Adaptive methods

Embedded pairs and step-size control, measured off-archive.

The rk run ended on 2026-11-10, and no more compute goes to it. Epoch 2 is frozen and these pages are its final snapshot: the site is no longer rebuilt. The overview site sums up the final findings, and the harness is public for anyone to download and run.

An adaptive method uses an embedded pair to estimate each step's error, and a step-size controller retries any step that misses the tolerance. A record holds one set of weights and a fixed step count, so the run searches pairs in a lane of its own: cycles set aside for this class that write an unscored archive. The lane reports the modeled cycles a method needs to reach an accuracy target.

At a glance

family
An embedded pair on the dyadic lattice (order 3 with an order-2 estimate) and a step-size controller
search
Deterministic enumeration, plus the side-track jobs
question
How many modeled cycles does a method need to reach an accuracy target? (cycles to tolerance)
problems
The validation problems
arithmetic
float64; the Q15 adaptive solver runs one fixed pair
checked by
Nothing: not order-verified and not scored
cost basis
Float64 attempt counts, each attempt priced by the Q15 cost model
sources
adaptive_archive/elites.json, the side-track ledger, benchmark/results.json
points measured
56
one per parameter point in the plan
jobs
4
each closes one open design question
highest rejection share
0.15748
largest over the adaptive.controller_gains points. Arithmetic: float64 only; no Q15 effects are included
side-track code hash
2b9f0b0f9005
the executor and prototypes, at the latest measurement

