Jedi first-damage source: original chain derived; live observer incomplete

The retained encounter cannot identify its first damage source. Original x86 establishes a damage dispatcher, health subtraction and generation-checked source ancestry. A task-local read-only decoder and bounded capture plan are retained, but readyForCampaign=false and usableDamageAttribution=false. The first unresolved branch is the damage-script call before health subtraction: its shield-bin write and event correlation are not yet byte-anchored. A synchronous capture/stop adapter is also required before another encounter campaign.

This source-only task used input main 3e24bee3762bc2ee87941a3faefee2daa20a4f3d. The four requested encounter/route notes were reviewed. No gameplay, build, browser, boat operation, restart, guest write, engine call or subworker occurred. Static route extraction was not repeated. Bossk266 remains a candidate, not an identified attacker; no civilian is classified as hostile.

What the retained evidence establishes

The last healthy sample is retrieved/session/emerge04-neutral-sample.json in scratch/runs/20261103-jedi-listener-ready-encounter/. At14:33:58.382Z it reports health100/shield100, sector265 and simulation accumulator808352ms. The rejected corridor01-pair-1-sample.json,14:34:07.081Z, reports health30.828529/shield0 and accumulator817033ms. The gap is8699 host ms /8681 simulation ms. The final PNG was reviewed in the predecessor task; this task does not claim another visual attacker identification.

Neither sample contains a damage-entry stack, writer trace, source pointer, source generation or projectile identity. The player's position/height change does not establish a fall as the cause. Resource requests for concuss5.wav, expl_09.wav and kshieldhit01.wav in the retained log are suggestive context, but lack victim/source identity and a first-event timestamp. They cannot prove an incoming projectile, identify its owner, exclude environmental damage or show which cause occurred first. Failed loose-file probes followed by archive access are not evidence of an audio defect.

The missing observation is a coherent record at the first local damage entry or first health/shield decrement, including original caller, victim, raw source, amount/type, live source slots and generations, with the matching health/shield write. A later screenshot or nearest-in-time actor is insufficient. retained-review.json in this task's ignored run records the exact sample hashes, gap calculation and relevant resource-request lines.

Original byte-anchored chain

All addresses below are original VAs at image base0x400000 in original JkDemo.exe,1388032 bytes, SHA256 0f8a1b01cfdcb196b043ebb4ebf42dc3a6d4fa394034ebfb8cad9e383e14aa46. The accepted Coords/health/inventory/clock proofs are reused unchanged, not requalified by another campaign. New static disassembly and exact code spans are retained in this task's run.

Original location Established behavior
0x4cca60 Entry stack is return,+4 victim,+8 source,+12 float amount,+16 damageType. Rejects nonpositive amount and victim flags, then dispatches on victim+0x0c.
0x4ccabf Calls 0x4dfda0 with message0x0a and the damage/source arguments under the victim flag gate. Its floating return replaces the amount before dispatch continues. This is an interception boundary, not proof of an actual loss.
0x4ccb11…0x4ccb19 Victim types2/10 call 0x4ecf40(victim,source,amount,type). Type3 takes a separate 0x4d4b30 path.
0x4ecf40 Saves victim in ESI; source is loaded into EDI at0x4ecf93. Applies actor flags/script and other gates.
0x4ecffd Calls 0x4cc930(source); EAX becomes resolved EDI. Raw source and resolved source can differ.
0x4cc930…0x4cc959 Follows source+0x80 only while source+0x84 equals the pointed slot's+0x74 generation. A failed comparison leaves the current source as the engine result.
0x4ed064…0x4ed074 Loads victim+0x134, subtracts stack+0x1c amount and stores back to victim+0x134. Exact store bytes: d9 9e 34 01 00 00.
0x4ed0c8 / 0x4ed2a6 Calls death path0x4ed290; that path explicitly zeroes victim+0x134.

At the health-store boundary after four saved registers, stack+0x10 is the actor function return, +0x14 victim, +0x18 raw source, +0x1c adjusted amount, +0x20 type. EDI may already be the resolved source. These offsets must not be used with an arbitrary later stack or an EBP-chain guess.

0x4cc960 validates a thing index against world+0xbc and derives its slot from world+0xc4 with stride0x2b0, requiring a nonzero slot+0x0c type. 0x4ccb50 independently allocates count*0x2b0. Slot+4 is initialized as the index. A fresh observer must read this pool and world+0x124 local player in the current process, rather than reuse the predecessor's numeric addresses.

Distinct original callers demonstrate why health loss alone is ambiguous:

Caller Arguments/context visible in original bytes
0x4d4ded Uses victim EDI, source ESI, amount ESI+0x13c and type ESI+0x134 in a collision path. The source object must be captured; static actor placement cannot replace it.
0x4e9799, 0x4e983e Pass victim as its own source and type0x40 after a squared excess-magnitude calculation in the movement/impact code; a fall/impact candidate path. Numeric type alone does not prove a fall.
0x4ecee7 Sector-flag/timer branch passes victim as source, amount10.0 and type0x20 after timer exceeds20000; an environmental candidate path.
0x50277b Script damage verb also calls the dispatcher. Other direct calls to0x4ecf40 exist, so a dispatcher-only trace is not complete write coverage.

