Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Animation pack evaluation: Protofactor Basic Locomotion Animset

Technical verdict: Usable with conditions

Evaluation completeness: partial — current mechanical and declared-contract evidence is exhaustive, but no current engine or visual acceptance ran.

Confidence: medium

Evaluation date: 2026-09-01

Current evaluator: AnimSmith 0.10.0

Report format: 2

Detailed evidence: Protofactor Basic Locomotion evidence

Technical decision

The verified AnimSmith 0.10.0 release loads all 179 delivered FBXs. Untouched lint finds 24,186 constant-track notes and 36 time-monotonic errors across 12 files. Current declared contracts cover 177 individual motions; 58 command invocations pass and 119 fail per format, so loop and continuity results are conditions rather than approval. The declared remediation rerun generated all 39 candidates: 12 slices removed the 36 time-ordering errors, 24 in-place gait anchors reduced the three measured circular spreads from 0.7156245/0.4630161/0.6597812 to 0.0501911/0.0938395/0.0724415, and three constant-track candidates were emitted. These are mechanically verified, unpromoted candidates; no engine, visual, retarget, contact, or gameplay run occurred.

Capability coverage

Complete core

  • Delivered filename families cover locomotion, cover, turns, airborne actions, and transitions; 70 labelled root-motion files have matching in-place partners.

Partial supporting gameplay

  • Mechanical input health is measured. The current slices remove their known time-ordering errors, while remaining contract findings require loop and continuity decisions before locomotion blend adoption.

Absent

  • No current evidence establishes additive, first-person, paired interaction, engine acceptance, or artistic readiness.

Runtime sets and authored motion

No important runtime sets were identified.

Integration recipe

  1. Members/topology: topology=not-evaluated; declare selected locomotion rings rather than infer them from names.
  2. Timing/synchronization: sync=not-evaluated; resolve contract failures and measure selected loop members.
  3. State ownership: owner=not-evaluated; declare IP/RM movement ownership per clip.
  4. Composition constraints: composition=full-body; do not approve masks or additive use.
  5. Acceptance gate: gate=engine-and-visual-review; import, blend, retarget, and playtest selected candidates.

Technical issue register

IDSeverityProblem and impactPrimary ownerCurrent actionFuture AnimSmith potentialEvidence/status
BL-010majorTime ordering and loop continuity require declared review before runtime use.artist-authorReview the 12 external slice candidates and decide loop policy before promotion.Current declared slicing, gait anchoring, and constant-track pruning are mechanical candidates only; no automatic artistic repair is established.observed-animsmith; slices remove 36 time errors, while original contracts still have 119 failing invocations per format.

Engine status

RuntimeEvidence levelTechnical resultRemaining gate
Unity unspecifiednot-evaluatedNo current import or playback run.Import, controller, visual, and build tests.
Unreal Engine unspecifiednot-evaluatedNo current import or playback run.Import, retarget, graph, and build tests.
Godot unspecifiednot-evaluatedNo current conversion, import, or playback run.Conversion/import, graph, and export tests.
Bevy unspecifiednot-evaluatedNo current glTF handoff or runtime run.Conversion, addressability, runtime, and performance tests.

Fit and limitations

Best fit is an engine project willing to declare and validate its locomotion contract. Cross-pack compatibility, visual loop quality, contacts, and artistic fit remain untested.

Changes between AnimSmith versions

AnimSmith 0.10.0 — official release revalidated the 179-FBX baseline and 177 declared contracts, then reran 12 declared slices, 24 in-place gait anchors, and three constant-track trials. Earlier 0.7.0 evaluator and any engine or offline evidence are historical only.

Evidence status

Current evidence uses the official 0.10.0 binary and the canonical readiness ladder. Commercial sources and derivatives remain external.

Sources