Work against accuracy across the tolerance sweep

Achieved error against rhs evaluations as the tolerance tightens, one panel per problem on shared axes, both log. Pair: Bogacki-Shampine 3(2).
1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorbattery_2rcbattery_2rc: tolerance 0.001, error 0.00343958 after 28 rhs evaluations, 0 rejected stepsbattery_2rc: tolerance 0.0001, error 0.000673436 after 37 rhs evaluations, 0 rejected stepsbattery_2rc: tolerance 1e-05, error 0.000100945 after 55 rhs evaluations, 0 rejected stepsbattery_2rc: tolerance 1e-06, error 1.22824e-05 after 97 rhs evaluations, 0 rejected stepsbattery_2rc: tolerance 1e-07, error 1.44501e-06 after 199 rhs evaluations, 1 rejected stepsbattery_2rc: tolerance 1e-08, error 1.52719e-07 after 421 rhs evaluations, 2 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorbicycle_lateralbicycle_lateral: tolerance 0.001, error 0.000357318 after 43 rhs evaluations, 0 rejected stepsbicycle_lateral: tolerance 0.0001, error 3.03138e-05 after 67 rhs evaluations, 0 rejected stepsbicycle_lateral: tolerance 1e-05, error 4.76471e-06 after 139 rhs evaluations, 2 rejected stepsbicycle_lateral: tolerance 1e-06, error 5.85663e-07 after 274 rhs evaluations, 2 rejected stepsbicycle_lateral: tolerance 1e-07, error 1.08566e-07 after 571 rhs evaluations, 3 rejected stepsbicycle_lateral: tolerance 1e-08, error 1.50442e-08 after 1195 rhs evaluations, 3 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorbuck_converterbuck_converter: tolerance 0.001, error 0.00151497 after 91 rhs evaluations, 0 rejected stepsbuck_converter: tolerance 0.0001, error 0.000447659 after 181 rhs evaluations, 1 rejected stepsbuck_converter: tolerance 1e-05, error 7.38994e-05 after 376 rhs evaluations, 2 rejected stepsbuck_converter: tolerance 1e-06, error 8.24192e-06 after 787 rhs evaluations, 3 rejected stepsbuck_converter: tolerance 1e-07, error 8.87062e-07 after 1666 rhs evaluations, 3 rejected stepsbuck_converter: tolerance 1e-08, error 9.10891e-08 after 3571 rhs evaluations, 4 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorenzyme_qssaenzyme_qssa: tolerance 0.0001, error 1.7963e-05 after 1162 rhs evaluations, 2 rejected stepsenzyme_qssa: tolerance 1e-05, error 4.17984e-07 after 1165 rhs evaluations, 2 rejected stepsenzyme_qssa: tolerance 0.001, error 0.000638856 after 1165 rhs evaluations, 3 rejected stepsenzyme_qssa: tolerance 1e-06, error 8.16411e-07 after 1645 rhs evaluations, 109 rejected stepsenzyme_qssa: tolerance 1e-07, error 1.67823e-08 after 1957 rhs evaluations, 8 rejected stepsenzyme_qssa: tolerance 1e-08, error 2.38565e-09 after 3301 rhs evaluations, 5 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorglucose_minimalglucose_minimal: tolerance 0.001, error 0.000325491 after 49 rhs evaluations, 0 rejected stepsglucose_minimal: tolerance 0.0001, error 4.03546e-05 after 88 rhs evaluations, 0 rejected stepsglucose_minimal: tolerance 1e-05, error 5.1956e-06 after 181 rhs evaluations, 2 rejected stepsglucose_minimal: tolerance 1e-06, error 5.34465e-07 after 373 rhs evaluations, 2 rejected stepsglucose_minimal: tolerance 1e-07, error 5.58226e-08 after 787 rhs evaluations, 3 rejected stepsglucose_minimal: tolerance 1e-08, error 5.6319e-09 after 1675 rhs evaluations, 3 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorpll_lockpll_lock: tolerance 0.001, error 0.000746124 after 52 rhs evaluations, 0 rejected stepspll_lock: tolerance 0.0001, error 0.000145968 after 85 rhs evaluations, 0 rejected stepspll_lock: tolerance 1e-05, error 2.60402e-05 after 175 rhs evaluations, 2 rejected stepspll_lock: tolerance 1e-06, error 3.16661e-06 after 358 rhs evaluations, 2 rejected stepspll_lock: tolerance 1e-07, error 3.52712e-07 after 757 rhs evaluations, 3 rejected stepspll_lock: tolerance 1e-08, error 3.67843e-08 after 1603 rhs evaluations, 3 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorrobertson_scaledrobertson_scaled: tolerance 0.001, error 0.000597569 after 2353 rhs evaluations, 2 rejected stepsrobertson_scaled: tolerance 0.0001, error 3.26777e-05 after 2380 rhs evaluations, 6 rejected stepsrobertson_scaled: tolerance 1e-05, error 8.16161e-06 after 2425 rhs evaluations, 7 rejected stepsrobertson_scaled: tolerance 1e-06, error 5.66517e-07 after 2536 rhs evaluations, 7 rejected stepsrobertson_scaled: tolerance 1e-07, error 8.94056e-08 after 2857 rhs evaluations, 9 rejected stepsrobertson_scaled: tolerance 1e-08, error 3.81611e-09 after 3883 rhs evaluations, 55 rejected steps1e11e21e31e41e-91e-71e-51e-3rhs evaluationsachieved errorservo_load_stepservo_load_step: tolerance 0.001, error 0.00187753 after 3004 rhs evaluations, 2 rejected stepsservo_load_step: tolerance 0.0001, error 0.000138677 after 3007 rhs evaluations, 3 rejected stepsservo_load_step: tolerance 1e-05, error 3.17866e-05 after 3016 rhs evaluations, 4 rejected stepsservo_load_step: tolerance 1e-06, error 2.84762e-06 after 3028 rhs evaluations, 4 rejected stepsservo_load_step: tolerance 1e-07, error 4.45106e-08 after 3064 rhs evaluations, 5 rejected stepsservo_load_step: tolerance 1e-08, error 6.00325e-09 after 3145 rhs evaluations, 7 rejected steps

Source: side-track job adaptive.suite_sweep, 8 points. Arithmetic: float64 only; no Q15 effects are included.

