Repository navigation
Physics Rigid Hinge Limit Diagnostic
Exactadoptions1280=b8d4fb1f,1264=41d3bf35,1294=63d20195a9d980a31930c682ed56e8be87d5a11f areverifiedbyRoot.1294actualparent18a30a46,44/44nativeownedmodes/blobs from6bd3376d→4e4f3070exactandadoptedHEADancestryretained. Original1279archiveisstillpreservedat /Users/user/.codex/worktrees/physics-multi-hinge/barocss HEAD0de27c51574e2da6b09966c7ef0c28435d9c9592; itisnotcopiedintosharedRoot. The1294historicalrunnerrequiresits originalexternalparentHEAD; nocurrentRootselfcontainedreplayclaimed. OldwholeCADindexpinischangedcurrentindex, notrewritten.
#1298 isIMPLEMENTING withthreehelpers atb8d4fb1. #1308 and#1309 have actualcodedependencyavailable; eachstillrequiresits ownbeforecodecontractselection/fullsource-input-error-referencefreezebeforeexecution. Detailedoriginalconsumer/numeric/copy/errorownershipplansarealreadyinWiki/Issues. Thisisnotanotherparenttrialrequestorintegratedscreenwait. #1264physicalNO-GO/#1279physicalNO-GO/#1294diagnosticGOonly remain. No publicengine/fullCCD/jointrepair/cost/Viewer/releaseGO. Currentcomposedvalidation remainsmerged; integration validation pending674/1254.
Issue #1294.
Parents #1279 numerical NO-GO, engine #1179/family #660. Exact source/archive/full-control freeze precedes any execution. Duplicate search for hinge angle found original joint/robot cards but no final accepted-angle diagnostic. This is a separate question; do not rerun or tune #1279.
#1279 source82c24a21 → native0de27c51574e2da6b09966c7ef0c28435d9c9592, base6bd3376d. Three graphs accept197/retain198states then independent angle .8000000413549193 exceeds limit/report .8 by4.135491926237744e-8rad (original1e-9 gate). Two actual public single-pair worlds match the private bodies/joints through that failure. Build/types/consumer0. Physical10sGO=false. Native receipt6097249893; manifest45175603c18bd1417f2bcd58a6af49b201c4c2da02dc8b9cedf19e0ab9bf9e2f. This does not identify the historical phase cause.
Can final-state report/limit/anchor residuals be reproduced independently on the exact last-pass196 and first-fail197 records, and does ONE explicitly detached original hinge projection evaluation demonstrate any post-report pose change? Source order shows clamp/report calculation precedes the anchor helper, which can rotate bodies. This is a hypothesis to measure on copied inputs; archived post-state cannot establish its hidden historical phase.
Only NEW examples/physics-rigid-hinge-limit-diagnostic/** in an isolated Physics lane. Original adopted6bd3376d Physics math/parser/joint/anchor sources remain unchanged. Native1279 is an immutable data archive, not an adopted public interface. Hash/bit validation of the archive and real original API are mandatory before execution; unrun Integration UI is not a wait. No original package/fixture/root/lock/Integration/preview edits.
Read only original source/source-repeat/genuinecompiled report + Float64LE traces at196/197, both groups, and corresponding two actual canonical public single-pair traces. Verify all275sourcepins,26artifacthashes/native manifest, rawframe layout/order and archive lengths first. Recorded private values retain Z0/Z1; public group1 is canonicalZ0 and only copied comparison positions get+1. Restore signed zero from raw13fields before live comparisons. Caller data and accepted archive remain untouched. Metres/kg/s/radians; original 1kg .1m spheres, anchors±.1, axisZ, limits±.8, motor.5/.0001/.01 anddt1/120 unchanged.
Derive raw and normalized quaternion relative angles, world anchors/gaps and axis error independently with scalar Python Decimal80/120 algebra and separately declared scalar atan2. Compare original math angle and anchor calculation against independent reference within1e-12rad/m; Decimal algebra80/120difference<=1e-25 per unit. Record both limit excess and reported-angle difference; retain original1e-9 gate, no widening. No unique force or load calibration claim.