The last two rows are reasons to retain caller/context. They are not claims that either branch executed in the retained encounter. The manifest pins16 direct dispatcher callsites found in the original disassembly, not every possible indirect entry or every health writer.

The new source-parent42-byte span has SHA256 915fb7abd68831e10f8559829a0469c985287891a3e645068590d9f082fcdf48; dispatcher225-byte span 485098a2eb8bad6c7c9c16f59ec58138ca39f3c52001aacfacda210b8cc8c905; health-store22-byte span 2557abd72cb371acb774839e6861c11079353dcd0130b76d9051daa1a7393bb4. original-anchors.json contains full bytes, file offsets and hashes for all15 code spans and16 callers. Its SHA256 is f84a59cc1b949121855c7cf638965a8fe1a712a8a59d9c7b74c1acba127f0c58.

Executable decoder and bounded observation recipe

Everything executable remains task-local under scratch/runs/20261103-jedi-damage-source/; no runtime or controller source was changed. anchors.cjs refuses a different original PE. decode.cjs accepts only a supplied read callback and stopped capture metadata. It checks the sealed manifest, all original code spans, exact entry/caller, process PID/launch token,100ms capture age, current world/player/pool, slot alignment, index/type, bounded ancestry, generations and stable rereads. Its maximum is 128 reads /16384 bytes /8 ancestry slots; no scan, write or engine invocation. An invalid generation is rejected for attribution even though the original resolver would stop traversal there.

The decoder emits raw source and resolved engine source separately. A null source remains unknown. Numeric type/index/signature/generation are retained; no model/template name, visible weapon or hostility is inferred. Every accepted record remains pre-damage, actualDamageProven=false and usableDamageAttribution=false: a script may veto it. Process identity and coherence labels must come from a trusted host adapter, not copied JSON. The offline tests do not establish authentic runtime provenance or sibling-thread coherence.

capture-plan.cjs prints an executable JSON recipe without launching anything. It identifies existing entry breakpoint, trace and watch facilities and a proposed30s/64-event arm bound. stopCandidate conservatively revokes controls, attempts all key releases, persists the record and requests stop at the first candidate or decoding error, before screenshot review. Failure controls show that revoke/release/persistence errors still attempt the remaining release/stop callbacks. These are callback tests; actual GetMessage keyup delivery was not tested here.

For eventual authorized integration, the event capture must run synchronously when the interpreter returns at the breakpoint/watch boundary, before another guest slice or control authorization. Capture first, latch stop immediately, release keys through existing timer/finally paths, then retain supplementary PNG. First error/death/health or shield loss and the existing original-code, identity, freshness, one-use approval,0.08-distance/4° and1s neutral-pair guards remain mandatory. No new gameplay keydown may follow the latch. A conservative stop before an intercepted damage attempt must be labeled an attempt, not an actual hit or combat result.

First concrete unresolved branch: 0x4ccabf → 0x4dfda0 can transform the amount before the health subtraction. The exact original shield-bin store and its script message/source correlation remain to be derived. Thus a health-only watch does not establish first-damage coverage. In addition, current CLI trace output has no synchronous task callback to copy dynamic source headers and invalidate controller approval. A debug prompt blocks normal control pumping; nonstopping trace output and external screenshot polling do not enforce the required stop/release behavior. Only the primary WAT watch halts per block; additional watch addresses are JS batch samples. These concrete limitations are reproduced from src/13-exports.wat and test/run.js, with excerpts retained in tracing-basis.txt. No ready live observer or authorized campaign is claimed.

Validation, recovery and lifecycle

34 offline checks pass. They cover exact original-byte matching, changed PE/ manifest/code, wrong process/token/player/source, stale/future/noncoherent capture, wrong thread/module/entry, invalid references/pool/amount/inventory, source reuse/cycles/generation changes, ambiguous callers, null-source unknown and stop/release failure paths. Positive ancestry fixtures are synthetic; no guest x86 was executed by these tests. They identify no live attacker, combat result, emulator defect or fix. A first test invocation had a .cjs module-resolution error; the suffix was corrected before the passing run. No build was run because this task changes only this note and ignored evidence and explicitly excludes builds.

RECOVERY.md gives exact decoder/test/disassembly commands. The own run retains the original compressed PE, static disassembly, full byte anchors, decoder, plan, tests, retained-sample analysis, full hashes and result/final records. It is ignored evidence and must be retained; the note alone cannot recover task-local code from Git.

No box was assigned or operated. Predecessor bx_zqdxvgxr remains recorded STOPPED, original expiry 2026-11-03T16:11:37.264Z, untouched; this task made no fresh lifecycle query. Coordinator retains lifecycle ownership. Next work is the shield/script correlation and synchronous capture/stop adapter qualification, followed only then by a separately authorized campaign. This single task ends with AUTHOR RELEASE and STOP, without reuse.