md).
|
docs/PROVISIONAL_TODO.md
CHANGED
|
@@ -852,3 +852,6 @@ Research progress: RE-616 [report](research/QVM_AIM_EXTRA_OFFSETS.md); remaining
|
|
| 852 |
|
| 853 |
|
| 854 |
Research progress: RE-617 [report](research/QVM_AIM_ENDPOINT.md); remaining native/content/runtime/platform scope stays open.
|
|
|
|
|
|
|
|
|
|
|
|
| 852 |
|
| 853 |
|
| 854 |
Research progress: RE-617 [report](research/QVM_AIM_ENDPOINT.md); remaining native/content/runtime/platform scope stays open.
|
| 855 |
+
|
| 856 |
+
|
| 857 |
+
Research progress: RE-618 [report](research/QVM_COMPLETE_AIM.md); remaining native/content/runtime/platform scope stays open.
|
docs/RESEARCH_LEDGER.md
CHANGED
|
@@ -767,3 +767,8 @@ IDs10/14/23 may add two RNG offsets; flag84 and the strict150-time boundary cont
|
|
| 767 |
## 2026-10-20 — RE-617 Aim endpoint construction and selected alias behavior (separate task commit)
|
| 768 |
|
| 769 |
The endpoint adds origin+16384×forward, then horizontal×global right and vertical×global up in distinct f32 write phases. 27 independent endpoint fixtures cover nonunit vectors and output aliases to origin/forward, retaining actual source effects. Full suite: 961 tests pass. Report regenerates byte-identically (417024 bytes). Storage/filter, links and whitespace checks pass. No assets/Rust runtime/shared VM evaluator changed; native/reference/legal/full platform scope remains open.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 767 |
## 2026-10-20 — RE-617 Aim endpoint construction and selected alias behavior (separate task commit)
|
| 768 |
|
| 769 |
The endpoint adds origin+16384×forward, then horizontal×global right and vertical×global up in distinct f32 write phases. 27 independent endpoint fixtures cover nonunit vectors and output aliases to origin/forward, retaining actual source effects. Full suite: 961 tests pass. Report regenerates byte-identically (417024 bytes). Storage/filter, links and whitespace checks pass. No assets/Rust runtime/shared VM evaluator changed; native/reference/legal/full platform scope remains open.
|
| 770 |
+
|
| 771 |
+
|
| 772 |
+
## 2026-10-20 — RE-618 Complete bounded shared aim function control flow (separate task commit)
|
| 773 |
+
|
| 774 |
+
A single bounded source interpreter now follows every branch and call through290493, including rejection loops, added jitter and final endpoint writes. 17 whole-function fixtures agree with independently checked staged calculations and endpoint formulas; zero seed stops at32 candidates with untouched output. Full suite: 965 tests pass. Report regenerates byte-identically (210344 bytes). Storage/filter, links and whitespace checks pass. No assets/Rust runtime/shared VM evaluator changed; native/reference/legal/full platform scope remains open.
|
docs/REVERSE_ENGINEERING_TODO.md
CHANGED
|
@@ -895,3 +895,6 @@ Purpose: turn the supplied installation into an evidence-backed specification of
|
|
| 895 |
|
| 896 |
|
| 897 |
- [x] RE-617 Aim endpoint construction and selected alias behavior ([report](research/QVM_AIM_ENDPOINT.md)).
|
|
|
|
|
|
|
|
|
|
|
|
| 895 |
|
| 896 |
|
| 897 |
- [x] RE-617 Aim endpoint construction and selected alias behavior ([report](research/QVM_AIM_ENDPOINT.md)).
|
| 898 |
+
|
| 899 |
+
|
| 900 |
+
- [x] RE-618 Complete bounded shared aim function control flow ([report](research/QVM_COMPLETE_AIM.md)).
|
docs/research/QVM_COMPLETE_AIM.md
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Complete bounded shared aim function control flow
|
| 2 |
+
|
| 3 |
+
RE-618. Reproduce with `python3 scripts/analyze_qvm_complete_aim.py`; check with `python3 scripts/test_qvm_complete_aim.py`. The [machine report](qvm-complete-aim.json) retains source identity and bounded fixtures.
|
| 4 |
+
|
| 5 |
+
The new driver interprets290494–291638 as one source range, retaining jumps, native call order and actual36717 state across all stages. It explicitly dispatches only recovered RNG and finite math natives, patches source LOCAL result destinations and rejects unexpected children. Calls and source trace hashes remain visible. Sixteen ordinary cases agree with the previously independently validated staged chain; direct f32 endpoint checks provide an additional vector oracle. A zero-seed case records128 RNG calls plus one sqrt,32 rejected candidates, no trig and unchanged output. The32-attempt limit is a harness boundary, not a successful function return. Completion stops before the void epilogue, and establishes aim data flow rather than a hit/damage outcome.