Every plotted point, in one table (48 rows)
problemtolerancerhs evaluationsachieved errorrejected steps
battery_2rc0.001280.003439580
battery_2rc0.0001370.0006734360
battery_2rc1e-05550.0001009450
battery_2rc1e-06971.22824e-050
battery_2rc1e-071991.44501e-061
battery_2rc1e-084211.52719e-072
bicycle_lateral0.001430.0003573180
bicycle_lateral0.0001673.03138e-050
bicycle_lateral1e-051394.76471e-062
bicycle_lateral1e-062745.85663e-072
bicycle_lateral1e-075711.08566e-073
bicycle_lateral1e-0811951.50442e-083
buck_converter0.001910.001514970
buck_converter0.00011810.0004476591
buck_converter1e-053767.38994e-052
buck_converter1e-067878.24192e-063
buck_converter1e-0716668.87062e-073
buck_converter1e-0835719.10891e-084
enzyme_qssa0.000111621.7963e-052
enzyme_qssa1e-0511654.17984e-072
enzyme_qssa0.00111650.0006388563
enzyme_qssa1e-0616458.16411e-07109
enzyme_qssa1e-0719571.67823e-088
enzyme_qssa1e-0833012.38565e-095
glucose_minimal0.001490.0003254910
glucose_minimal0.0001884.03546e-050
glucose_minimal1e-051815.1956e-062
glucose_minimal1e-063735.34465e-072
glucose_minimal1e-077875.58226e-083
glucose_minimal1e-0816755.6319e-093
pll_lock0.001520.0007461240
pll_lock0.0001850.0001459680
pll_lock1e-051752.60402e-052
pll_lock1e-063583.16661e-062
pll_lock1e-077573.52712e-073
pll_lock1e-0816033.67843e-083
robertson_scaled0.00123530.0005975692
robertson_scaled0.000123803.26777e-056
robertson_scaled1e-0524258.16161e-067
robertson_scaled1e-0625365.66517e-077
robertson_scaled1e-0728578.94056e-089
robertson_scaled1e-0838833.81611e-0955
servo_load_step0.00130040.001877532
servo_load_step0.000130070.0001386773
servo_load_step1e-0530163.17866e-054
servo_load_step1e-0630282.84762e-064
servo_load_step1e-0730644.45106e-085
servo_load_step1e-0831456.00325e-097

Controller gains and rejected steps

Share of attempted steps the controller rejected, by gains alpha (rows) and beta (columns). Every cell prints its share; on a log scale, the further a cell's color is from the page background, the larger the share. n/a means no run finished. Each point runs buck_converter, pll_lock, glucose_minimal at tolerance 1e-06.
betaalpha01/161/81/41/81/43/81/2alpha 1/8, beta 0: rejection share 0.02534; 13 rejected and 500 accepted steps, 1542 rhs evaluations, largest error 8.18309e-060.0253alpha 1/8, beta 1/16: rejection share 0.01793; 13 rejected and 712 accepted steps, 2178 rhs evaluations, largest error 2.7001e-060.0179alpha 1/8, beta 1/8: no run finished (status underflow)n/aalpha 1/8, beta 1/4: no run finished (status underflow)n/aalpha 1/4, beta 0: rejection share 0.01663; 7 rejected and 414 accepted steps, 1266 rhs evaluations, largest error 1.51356e-050.0166alpha 1/4, beta 1/16: rejection share 0.01562; 7 rejected and 441 accepted steps, 1347 rhs evaluations, largest error 1.23288e-050.0156alpha 1/4, beta 1/8: rejection share 0.01386; 7 rejected and 498 accepted steps, 1518 rhs evaluations, largest error 8.24192e-060.0139alpha 1/4, beta 1/4: no run finished (status underflow)n/aalpha 3/8, beta 0: rejection share 0.01018; 4 rejected and 389 accepted steps, 1182 rhs evaluations, largest error 1.87781e-050.0102alpha 3/8, beta 1/16: rejection share 0.00993; 4 rejected and 399 accepted steps, 1212 rhs evaluations, largest error 1.69917e-050.00993alpha 3/8, beta 1/8: rejection share 0.00957; 4 rejected and 414 accepted steps, 1257 rhs evaluations, largest error 1.51866e-050.00957alpha 3/8, beta 1/4: rejection share 0.00797; 4 rejected and 498 accepted steps, 1509 rhs evaluations, largest error 8.4292e-060.00797alpha 1/2, beta 0: rejection share 0.01309; 5 rejected and 377 accepted steps, 1149 rhs evaluations, largest error 2.1067e-050.0131alpha 1/2, beta 1/16: rejection share 0.01295; 5 rejected and 381 accepted steps, 1161 rhs evaluations, largest error 1.94951e-050.0129alpha 1/2, beta 1/8: rejection share 0.01269; 5 rejected and 389 accepted steps, 1185 rhs evaluations, largest error 1.82134e-050.0127alpha 1/2, beta 1/4: rejection share 0.15748; 80 rejected and 428 accepted steps, 1527 rhs evaluations, largest error 1.55843e-050.157

Source: side-track job adaptive.controller_gains, 16 points. Arithmetic: float64 only; no Q15 effects are included.

