Claude finished @Yeraze's task —— View job

PR Review: docs(site): publish the Estimated Accuracy algorithm page

Tasks:

• Gather context and analyze PR
• Read new documentation file
• Verify formulas against source implementation
• Review sidebar config and links
• Provide comprehensive feedback

Overall Assessment

Excellent documentation PR. The page is technically accurate, well-structured, and clearly written. I verified all formulas and example numbers against the actual implementation in positionEstimationService.ts. No code changes — docs only.

Formula Accuracy Verification ✅

Time decay: The docs say time_decay = 0.5^(age_h/24). The implementation uses Math.exp(-DECAY_CONSTANT * ageMs) where DECAY_CONSTANT = Math.LN2 / HALF_LIFE_MS. These are mathematically identical — e^(-ln2·t/24h) = 0.5^(t/24h). The simpler form in the docs is more readable for users and correct.

SNR weight: 10^(SNR_dB / 10) — matches Math.pow(10, obs.snrDb / 10) in observationWeight() exactly.

Kish n_eff: (Σw)²/Σw² — matches (wSum * wSum) / w2Sum exactly.

Confidence blend: clamp(n_eff − 1, 0, 1) — matches Math.min(1, Math.max(0, nEff - 1)) exactly.

Final uncertainty: 5·(1−c) + stat·c — matches the implementation exactly.

Worked examples (B): I independently verified the math — centroid ≈ 0.025 km from anchor 1, rms ≈ 0.32 km, confidence ≈ 0.013, uncertainty ≈ 4.94 km. All correct.

Minor Observations

1. Slightly confusing notation in Example A (estimated-accuracy.md, line 213)

wSum = 0.063      w²Sum = 0.063²        n_eff = 0.063² / 0.063² = 1.00

Writing w²Sum = 0.063² rather than w²Sum = 0.003969 shows the symbolic form but is a bit unusual — a reader has to mentally evaluate 0.063² to get the actual value. It's not wrong, just slightly harder to follow than Example B which shows evaluated numbers. Consider:

wSum = 0.063   w²Sum = 0.003969   n_eff = 0.063² / 0.003969 = 1.00

2. The rms_km ≈ 0 claim in Example A (line 215)

centroid = the anchor itself  ⇒  rms_km = 0

With a single observation, the centroid IS the anchor, so distance from centroid to anchor = 0, meaning rms_km = 0. This is mathematically correct. However, note that statistical_km = max(0.05, 0/√1) = 0.05, not 0. The example skips this step and jumps to the final uncertainty = 5 km, which is correct because confidence = 0, but showing statistical_km = 0.05 km would be consistent with Example B's level of detail.

3. Sidebar section name (config.mts)

The sidebar entry is in "Maps & Geography" which is correct and appropriate.

Links & Structure ✅

• All cross-links (/features/position-estimation, /features/estimated-accuracy, /features/maps, etc.) match existing pages.
• Both pages now cross-link each other bidirectionally.
• The sidebar placement after "Position Estimation" is the natural reading order.

Summary

No blocking issues found. The math is accurate, the examples are correct, the cross-links are valid, and the content genuinely helps users understand when to trust an accuracy circle. The two minor observations above are optional polish.