A jump-navigation loop for alpha Centauri A converges in three jumps to the quality of the orbit solution, and no further: with the catalogue as the only input, jump 1 misses by 486 au RMS — priced almost entirely by the adopted parallax, along the line of sight. One onboard astrometric re-fix later (1 mas assumed, parameterised), the AB pair itself acts as a known-baseline rangefinder — the pair's angular separation grows as 1/d, so range error falls as sigma_fix * d^2/separation — bringing jump 2 to 1.15 au. Jump 3 lands on the 0.300 au floor set by the quality of the Pourbaix-Boffin orbital elements, and jumps 4-5 buy nothing: past that point the limit is the ephemeris, not the instrument. A 100-fold better sensor changes nothing at jump >= 3; only a better AB orbit solution does. Monte Carlo (4000 samples, persistent catalogue errors, exact 1/parallax nonlinearity) agrees with the closed-form covariance budget to 0.982-0.996 across all legs; with a perfect sensor the loop converges to 0.3001 au = the ephemeris floor, not to zero. The instant-jump image-aiming offset of 33.54 au reproduces targeting_sim exactly. Legs shorter than ~10x the pair separation are flagged out-of-regime for the small-angle model rather than reported.

Ran the full test suite (10/10 pass, 0.12s) and re-ran compute/jump_targeting.py from a clean shell to a scratch JSON: it reproduced report/jump_targeting.json digit-for-digit. Verified every number the claim states: leg-1 miss 485.7265 au (= d*sigma_plx/plx, radial 485.73 vs transverse 0.227 — "priced almost entirely by the adopted parallax, along the line of sight" holds); leg-2 miss 1.1505 au; legs 3-5 at 0.3001/0.3001/0.3001 au; floor_check with sigma_fix=0 converges to 0.3001 = sqrt(3)*(0.3/sqrt(3)); MC ratios 0.9914/0.9816/0.9961/0.9962/0.9962 match the claimed 0.982-0.996 band. Ran targeting_sim.py offline (PDB_HTTP_OFFLINE=1, cache sha a187495ca935) and compared the instant-jump aim-vs-image number at full precision: 33.543872344196 in both modules, difference 0.000e+00, and the embedded SIMBAD constants equal the cached fetch exactly — "reproduces targeting_sim exactly" is literally true. Confirmed S-0012 fulltext quotes 743+/-1.3 mas, P=79.91 yr, masses 1.13/0.97 (element sigmas sit in the stripped Table 1, acknowledged gap PDB-26), and the 0.300 au floor appears as "0.223 arcsec RMS = 0.300 au" in navigation_requirements_alpha_cen.py section 2. Main attack (attack_terms.py in scratch): dissected the leg-2 range budget term-by-term and swept sigma_fix over {0, 0.01, 1, 100, 1000} mas. The claimed rangefinder law sigma_fix*d^2/sep evaluates to 8.48e-5 au at leg 2, while the actual 1.124 au range error is f_sep*d = 1.111 au (f_sep = 0.23% baseline-scale error from sigma_a/a and sigma_plx/plx); the full miss table is IDENTICAL for sigma_fix from 0 to 100 mas (leg 2: 1.150522 in all cases; sensor only starts to matter above ~13 arcsec). The 100x-better-sensor statement verified (delta <= 6e-14 at legs 3-5). Seed-robustness check: MC/closed-form ratios at seeds 1, 2, 42, 777 span 0.9816-1.0095 — the agreement is not seed-cherry-picked. None of these attacks flipped a stated number or the conclusion; the mechanism-attribution hit is a wording defect, not a numerical break. || FINDINGS: Every number asserted by C-0029 reproduces: 486 au (485.7265), 1.15 au (1.1505), 0.300 au floor (0.3001), MC band 0.982-0.996 (0.9816-0.9962 at the stated seed), 33.54 au exact to the last bit, 10/10 tests. The convergence conclusion is confirmed and actually STRENGTHENED by the attack: the sensor is inert not just at jump >= 3 but at jump 2 as well (miss table flat in sigma_fix from 0 to ~1 arcsec; breakeven at ~13 arcsec), so "only a better AB orbit solution does" holds even more broadly than claimed — jump 2's 1.15 au is set by the 0.23% baseline-scale knowledge (sigma_a, sigma_plx), which is catalogue/orbit quality, not instrument performance. The one real hit: the clause "range error falls as sigma_fix * d^2/separation — bringing jump 2 to 1.15 au" names a term that contributes 8.5e-5 au of the 1.12 au; the quadratic sensor-limited law never dominates at any leg of this mission (it would first matter at d ~ 6e6 au for 1 mas). This misattribution does not change any stated number or the headline truth value.