Every measured cell, in one table (16 rows)
alphabetarejected sharerejected stepsaccepted stepsrhs evaluations
1/800.02534135001542
1/81/160.01793137122178
1/81/8no run finishedn/an/an/a
1/81/4no run finishedn/an/an/a
1/400.0166374141266
1/41/160.0156274411347
1/41/80.0138674981518
1/41/4no run finishedn/an/an/a
3/800.0101843891182
3/81/160.0099343991212
3/81/80.0095744141257
3/81/40.0079744981509
1/200.0130953771149
1/21/160.0129553811161
1/21/80.0126953891185
1/21/40.15748804281527

Embedded pairs on the dyadic lattice

Lattice matrices that meet each condition an order-3 pair with an order-2 estimate needs, by stage count and lattice depth s. Counts are exact; the axis is log, so 0 has no bar. Neighboring conditions with the same count share a bar.
1101001,00010,000100,0003 stages, s ≤ 1matrices, order-2 consistent3 stages, s ≤ 1: 64 matrices, order-2 consistent64order-3 consistent, both solvable03 stages, s ≤ 2matrices, order-2 consistent3 stages, s ≤ 2: 512 matrices, order-2 consistent512order-3 consistent, both solvable3 stages, s ≤ 2: 4 order-3 consistent, both solvable43 stages, s ≤ 3matrices, order-2 consistent3 stages, s ≤ 3: 4,096 matrices, order-2 consistent4,096order-3 consistent, both solvable3 stages, s ≤ 3: 6 order-3 consistent, both solvable63 stages, s ≤ 4matrices, order-2 consistent3 stages, s ≤ 4: 32,768 matrices, order-2 consistent32,768order-3 consistent, both solvable3 stages, s ≤ 4: 24 order-3 consistent, both solvable244 stages, s ≤ 1matrices, order-2 consistent4 stages, s ≤ 1: 4,096 matrices, order-2 consistent4,096order-3 consistent, both solvable4 stages, s ≤ 1: 3,646 order-3 consistent, both solvable3,646

Not counted (space over the census cap of 100,000 matrices): 4 stages, s ≤ 2; 4 stages, s ≤ 3; 4 stages, s ≤ 4.

Source: side-track job adaptive.pair_census, 8 points. Arithmetic: exact over Fractions; no float and no Q15 arithmetic is involved.

Every plotted count, in one table (20 rows)
stages and lattice depthconditionmatrices
3 stages, s ≤ 1matrices64
3 stages, s ≤ 1order-2 consistent64
3 stages, s ≤ 1order-3 consistent0
3 stages, s ≤ 1both solvable0
3 stages, s ≤ 2matrices512
3 stages, s ≤ 2order-2 consistent512
3 stages, s ≤ 2order-3 consistent4
3 stages, s ≤ 2both solvable4
3 stages, s ≤ 3matrices4,096
3 stages, s ≤ 3order-2 consistent4,096
3 stages, s ≤ 3order-3 consistent6
3 stages, s ≤ 3both solvable6
3 stages, s ≤ 4matrices32,768
3 stages, s ≤ 4order-2 consistent32,768
3 stages, s ≤ 4order-3 consistent24
3 stages, s ≤ 4both solvable24
4 stages, s ≤ 1matrices4,096
4 stages, s ≤ 1order-2 consistent4,096
4 stages, s ≤ 1order-3 consistent3,646
4 stages, s ≤ 1both solvable3,646

Measured against real counterparts, at matched accuracy

rk-work/benchmark/results.json has not been written in this work directory, so the matched-accuracy comparison is not stated here. It appears once rk_harness.benchmark has been run against the current code; nothing is inferred in its place. The epoch 1 comparison, run under that epoch's own pin, is on the frozen epoch 1 site.

