Blame for docs/research/QVM_LOCATION_ALLOCATION_BATCH.md

# Location, formatting and objective allocation: RE-1034–1043

Ten tasks recover five source/Rust contracts across **115 paired scenarios**. They extend the [resolved effect-feedback batch](QVM_EFFECT_FEEDBACK_BATCH.md) by executing location selection/text and the actual radio formatting chain, then connecting successful pool allocation to defuse event construction. Every task has its own commit and focused validation.

| Tasks | Contract | Paired scenarios |
| --- | --- | ---: |
| RE-1034 / RE-1035 | [Source location selection](QVM_LOCATION_NEAREST.md) / [Rust selection](RUST_LOCATION_NEAREST.md) | 16 |
| RE-1036 / RE-1037 | [Source bounded location text](QVM_LOCATION_TEXT.md) / [Rust text](RUST_LOCATION_TEXT.md) | 24 |
| RE-1038 / RE-1039 | [Source located radio](QVM_LOCATION_RADIO.md) / [Rust radio](RUST_LOCATION_RADIO.md) | 35 |
| RE-1040 / RE-1041 | [Source allocator](QVM_LOCATION_ALLOCATION.md) / [Rust allocator](RUST_LOCATION_ALLOCATION.md) | 16 |
| RE-1042 / RE-1043 | [Source allocated defuse event](QVM_LOCATION_ALLOCATED_DEFUSE.md) / [Rust composition](RUST_LOCATION_ALLOCATED_DEFUSE.md) | 24 |

## Cached, trigger and nearest location268248

An existing actor848 pointer returns immediately only while signed current tick is strictly before cached entity596 expiry. Equality is expired. No query, classname comparison or fallback geometry is evaluated on a live cache hit. The function does not update the actor's cache field on any recovered path.

On a miss, engine-33 receives an origin-centered query box with extents40/40/52 and a1024-entry response buffer. The first returned entity whose classname compares case-insensitively equal to `trigger_location` wins immediately. Actual69386 performs the comparison. Query results and their order are supplied engine evidence; the routine does not independently prove geometric trigger containment.

Otherwise it traverses the list at34999640. Each entity524 points to a structure whose24 field holds the next location. Squared f32 distance must be at most the current best, initially201326592. Engine-27 PVS is queried only within that bound. A visible candidate replaces the best, including equal distances, so later visible ties win. Supplied identity/order, finite positions and native box/PVS responses bound this contract. Tests include cache equality/wrap, trigger precedence, hidden locations, empty lists and distance-bound rejection.

## Location text268637 and actual formatting

The wrapper always calls selection once. A null result returns0 without clearing the destination. A selected location formats its592 string through actual70386/37914 and bounded-copy helpers, then returns1 regardless of empty text or clipping. The original warning5123 occurs when formatted length reaches capacity. The destination is zero-padded within capacity; neighboring bytes survive.

Twenty-four cases cover missing/found location, empty/quoted text and capacities1/4/16/150. Rust uses the existing recovered bounded formatter. A missing location can bypass otherwise unused zero-capacity input; a found unsupported capacity rejects without mutation. This wrapper contract supplies the selected pointer; the next composition executes actual selection.

## Radio location/formatting composition

Thirty-five scenarios execute193620,268637,268248,70386,70458,37914 and their recovered comparison/classification/formatting helpers. The location label now derives from actual cache/trigger/list/PVS selection instead of a supplied text response. A missing location reaches the original caller's first-byte clearing path and produces empty radio text.

The actual ring formatter alternates buffers and advances its counter for each recipient. Source/Rust cases cover counter wrapping, mode/team/sender gates, no recipients and64-entry tables. Quotes remain literal bytes. A160-byte location label clips to149 bytes in the150-byte location buffer, emits one warning and is then inserted into the radio command. Marker deadlines change only after sends as established in the prior batch.

Rust stages marker/ring updates and reuses the existing formatting and radio policies. Engine box/PVS responses and native network sends remain boundaries. The downstream client interpretation/audio behavior of radio messages is not exercised.

## Allocator277165 and initializer277125

The allocator scans non-client entities from slot64. A first-pass free record is skipped when its signed free timestamp is later than start+2000 and its wrapped signed age is below1000. A pool with count1024 retries once without the recent-free gate. Successful reuse preserves count and requests no registration.