Construct copied original Bodies and Joint for each archived pair using original parseBody/parseJoints and unchanged metadata/default inertia. Do not construct/step a PhysicsWorld or replay hidden motor/cache/history. Invoke original projectDynamicHingeJoint once on the copies atdt1/120; persist before/returned-state/after rawposes and independent angles/anchors/momentum/energy. Separate returned clamp/report, final copied pose and original archived result. Qualify source/repeat/genuinecompiled original-helper parity exactly. Check finite/unit raw data and copy isolation; a failure of original bounds is recorded, not repaired. This one-stage result is a counterfactual diagnostic and never establishes10s/world/joint-networkGO or historical causation.
Freeze literal observer controls: resting pair; exactly at±.8 limits; slight known angle above/below; sign-equivalent quaternion; Ztranslation; changed reported angle without changed pose; nonfinite/null/IDs/raw length/hash/archive refusals. Freeze expected metadata, errors, source graph, operation-order and raw-zero conventions. Unexpected harness/reference/protocol failures remain execution failures, not physicalNO-GO. Actual affected CAD/Physics build, strict adaptertypes and ONE bounded diagnostic consumer precede native handoff. No parameter sweep, more projection passes, model changes or world/refinement/cost/Chrome/Viewer campaign.
Persist all selected records, independent residuals, exact one-call outputs, negative controls, graphs/hashes and first failure with immutable source/native commits. A successful diagnostic can justify a separate coupled limit/anchor correction contract; it is not repair acceptance. Public/wholeworld correction requires separately frozen API/model/10s gates. Keep #1279 physicalNO-GO and #1179/#660 OPEN. Sole Integration adopts a scoped native delta once; exact adoption supports leaf closure while current integrated validation is tracked under#674. No productpush/PR/deploy/npm/release.
Actualbase6bd3376d6943a8d84b83ca62b0031a9ec8927bea → measured3245bf5ae319990732e725a8eadee7af0f76f34b,14onlynewownedfiles. frozen-source286pins SHAeb679d523ad0815bd36425191d94b429262c8f364d235bd16e5b9e65f3b3a333, receipt posted before all execution. Independent source review confirms allpins, raw13 restore order,24one-call bound and noWorld/motor/tuning. Actual emittedbytes will be pinned before one diagnostic consumer. No numerical diagnostic outcome is claimed here yet.
Actual CAD/Physics build and stricttypes exit0; 502emittedbyte freeze, SHA06fe669fc86a7d4df28b2c5c422c83100152821cec906750fdb4c8095edb2725, was independently reread before the oneconsumer. All21parenthelper emittedhashesmatch. Consumer completed24one-callobservations with independentdiagnosticagreement and exactsource/repeat/compiledhelper parity; worldCreated/stepped/motorDrivenfalse. Originalarchives/source/controlbytesretained.
Step196detachedprojection returns angle0.797128646551988rad with independent residual0. Step197archive report0.8 differs from rawindependent angle by4.1354919266822776e-8rad. ONE fresh originalprojection oncopiedstep197 returns0.8 but finalcopiedraw/normalizedangle residual is1.136839100635001e-8rad, stillaboveoriginal1e-9gate. Bothpairs/privateZ0/Z1/publiccanonicalZ0/all3graphsagree. This proves the freshone-call postcondition mismatch on theseinputs. It does not establish the historical world-phase cause, solve theparent10sNO-GO, or authorize extraoriginalpasses/tuning. Next candidate must define a separate angle-and-anchor law, frozenreference andadoptiongates.
Measured3245bf5a → native 4e4f307097eb54b9173552edad68a704457a7f76,2commits/44onlynewownedpaths/clean. Manifest4053c025ac041b468f6b68fafd676b4fce382ede58f11b9d9f82edf2c8075fe0,27artifacthashes/28files. Root verifiedallartifact/decoded/sourcehashes and addedpathset. Native receipt. Build/types/oneconsumerexit0. Controls25independent/17typed/10archive pergraph;source8originalsrc/compiled8genuineemittedhelpers. Maxoriginal/reference mathdelta1.439070499421966e-17,Decimalalgebra4.377246663946343e-80 (atan2separate). DiagnosticGO is distinct from physical1279repair/10sGO; actualadoptionpending.
#1309 coupled hinge reference defines one clamp/held-orientation/COM-preserving anchor correction while preserving v/omega bit-for-bit in the detached first candidate. No reactions are invented. A velocity/motor law remains a separate later design. It tests original196/197archives and independentpose/momentum controls first. The law is untested; no extraoriginalpasses/tolerancetuning or worldrepairclaim. A laterworld/motorconsumerrequiresitsownIssueandGO.