|
| 6 |
+
|
| 7 |
+
No original executable/full QVM/native collision/physics/rendering executes. Native responses and legal input domains are supplied or unverified where stated; full runtime/macOS/Linux Rust/Bevy/Solari gameplay remains open.
|
docs/research/README.md
CHANGED
|
@@ -453,3 +453,6 @@ The extension collector adds five focused tests for native tail calls, command-r
|
|
| 453 |
|
| 454 |
|
| 455 |
[RE-617: Aim endpoint construction and selected alias behavior](QVM_AIM_ENDPOINT.md). Reproduce with `python3 scripts/analyze_qvm_aim_endpoint.py`; check with `python3 scripts/test_qvm_aim_endpoint.py`.
|
|
|
|
|
|
|
|
|
|
|
|
| 453 |
|
| 454 |
|
| 455 |
[RE-617: Aim endpoint construction and selected alias behavior](QVM_AIM_ENDPOINT.md). Reproduce with `python3 scripts/analyze_qvm_aim_endpoint.py`; check with `python3 scripts/test_qvm_aim_endpoint.py`.
|
| 456 |
+
|
| 457 |
+
|
| 458 |
+
[RE-618: Complete bounded shared aim function control flow](QVM_COMPLETE_AIM.md). Reproduce with `python3 scripts/analyze_qvm_complete_aim.py`; check with `python3 scripts/test_qvm_complete_aim.py`.
|
docs/research/qvm-complete-aim.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
scripts/analyze_qvm_complete_aim.py
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""Interpret the entire290493 body with source control flow and bounded sampling."""
|
| 3 |
+
import itertools,math
|
| 4 |
+
from pathlib import Path
|
| 5 |
+
from analyze_deferred_auth import trace_slice
|
| 6 |
+
from analyze_bsp_references import trace_fields
|
| 7 |
+
from qvm_research_support import source
|
| 8 |
+
from qvm_server_fire_support import seed,ENTITY,RIGHT,UP
|
| 9 |
+
from qvm_shot_math_support import TRIG
|
| 10 |
+
from analyze_qvm_server_shot_setup import random_word
|
| 11 |
+
from analyze_qvm_spread_parameters import LOCAL
|
| 12 |
+
from analyze_qvm_aim_endpoint import inputs,endpoint,ORIGIN,DIRECTION,OUTPUT
|
| 13 |
+
from analyze_qvm_acceleration import PS,word,number
|
| 14 |
+
from analyze_qvm_slide_prefix import vec
|
| 15 |
+
from analyze_qvm import write_json
|
| 16 |
+
|
| 17 |
+
def complete_aim(ins,m,origin=ORIGIN,direction=DIRECTION,output=OUTPUT,attempt_limit=32):
|
| 18 |
+
if not 1<=attempt_limit<=64:raise ValueError('Attempt limit outside1..64')
|
| 19 |
+
m=dict(m);initial={'origin':vec(m,origin),'forward':vec(m,direction),'right':vec(m,RIGHT),'up':vec(m,UP),'output':vec(m,output),'seed':m[21044]&0xffffffff};m.update({LOCAL+216:ENTITY,LOCAL+220:origin,LOCAL+224:direction,LOCAL+228:output});pc=290494;trace=[];calls=[];effects=[];candidates=[];points=[x['index']+2 for x in ins[290494:291639] if x['opcode']=='CALL']+[291079,291081,291639]
|
| 20 |
+
for _ in range(512):
|
| 21 |
+
t=trace_slice(ins,290494,291639,m,{},resume_at=pc,stop_targets=[p for p in points if p!=pc],local_base=LOCAL,capture_stack=True,limit=8192);m=t['memory'];trace+=t['trace'];pc=t['stop'];calls+=t['calls']
|
| 22 |
+
for c in t['calls']:
|
| 23 |
+
target=c['target']
|
| 24 |
+
if target==36717:m,e=random_word(ins,m);value=e['result'];effect={k:e[k] for k in ['initial_seed','final_seed','result']};effect['trace_sha256']=e['execution']['trace_sha256']
|
| 25 |
+
elif target==-107:value=word(math.sqrt(number(c['args'][8])));effect={'model':'sqrt','input':c['args'][8]&0xffffffff,'result':value}
|
| 26 |
+
elif target in TRIG:m,_,value,effect=TRIG[target](ins,m,c['args'],(0,0,0))
|
| 27 |
+
else:raise ValueError('Unrecovered aim child '+str(target))
|
| 28 |
+
assert ins[c['instruction']+1]['opcode']=='STORE4' and ins[c['instruction']-2]['opcode']=='LOCAL'