Lane elites, ranked by cycles to tolerance

The archive fixes a cycle budget and measures the error a method reaches. The lane fixes an accuracy target on the axis-T ladder and measures the modeled cycles a candidate needs to reach it, so neither number converts into the other. Every run is float64. The pinned order-condition checker never saw these coefficients, so none is order-verified.

candidates measured
640,578
under the lane search code hash below
candidates ranked
123
the subset this document put in order; the leading 32 are listed below, the document's cap
source
rk-work/adaptive_archive/elites.json
elite target
0.0009765625 final-state error
lane search code hash
d99e4d2dbe284c58
document written
cycle 17363, 2026-11-10 13:29 CT
The 32 entries plotted here, each at its median modeled cycles at the elite target against its best achieved error (log axis); left is cheaper and lower is more accurate.
330034003500360037001e-81e-71e-61e-51e-41e-3median cycles at the elite targetbest achieved errorrank 1: median 3388 cycles at the elite target, best achieved error 9.01642e-06rank 2: median 3400 cycles at the elite target, best achieved error 8.9257e-05rank 3: median 3416 cycles at the elite target, best achieved error 3.77813e-05rank 4: median 3444 cycles at the elite target, best achieved error 7.54048e-05rank 5: median 3472 cycles at the elite target, best achieved error 2.62007e-05rank 6: median 3472 cycles at the elite target, best achieved error 4.29037e-05rank 7: median 3490 cycles at the elite target, best achieved error 6.27542e-05rank 8: median 3500 cycles at the elite target, best achieved error 8.15011e-06rank 9: median 3528 cycles at the elite target, best achieved error 5.02352e-06rank 10: median 3528 cycles at the elite target, best achieved error 6.46039e-08rank 11: median 3530 cycles at the elite target, best achieved error 5.87563e-05rank 12: median 3556 cycles at the elite target, best achieved error 4.55622e-05rank 13: median 3567 cycles at the elite target, best achieved error 7.82133e-07rank 17: median 3584 cycles at the elite target, best achieved error 2.02167e-06rank 15: median 3584 cycles at the elite target, best achieved error 2.51696e-05rank 14: median 3584 cycles at the elite target, best achieved error 5.45847e-05rank 16: median 3584 cycles at the elite target, best achieved error 7.07468e-06rank 18: median 3590 cycles at the elite target, best achieved error 2.19819e-06rank 19: median 3590 cycles at the elite target, best achieved error 5.34533e-05rank 21: median 3612 cycles at the elite target, best achieved error 2.4653e-05rank 20: median 3612 cycles at the elite target, best achieved error 7.30608e-07rank 22: median 3620 cycles at the elite target, best achieved error 4.09451e-05rank 28: median 3640 cycles at the elite target, best achieved error 0.000101556rank 23: median 3640 cycles at the elite target, best achieved error 1.42391e-06rank 24: median 3640 cycles at the elite target, best achieved error 4.73485e-05rank 25: median 3640 cycles at the elite target, best achieved error 6.33567e-05rank 27: median 3640 cycles at the elite target, best achieved error 6.77336e-07rank 26: median 3640 cycles at the elite target, best achieved error 8.28288e-06rank 29: median 3645 cycles at the elite target, best achieved error 7.09109e-05rank 30: median 3650 cycles at the elite target, best achieved error 4.4025e-05rank 31: median 3660 cycles at the elite target, best achieved error 0.000130083rank 32: median 3668 cycles at the elite target, best achieved error 1.73117e-06

Source: rk-work/adaptive_archive/elites.json. float64 runs, cycles priced on float_trajectory_q15_attempt_cost; not order-verified.

Every plotted mark, in one table (32 rows)
rankmedian cycles at the elite targetbest achieved errorentries
133889.01642e-061
234008.9257e-051
334163.77813e-051
434447.54048e-051
534722.62007e-051
634724.29037e-051
734906.27542e-051
835008.15011e-061
935285.02352e-061
1035286.46039e-081
1135305.87563e-051
1235564.55622e-051
1335677.82133e-071
1735842.02167e-061
1535842.51696e-051
1435845.45847e-051
1635847.07468e-061
1835902.19819e-061
1935905.34533e-051
2136122.4653e-051
2036127.30608e-071
2236204.09451e-051
2836400.0001015561
2336401.42391e-061
2436404.73485e-051
2536406.33567e-051
2736406.77336e-071
2636408.28288e-061
2936457.09109e-051
3036504.4025e-051
3136600.0001300831
3236681.73117e-061
The 32 listed entries, the ranking rule and how the lane prices a cycle

Ranking rule, quoted from the document: lowest median cycles at the elite target across the problems where the status is reached; a candidate that reached the elite target nowhere ranks below every candidate that reached it somewhere; ties broken by the count of targets reached, highest leading, then by record_hash. No clock is read and no value is sampled, so the order is total and two runs over the same records agree.

