The P2 shortlist has zero passing designs.

Both attempted corrections are rejected. Their worst linear RMS values are 803.0 µm before repair and 803.0 µm after repair, down from 825.0 µm. The repair passes physical constraints, but fails the separate, denser search check of per-field blur, overrun and containment. It contains 9,728 of 16,341 search rays, compared with 9,757 in the input.

The retained input has 825.0 µm RMS in the worst uniform-disk linear prediction. Measured full-beam RMS on the finest independent pupil remains 855.9 µm. 261,053/261,053 rays survive; 155,738 intercepts fit the detector. Detector misses remain in the blur measurement.

Focus spread is 7,059.6 µm; both imaging limits remain 24.2 µm. Finite-field focal length remains 99.143 mm; local angular scales are 105.000/95.000 mm per slope. Package span is 128.1 mm and maximum chief incidence 61.20°.

The selected endpoint is “RFS-120 retained input”. A lower linear prediction does not establish improved finite-beam imaging. Denser and rotated pupils cannot select a replacement; their failures are retained. The 0.25 mm sampled clearance reserve and conservative 250 µm full-aperture departure bound do not establish manufacturing tolerances.

Imaging still fails; no candidate is promoted. The historical shortlist remains 34 runs and zero passes. Flat blockers omit substrates, mounts, windows, baffles and package walls.

The RFS-120 input also maps the wider horizontal sky field onto the detector’s shorter, 8.704 mm side. Its two horizontal edge centroids lie at about ±5.086 mm along that axis, beyond the ±4.352 mm active boundary. Detector roll is absent from this parameter family. Testing its physical orientation is the next useful containment experiment; it requires fresh obstruction checks and will not by itself resolve the large optical blur.

Download the minimax fit and independent checks · Read the calculation and dated reflection

Worst detector RMS changes from 1,114.6 to 855.9 µm on the finest independent pupil. 261,053/261,053 rays survive; 155,738 intercepts fit the detector, compared with 146,951 before fitting. Misses remain in the blur measurement.

Focus spread is 7,059.6 µm; both imaging limits are 24.2 µm. Finite-field focal length changes from 96.493 to 99.143 mm. Local angular scales are 105.000/95.000 mm per slope. Maximum chief incidence is 61.20° and sampled polynomial departure is 201.2 µm.

The fit preserves the 0.25 mm sampled clearance reserve and a conservative 250 µm departure bound over each full aperture. The latter is measured from each trial conic. Response bounds prevent a worse first-order result; they do not require perfect imaging. The unchanged input is retained if no improving endpoint passes the search constraints. Denser and rotated pupils cannot select a replacement.

Imaging still fails; no candidate is promoted. The historical shortlist remains 34 runs and zero passes. Flat blockers omit substrates, mounts, windows, baffles and package walls. Sampled clearance and bounded sag do not establish tolerances or manufacturability.

Download the coupled fit and independent checks · Read the calculation and dated reflection

On the finest independent pupil, worst detector RMS changes from 1,311.2 to 1,114.6 µm. Survival is 261,053/261,053 rays; 146,951 intercepts fit the finite detector, compared with 147,295 in the conic input. Detector misses remain in the blur measurement.

Flat-focus spread is 7,595.7 µm. RMS and focus limits remain 24.2 µm, and full detector containment is required. Finite-field focal length changes from 95.766 to 96.493 mm; local angular scales are 98.774/95.003 mm per input slope. Sampled surface departure is 211.9 µm; maximum chief incidence is 62.79°.

The 0.25 mm sampled clearance reserve is a numerical constraint, not a manufacturing tolerance. The departure bound limits surface sag; it does not establish acceptable slopes, tooling or alignment. The unchanged conic input remains available if fitted endpoints fail. Denser and rotated pupils cannot select a replacement.

Imaging still fails; no candidate is promoted. The historical shortlist remains 34 runs and zero passes. Flat blockers omit substrates, mounts, windows, baffles and package walls. Continuous clearance and tolerance-supported hardware feasibility remain open.