|
| 29 |
+
m[LOCAL+ins[c['instruction']-2]['operand']]=value;effects.append({'call':c,'effect':effect})
|
| 30 |
+
if 291066 in t['trace']:candidates.append({'x':m[LOCAL+28]&0xffffffff,'y':m[LOCAL+32]&0xffffffff,'radius_squared':m[LOCAL+128]&0xffffffff,'accepted':pc==291081})
|
| 31 |
+
if pc==291639:status='void-epilogue';break
|
| 32 |
+
if pc==291079 and len(candidates)>=attempt_limit:status='attempt-budget-exhausted';break
|
| 33 |
+
else:raise ValueError('Aim segment bound exceeded')
|
| 34 |
+
return m,{'initial':initial,'attempt_limit':attempt_limit,'candidates':candidates,'status':status,'execution':{'calls':calls,'children':effects,**trace_fields({'trace':trace},False)},'offsets':{str(o):m[LOCAL+o]&0xffffffff for o in [76,52]} if status=='void-epilogue' else None,'output':vec(m,output),'final_seed':m[21044]&0xffffffff}
|
| 35 |
+
|
| 36 |
+
def main():
|
| 37 |
+
q,raw,h,ins=source();rows=[]
|
| 38 |
+
for ident,rng,ground in itertools.product([2,10,14,23],[1,0x12345678],[0,1023]):
|
| 39 |
+
m=seed(raw,h,ident);m[PS+144]=0;m[PS+312]=ident;m[PS+68]=ground;m[PS+216]=10;m[21044]=rng;m=inputs(m,direction=(.6,-.25,.8));_,r=complete_aim(ins,m);_,stages=endpoint(ins,m);r['staged_comparison']={'output':stages['output'],'final_seed':stages['final_seed'],'status':stages['status']};rows.append(r)
|
| 40 |
+
m=inputs(seed(raw,h));m[21044]=0;rows.append(complete_aim(ins,m)[1])
|
| 41 |
+
write_json(Path('docs/research/qvm-complete-aim.json'),{'source':q['source'],'sha256':q['sha256'],'probes':rows,'limits':'Complete290494–291638 source control flow; stop before void epilogue. Actual36717, finite supplied math natives;32-candidate limit. No fabricated completion on budget exhaustion. No engine trace, hit or damage.'})
|
| 42 |
+
if __name__=='__main__':main()
|
scripts/test_qvm_complete_aim.py
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json,unittest
|
| 2 |
+
from pathlib import Path
|
| 3 |
+
from test_qvm_aim_endpoint import expected
|
| 4 |
+
class CompleteAimTests(unittest.TestCase):
|
| 5 |
+
@classmethod
|
| 6 |
+
def setUpClass(cls):cls.r=json.loads(Path('docs/research/qvm-complete-aim.json').read_text())
|
| 7 |
+
def test_independent_endpoint_and_staged_agreement(self):
|
| 8 |
+
self.assertEqual(len(self.r['probes']),17)
|
| 9 |
+
for p in self.r['probes'][:-1]:
|
| 10 |
+
self.assertEqual(p['output'],expected({'initial':p['initial'],'offsets_stage':{'offsets':p['offsets']}}));self.assertEqual({k:p[k] for k in ['output','final_seed','status']},p['staged_comparison'])
|
| 11 |
+
def test_four_rng_calls_per_candidate_plus_optional_pair(self):
|
| 12 |
+
for p in self.r['probes'][:-1]:
|
| 13 |
+
calls=[c for c in p['execution']['calls'] if c['target']==36717];self.assertIn(len(calls)-4*len(p['candidates']),[0,2]);self.assertTrue(p['candidates'][-1]['accepted']);self.assertFalse(any(c['accepted'] for c in p['candidates'][:-1]))
|
| 14 |
+
def test_zero_seed_no_output_or_trig(self):
|
| 15 |
+
p=self.r['probes'][-1];self.assertEqual(p['status'],'attempt-budget-exhausted');self.assertEqual(p['output'],p['initial']['output']);self.assertEqual(len(p['candidates']),32);self.assertEqual(len(p['execution']['calls']),129);self.assertFalse(any(c['target'] in [-104,-105] for c in p['execution']['calls']))
|
| 16 |
+
def test_unexpected_children_are_not_supplied(self):
|
| 17 |
+
for p in self.r['probes']:self.assertLessEqual(set(c['target'] for c in p['execution']['calls']),{36717,-107,-104,-105})
|
| 18 |
+
if __name__=='__main__':unittest.main()
|