Date: 2026-05-04 Scope: libs/oya/**/*.{ts,py} (excluding *.test.ts,
*.spec.ts, node_modules, dist) Tool:
scripts/stub-indicator-scan.sh --mode=stubs|random
Status: clean (qualitative stub markers)#
./scripts/stub-indicator-scan.sh --mode=stubs $(...oya files...) returns
zero actionable hits after this audit. Every prior hit was either fixed
(real stub → real implementation) or marked // stub:legitimate <reason> when
the qualitative phrase appears in user-visible text, JSDoc parameter
descriptions, or physics edge-case comments.
Per-file actions taken#
| File | Line | Original | Action |
|---|---|---|---|
core/src/gps-navigation.ts |
3636 | let val = -KH[i*n+j] * this.P[j]; // placeholder, need full multiply<br>followed by val = 0 and the real (I-KH)*P loop |
Fixed: removed the dead placeholder line; the real Kalman covariance update was already in the inner loop. |
core/src/multi-view-fusion.ts |
608 | // For simplicity, store a placeholder fundamental matrix<br>F = [t]x only — ignored intrinsic matrices and relative rotation |
Fixed: implemented full F = K2^-T _ E _ K1^-1 with E = [t_rel]x _ R_rel. Added mat3Mul and mat3InvUpperTri helpers. R_rel = R2 _ R1^T, t_rel = t2 - R_rel*t1. K1, K2 built from CameraIntrinsics. 150 prior multi-view-fusion tests still pass. |
core/src/edge-deployment.ts |
481 | 'Model exceeds on-chip SRAM; will use external memory (slower)' |
Marked legitimate (// stub:legitimate user-facing reason text describing real EdgeTPU behaviour). User-visible string. |
core/src/mission-execution.ts |
438 | Mission will use ~${...}% battery, leaving less than ${...}% for RTH |
Marked legitimate (// stub:legitimate user-facing validation warning text). UI message. |
core/src/kinematics.ts |
866 | // Cannot hover — would need >100% throttle |
Marked legitimate (// stub:legitimate physics edge-case description (TWR<1 cannot hover)). Comment describes a physical impossibility, not deferred work. |
core/src/inter-drone-communication.ts |
194 | @param signed - Whether to include signature placeholder (JSDoc) |
Marked legitimate in expanded JSDoc (stub:legitimate the placeholder bytes are filled in by sign()). The signed flag controls real placeholder injection by the MAVLink v2 signing path. |
core/src/sapiens-integration.ts |
1587 | // No mapping — emit a zero-confidence placeholder |
Marked legitimate (stub:legitimate sentinel keypoint). The zero-confidence sentinel is a real fallback that lets downstream code index by mediapipe index without bounds checks. |
Random scan (--mode=random)#
--mode=random reports 202 actionable hits across the same file set; all are
legitimate by inspection:
- UUID generation (
mission-execution.ts:138-141,enterprise-platform.ts:1708-1711):Math.randomfor UUIDv4-shaped identifiers. Acceptable for non-security IDs. - Box-Muller / inverse-CDF Gaussian sampling (
physics-simulation.ts,kinematics.ts):Math.randomis the correct primitive for stochastic noise injection. - RANSAC sampling (
visual-slam.ts:3087,obstacle-avoidance.ts:2443): random sample selection is the algorithm's foundational step. - Simulation scenario generation (
modern-simulation-platforms.ts,physics-simulation.ts): SITL/simulation tooling intentionally injects noise/jitter. - Canary deployment routing (
edge-deployment.ts:2327): A/B traffic split. - Phase offset / clock jitter (
multi-camera-coordination.ts:698).
These do not need per-line // random:legitimate annotations because they are
not staged in this commit; the pre-commit hook only runs the stubs mode
(--mode=stubs) per .husky/pre-commit. The whole-repo random scan runs in CI
(scripts/stub-indicator-scan.precommit.test.mjs / nightly job) and currently
classifies all 202 hits as expected stochastic primitives.
Summary#
- Real stubs found and fixed: 2 (gps-navigation Kalman, multi-view-fusion fundamental matrix).
- Legitimate qualitative markers annotated: 5.
- Random-source false positives logged but not annotated: 202 (no action needed; pre-commit hook is stubs-mode only).
- Adversarial-clean status:
--mode=stubsreturns zero hits acrosslibs/oya/.
V1-P2-2100..V1-P2-2107 wirings (this session) all left their files adversarial-clean per the same scan; this audit confirms the surrounding file set has no remaining qualitative stub markers.