Rank is the position the document gives under the rule above. Each row shows the candidate's shape, the median modeled cycles it needed at the elite target, how many problems it reached that target on, and the cost basis its cycles were priced on.

rankcandidateorderembedded orderstagessafetyalphabetamedian cycles at targetproblems at targettargets reachedcost basis
16de647f4a17b6f7a32415/163/8033888 of 832 of 32float_trajectory_q15_attempt_cost
28c8104c35a37c56f32415/163/161/1634007 of 830 of 32float_trajectory_q15_attempt_cost
32bc88eb7aca6c5d132415/163/8034168 of 832 of 32float_trajectory_q15_attempt_cost
4cbd48b68efc7b87932415/163/8034448 of 832 of 32float_trajectory_q15_attempt_cost
5b4e320acf5af7bc73243/43/81/834728 of 832 of 32float_trajectory_q15_attempt_cost
69a522fc83fb27c4632415/161/41/1634728 of 832 of 32float_trajectory_q15_attempt_cost
7fb6aafddb0db2e063247/81/8034907 of 830 of 32float_trajectory_q15_attempt_cost
830a681c43ebc6bed32415/163/8035008 of 832 of 32float_trajectory_q15_attempt_cost
954a0975db1e0b8e732415/163/8035288 of 832 of 32float_trajectory_q15_attempt_cost
1032a60583f94dec9b32415/163/8035288 of 832 of 32float_trajectory_q15_attempt_cost
11b1794656aefd5a853247/83/83/1635308 of 832 of 32float_trajectory_q15_attempt_cost
1244647068bbf437053247/83/81/1635568 of 832 of 32float_trajectory_q15_attempt_cost
139dc244a7d721c78b32415/163/8035678 of 832 of 32float_trajectory_q15_attempt_cost
149d937289d9603f8c32415/163/8035848 of 832 of 32float_trajectory_q15_attempt_cost
156ed785c91ebbede83243/43/81/835848 of 832 of 32float_trajectory_q15_attempt_cost
165458f4da3de3187432415/161/41/1635848 of 832 of 32float_trajectory_q15_attempt_cost
173e2f1a6bccd17b0a32415/163/8035848 of 832 of 32float_trajectory_q15_attempt_cost
1847d82adcaa7d02cb3247/83/83/1635908 of 832 of 32float_trajectory_q15_attempt_cost
193b43487937ebcead3247/83/83/1635908 of 832 of 32float_trajectory_q15_attempt_cost
20cafe2b7a70fd50763247/83/16036128 of 832 of 32float_trajectory_q15_attempt_cost
219cd0ec9fcbe980b23247/83/81/1636128 of 832 of 32float_trajectory_q15_attempt_cost
224e063a3c7a6c18993247/83/83/1636208 of 832 of 32float_trajectory_q15_attempt_cost
2311960e2d0409354b32415/163/8036408 of 832 of 32float_trajectory_q15_attempt_cost
24650a40fe4a8ad48132415/161/41/1636408 of 832 of 32float_trajectory_q15_attempt_cost
2570540246b86bdedc3243/43/81/836408 of 832 of 32float_trajectory_q15_attempt_cost
266af1d2d65dcc457832415/163/8036408 of 832 of 32float_trajectory_q15_attempt_cost
27d748277d844255f232415/163/8036408 of 832 of 32float_trajectory_q15_attempt_cost
280a12a65ee5557d1c32415/161/8036407 of 830 of 32float_trajectory_q15_attempt_cost
2932b36bf83cf943863247/83/81/836458 of 832 of 32float_trajectory_q15_attempt_cost
30036936ee657e3b5c3247/83/83/1636508 of 832 of 32float_trajectory_q15_attempt_cost
31a92dd66ca936d30e32415/163/8036608 of 832 of 32float_trajectory_q15_attempt_cost
322403f6837d3ac8fc32415/163/8036688 of 832 of 32float_trajectory_q15_attempt_cost

