Table: reproduction and benchmark checks (verify.py).

| Check | Measured | Tolerance | Result |
|---|---:|---:|:---:|
| Repricing: every input within the house tolerance, method A, largest abs error (tolerance read from src/house/evidence-policy.md) | 3.86e-09 bp | at most 0.01 | pass |
| Repricing: every input within the house tolerance, method B, largest abs error (tolerance read from src/house/evidence-policy.md) | 2.76e-10 bp | at most 0.01 | pass |
| Anchor assumption, ex post: largest gap between the fixings realised for 2026-09-18 and 2026-09-21 and the assumed level (the snapshot holds the fixings published after the as-of date) | 0 bp | at most 0 | pass |
| Paper example p. 102 (cubics go negative): discrete forwards from the capitalisation factors against the printed ones (tolerance is half a unit of the printed precision (0.01 %); 4Y printed 4.40 %, factor implies 5.00 %) | 4.55e-05 abs | at most 5e-05 | pass |
| Paper example p. 102 (cubics go negative): minimum instantaneous forward of method B on (0, 30] | 0.0087 rate | greater than 0 | pass |
| Paper example p. 102 (cubics go negative): every discrete forward recovered by method B | 0 rate | at most 1e-14 | pass |
| Paper example p. 107 (tamer curve): discrete forwards from the capitalisation factors against the printed ones (tolerance is half a unit of the printed precision (0.01 %); printed yields agree with the factors) | 9.09e-06 abs | at most 5e-05 | pass |
| Paper example p. 107 (tamer curve): minimum instantaneous forward of method B on (0, 30] | 0.0366 rate | greater than 0 | pass |
| Paper example p. 107 (tamer curve): every discrete forward recovered by method B | 0 rate | at most 1e-14 | pass |
| Method B, condition (i): midpoint-rule average of f on each interval equals its discrete forward | 6.26e-10 rate | at most 1e-07 | pass |
| Method B, condition (ii): minimum instantaneous forward | 0.039 rate | greater than 0 | pass |
| Method B, condition (iii): largest jump of f across a node | 6.06e-11 rate | at most 1e-06 | pass |
| Method B, condition (iv): intervals that break local monotonicity | 0 count | at most 0 | pass |
| Round trip: flat 4 % curve recovered by method A, max abs zero-rate error (monthly to 30Y) (each deposit and swap is priced off the flat curve and bootstrapped back) | 9.07e-11 bp | at most 1e-06 | pass |
| Round trip: flat 4 % curve recovered by method B, max abs zero-rate error (monthly to 30Y) (each deposit and swap is priced off the flat curve and bootstrapped back) | 5.97e-12 bp | at most 1e-06 | pass |
| Method A: QuantLib against hagan_west.py log-linear bootstrap, max abs discount-factor difference (daily) | 1.93e-13 abs | at most 1e-09 | pass |
| Pricer: hagan_west.py par rate against QuantLib OISRateHelper.impliedQuote on the QuantLib curve | 2.08e-13 bp | at most 1e-06 | pass |
| Method B through QuantLib's swap engine: max abs par-rate error of the inputs (QuantLib values method B's curve on a daily discount-factor grid (log-linear between business days)) | 1.27e-11 bp | at most 0.0001 | pass |
| Method B interpolation: hagan_west.py against QuantLib ConvexMonotoneInterpolation(0, 1), max abs forward difference (same node times and discrete forwards; QuantLib's forcePositive floors only the end points, the paper's collar is not binding here) | 3.47e-13 bp | at most 1e-06 | pass |
| Out-of-sample: median absolute residual of prints the bootstrap never saw, method B (460 prints; residual is close-adjusted print rate minus the curve's par rate; method A: 0.460 bp) | 0.405 bp | at most 1 | pass |

Table: out-of-sample residuals, close-adjusted print rate minus curve par rate (bp).

| Prints | n | A median | A p25 / p75 | A within 1 bp | B median | B p25 / p75 | B within 1 bp |
|---|---:|---:|---:|---:|---:|---:|---:|
| spot start, input tenor | 523 | +0.00 | -0.28 / +0.24 | 91% | +0.00 | -0.28 / +0.25 | 91% |
| spot start, other maturity | 140 | -0.21 | -0.69 / +0.17 | 76% | -0.18 | -0.51 / +0.14 | 86% |
| forward start | 320 | -0.15 | -0.74 / +0.24 | 79% | -0.17 | -0.77 / +0.16 | 78% |
| all | 983 | -0.04 | -0.49 / +0.24 | 85% | -0.04 | -0.45 / +0.21 | 86% |
| held out, last two hours, unadjusted | 49 | -0.76 | -1.27 / +0.02 | 55% | -0.47 | -1.21 / +0.04 | 59% |

Source: out/verify.json, out/oos_residuals.csv, verify.py.

Table: paired comparison on the held-out prints: which method's par rate is closer to each print, by years to the print's expiration.

| Prints | n | B closer | A closer | Sign-test p (two-sided) | Median of abs(A) minus abs(B) (bp) | Within 1 bp: A | Within 1 bp: B |
|---|---:|---:|---:|---:|---:|---:|---:|
| spot start, other maturity, under 1Y | 41 | 30 | 11 | 0.004 | +0.277 | 23 | 34 |
| spot start, other maturity, 1Y to 10Y | 70 | 37 | 33 | 0.720 | +0.003 | 63 | 64 |
| spot start, other maturity, 10Y to the last node | 28 | 11 | 17 | 0.345 | -0.041 | 21 | 23 |
| spot start, other maturity, beyond the last node | 1 | 1 | 0 | 1.000 | +3.668 | 0 | 0 |
| forward start, under 1Y | 27 | 11 | 15 | 0.557 | -0.000 | 23 | 20 |
| forward start, 1Y to 10Y | 172 | 101 | 70 | 0.022 | +0.001 | 138 | 135 |
| forward start, 10Y to the last node | 96 | 60 | 36 | 0.018 | +0.038 | 70 | 71 |
| forward start, beyond the last node | 25 | 18 | 7 | 0.043 | +0.082 | 22 | 22 |
| all held-out prints | 460 | 269 | 189 | <0.001 | +0.007 | 360 | 369 |
| spot start, other maturity, under 1Y, one per expiration date | 22 | 16 | 6 | 0.052 | +0.241 | 14 | 19 |
| spot start, other maturity, under 1Y, raw rates | 41 | 35 | 6 | <0.001 | +0.353 | 11 | 17 |
| spot start, other maturity, 6M to 1Y (exploratory) | 28 | 24 | 4 | <0.001 | +0.804 | 13 | 26 |

Source: out/verify.json, verify.py. A positive median means B's residual is smaller. The one-per-expiration row treats each distinct expiration date as one observation (the median absolute residual of its prints); the raw-rates row uses unadjusted print rates. Eight maturity buckets are tested, so a p-value near 0.05 in one bucket is weak evidence on its own.