Extension increments count, requests native-16 registration and initializes the selected slot. At count1022 with no reusable record, the source resets its scan index to0 and reaches the diagnostic-loop boundary. At a full1024-entry pool with no reusable record, it retries, increments count to1025 and reaches the fatal boundary before registration/initialization. These prefixes are preserved rather than clamped or invented as successful allocations.

Initialization writes index0, in-use532, class536/hash540 and field512=1023. It does not zero the record. Sixteen cases compare initialization and retained canary/free-time fields, exact age/start thresholds, signed wraps, extension/reuse/retry and terminal prefixes. Initial bytes outside the current count are explicitly supplied; the64-entry fixture already seeds slot64 with free-time0, which initialization preserves.

Rust uses caller-supplied fixed pool records. Terminal outcomes expose the preceding source count/scan writes and do not execute diagnostic/fatal effects. A subsequent allocation with terminal count1025 rejects. Native registration/lifetime and allocation concurrency remain external.

## Allocated defuse start through153917

Twenty-four source cases compose153774 with actual277695→277165→277125 and278138. Successful reuse, extension, recent-free extension and full-pool retry return real selected records to the constructor and caller. The constructor snaps origin and initializes event metadata; the caller then writes configured duration to184 and ORs mask32 into424.

Reports compare the initialized record separately from the final event, registration counts, selected slots, link requests, caller flags/clock/animation and retained neighboring words. Raw old record flags feed the caller event state; they are not replaced with synthetic fixed flags. Source execution stops at153917 before radio routing. Located radio is verified separately above; a complete combined allocator/radio/client path remains future work.

The Rust composition stages defuse and pool state until allocation/construction succeeds. Its successful-allocation source domain excludes allocator terminal prefixes, which are separately exposed by the generic allocator. Unsupported composition inputs reject atomically; that typed adapter guarantee does not assert rollback behavior for original fatal/undefined domains. Native linking and registration remain plans.

## Validation and remaining boundaries

Every task runs the matching `cargo xtask test-group qvm-location-*` or `rust-location-*` before committing. Selected reports regenerate twice identically. Source reports retain verified QVM identity and trace summaries; independent cache/format/arithmetic/pool expectations check results. Rust reports retain source-report SHA256 and compare all projected fields. Ten additional Rust tests cover ties, cache/trigger bypass, missing/clipped text, radio ring state, terminal allocator counts, preserved records and atomic composition errors.

Dependency mappings include the location module/CLI, formatter, radio, defuse and event-constructor consumers plus gameplay test-count metadata. The registry regression verifies selection. One full-suite check follows task ten: `cargo xtask test-full --since a1fd4e2b`. Python uses only the flake-declared Nix runtime, including Git hooks. Original assets are unchanged.

Native engine box/PVS, linking, registration, sending, allocation lifecycle and full ranking/log feedback remain explicit boundaries. These are finite QVM/Rust policies, not native runtime/platform conformance.

The single `cargo xtask test-full --since a1fd4e2b` run at the tenth-commit boundary passed: **2,018 Python tests** across559 modules, **339 Rust tests** (assets10, gameplay299, xtask30), **146 deterministic affected report producers**, storage policy, links across **623 documentation files** and whitespace guards. Regeneration left no changes, including earlier radio consumers. Focused dependency-registry regression passed. Validation notes are amended into task ten, preserving ten commits.

## Next work: RE-1044

- [ ] Combine allocated objective events and located radio with the complete defuse callback and longer lifecycle sequences, including allocation-terminal caller prefixes.
- [ ] Connect remaining ranking/log feedback and native box/PVS/link/register/send adapters, reusing existing engine research where applicable.
- [ ] Verify location/cache ownership, real spatial candidate generation and entity/linked identity/aliasing through allocation/free cycles.
- [ ] Migrate historical effect labels to semantic APIs while retaining evidence compatibility; connect radio/objective client rendering/audio consumers.
- [ ] Compose full154089 preparation, assistance/self recovery, defusing, movement54475 and timer155714; continue asset conversion, Bevy/Solari rendering and macOS/Linux acceptance.