How this lane prices a cycle

float_trajectory_q15_attempt_cost

direction
upper bound within the model, over a trajectory the model did not produce
excludes
conditional branches, which costmodel prices for no class, so the bit scan is booked at its worst case; derivative evaluations, so FSAL is worth zero cycles here; and the Q31 time register terms
grade
an upper bound in the Q15 cost model over an attempt count measured in float64, which is a mixed number and is labelled as one
includes
the stage work of the propagating formula through the pinned cycle_count, the error estimate at one priced block per nonzero d term scaled by the state count, and one controller update, charged on every attempt whether accepted or rejected
source
costmodel.cycle_count plus costmodel.count_sequence over adaptive_q15.reference_sections, with adaptive.solve_adaptive supplying the attempt count
why_mixed
adaptive_q15.solve_adaptive_q15 is hard-wired to the BS32 pair through A_REPS, B_REPS and D_REPS, so a candidate pair has no Q15 run available. The attempt count comes from the float64 controller and the price of an attempt comes from the Q15 model. A Q15 run of the same pair would reject more attempts than the float run does, so the attempt count is the optimistic half of the figure and the per-attempt price is the pessimistic half

What these numbers are not

Every measured point (56 points)

One table per job and one row per point, from its newest ledger line. A point counts as measured only under the code hash that measured it, so an older line from earlier code is left out; that rule and the retry policy are in the measurement ledger. Plan, job catalogue and invariants: docs/SIDETRACK-AUTOMATION.md in the harness repository. Designs these feed: docs/EPOCH2-DESIGN.md and docs/EPOCH3-DESIGN.md.

adaptive.controller_gains

16 points

Closes: EPOCH2-DESIGN section 4: the dyadic PI gains rest on one measurement, and the classical alpha for a 3(2) pair is 1/3 rather than 1/4

Arithmetic: float64 only; no Q15 effects are included

pointcyclepointsfinishedmax_errorrejection_ratetotal_acceptedtotal_fevalstotal_rejected
a1o2_b0o160332.1067e-050.0130937711495
a1o2_b1o1660331.94951e-050.0129538111615
a1o2_b1o460331.55843e-050.15748428152780
a1o2_b1o860331.82134e-050.0126938911855
a1o4_b0o160331.51356e-050.0166341412667
a1o4_b1o1660331.23288e-050.0156244113477
a1o4_b1o46030n/an/an/an/an/a
a1o4_b1o860338.24192e-060.0138649815187
a1o8_b0o160338.18309e-060.02534500154213
a1o8_b1o1660332.7001e-060.01793712217813
a1o8_b1o46030n/an/an/an/an/a
a1o8_b1o86030n/an/an/an/an/a
a3o8_b0o160331.87781e-050.0101838911824
a3o8_b1o1660331.69917e-050.0099339912124
a3o8_b1o460338.4292e-060.0079749815094
a3o8_b1o860331.51866e-050.0095741412574

adaptive.pair_census

8 points

Closes: EPOCH2-DESIGN section 2: the search-space note assumes a dyadic A admits both an order-p b and an order-(p-1) b_hat with a free parameter left, and that has not been counted

Arithmetic: exact over Fractions; no float and no Q15 arithmetic is involved

pointcycleb_hat_exactb_hat_freeboth_solvabled_forced_zerofsalmatricesspace_sizestatus
st3_s16064640006464ok
st3_s260384512400512512ok
st3_s3602048409660040964096ok
st3_s460102403276824003276832768ok
st4_s1604096409636460040964096ok
st4_s260n/an/an/an/an/a0262144capped
st4_s360n/an/an/an/an/a016777216capped
st4_s460n/an/an/an/an/a01073741824capped

adaptive.q15_estimate_floor

24 points

Closes: EPOCH2-DESIGN section 3: the roughly 2 LSB floor-bias figure for the Q15 error estimate is an estimate, and section 5 needs the scored tolerance ladder to sit above whatever that floor turns out to be

Arithmetic: Q15 fixed point throughout, on the pinned primitives in rk_harness.fixedpoint, with floor multiply; the reference float mode used to reproduce the frozen curve is float64

the bit scan is booked at its worst case and its branches cannot be priced by the pinned cost model, so the sequence figure is an upper bound rather than an exact count. The factor table is held in Q14 because a Q15 table cannot represent a factor of 1.0 at all; the encoding is an open item in EPOCH2-DESIGN section 4.

pointcyclepointsfinishedstatusesflattening_tol_lsbmean_bias_lsb_pooledmean_bias_lsb_totaloverflow_rejectedtables_agree
battery_2rc_s160108attempt_cap, ok512-0.00486615-2.035640false
battery_2rc_s260108attempt_cap, ok512-0.00466504-1.895320false
battery_2rc_s460108attempt_cap, ok512-0.0126933-1.9430false
bicycle_lateral_s160108attempt_cap, ok256-0.0168511-1.594510false
bicycle_lateral_s260108attempt_cap, ok512-0.00646408-1.289540false
bicycle_lateral_s460108attempt_cap, ok512-0.00734325-1.228970false
buck_converter_s160108attempt_cap, ok512-0.0504386-1.81190false
buck_converter_s260108attempt_cap, ok512-0.0101078-1.840620false
buck_converter_s460108attempt_cap, ok512-0.016269-1.813960false
enzyme_qssa_s1120107attempt_cap, ok, step_underflow256-0.200438-1.8909135false
enzyme_qssa_s2120107attempt_cap, ok, step_underflow512-0.12472-1.84144136false
enzyme_qssa_s4120107attempt_cap, ok, step_underflow512-0.0850521-1.81784140false
glucose_minimal_s160107attempt_cap, ok, step_underflow512-0.0394501-1.917370false
glucose_minimal_s260107attempt_cap, ok, step_underflow512-0.0237129-1.938910false
glucose_minimal_s4120107attempt_cap, ok, step_underflow512-0.0134941-1.904170false
pll_lock_s160107attempt_cap, ok, overflow512-0.104358-1.947281false
pll_lock_s260107attempt_cap, ok512-0.00373301-1.895910false
pll_lock_s460108attempt_cap, ok512-0.00638674-1.820130false
robertson_scaled_s1120107attempt_cap, ok, step_underflow128-0.166833-1.8887110false
robertson_scaled_s2120106attempt_cap, ok, overflow, step_underflow64-0.112156-1.8856520false
robertson_scaled_s4120105attempt_cap, ok, overflow, step_underflow32-0.0732931-1.757532false
servo_load_step_s1120107ok, overflow, step_underflow256-1.58357-1.727873false
servo_load_step_s2120107ok, overflow, step_underflown/a-1.52041-1.591913false
servo_load_step_s4120107ok, overflow, step_underflown/a-1.45251-1.481714false

adaptive.suite_sweep

8 points

Closes: EPOCH2-DESIGN section 5: the scored tolerance ladder is unchosen, and the published curve covers 3 of the 8 validation problems

Arithmetic: float64 only; no Q15 effects are included

pointcyclepointsfinishedstatusesfevals_maxfevals_minrejection_rate_max
battery_2rc6066ok421280.0151515
bicycle_lateral6066ok1195430.0434783
buck_converter6066ok3571910.0166667
enzyme_qssa6066ok330111620.198905
glucose_minimal6066ok1675490.0333333
pll_lock6066ok1603520.0344828
robertson_scaled6066ok388323530.0425039
servo_load_step6066ok314530040.00667939
Further evidence, not published here
documentstatewhat it holds
rk-work/validation/axes.jsonnot written yetthe cycles-to-tolerance rows for this class, and the measured floor of the Q15 error estimate that bounds how tight a tolerance can mean anything
rk-work/adaptive_archive/YYYY-MM-DD.jsonlwrittenthe lane's per-day search log: one line per candidate the lane enumerated, ranked against nothing
rk-work/schedule/shares.jsonwrittenthe per-cycle lane log summarised per method class

The traceability rule lists key_findings.json, validation/results.json, benchmark/results.json, the side-track ledger with its artifacts, and the two lane elites documents, adaptive_archive/elites.json and implicit_archive/elites.json. Each document named above sits outside that list and says so in its own schema, so this page names it and reports whether it exists without publishing a number from it. Only the owner can add one to the list.