Download the fit and independent checks · Read the calculation and dated reflection

On the finest independent pupil, worst detector RMS is 1,311.2 µm, compared with 6,245.7 µm at the RFS-117 starting seed and 1,765.6 µm for the RFS-113 imaging reference. Survival is 261,053/261,053 rays; 147,295 intercepts fit the finite detector.

Flat-focus spread is 8,980.1 µm. RMS and focus limits remain 24.2 µm, and full detector containment is still required. Finite-field focal length is 95.766 mm; local angular scales are 95.087/95.000 mm per input slope. Package span is 119.9 mm. The starting seed's focal length was 102.341 mm; maximum chief incidence increases from 54.06° to 62.97°.

The search preserves a 0.25 mm sampled clearance reserve and keeps the original seed available if a fitted endpoint fails. Worst-margin points from a 1,257-ray search disk refine the constraints. Larger and rotated pupils are checked after selection. All three raw blur endpoints fail the reserve or scale screen; a bounded feasibility repair retains the lower blur and passes those screens. All independent pupils retain every ray and the reserve. The numerical reserve is not a tolerance or manufacturing allowance.

Imaging still fails; no candidate is promoted. The historical shortlist remains 34 runs and zero passes. This bounded fit uses flat mirror blockers and the active detector rectangle. Substrates, mounts, windows, baffles, package walls, continuous clearance and tolerance-supported manufacturing remain unclosed.

Download the fit and independent checks · Read the calculation and dated reflection

261,053/261,053 rays survive on the finest independent pupil. Both 5,025-ray disks, including the 17° rotation, also retain every ray at all 13 fields. Minimum sampled mirror-edge margin is 0.250 mm; minimum active obstacle margin on the finest disk is 11.885 mm. The projected detector shadow is zero. These margins use the existing flat-blocker model and do not establish continuous or hardware clearance.

Worst RMS is 6,245.7 µm versus 1,765.6 µm for RFS-113; flat-focus spread is 16,562.3 µm. Both limits remain 24.2 µm. Only 98,141/261,053 intercepts fit the finite detector. The clearance seed passes the selection screens while RMS, focus spread and detector containment still fail.

Finite-field focal length is 102.341 mm; local angular scales are 104.998/96.736 mm per input slope. Package span is 136.3 mm and optics mass proxy is 0.375 kg. The 0.25 mm search reserve is not a manufacturing or alignment allowance.

One of two starts reaches sampled feasibility; the other remains infeasible after three refinement rounds. A successful solver termination with negative margin is still rejected. The 1,257-ray refinement disk belongs to the search; the larger and rotated checks above do not influence selection.

RFS-113 remains the imaging reference; no candidate is promoted. The RFS-118 fit above uses this geometry with hard clearance, package and both scale constraints active. The historical shortlist remains 34 runs and zero passes. Substrates, mounts, window, baffles, package walls, tolerances and manufacturing remain unclosed.

Download constraints, refinement and independent checks · Read the calculation and dated reflection

The selected trial loses 1,351 of 261,053 rays on the finest disk. Worst detector RMS is 1,261.2 µm, compared with 1,765.6 µm for RFS-113 and a 24.2 µm limit. Its changed surviving population and 94.136 mm finite-field focal length prevent calling this an acceptable imaging improvement.

Two starts stop immediately with invalid full-beam mirror or detector intersections; the two fitted endpoints also fail clearance and plate-scale screens. Local angular scales are 101.979/89.323 mm per input slope, and flat-focus spread is 8,211.5 µm. RMS includes detector misses. The 0.25 mm fitting margin is a soft numerical preference, not a tolerance budget or continuous clearance proof.

RFS-113 remains the reference; no candidate is promoted. The historical shortlist remains 34 runs and zero passes. The RFS-117 calculation above now establishes a clear conic seed with hard ray-retention constraints and worst-margin refinement. Imaging improvement remains open. Substrates, mounts, window, baffles, package-wall clearance and tolerances remain unclosed.

Download endpoints, signed margins and dense checks · Read the calculation and dated reflection

Eight starts remove the old fitted freeforms and vary conic mirrors, folds, spacings, optical powers and detector distance. The objective constrains all four pupil-to-image derivatives and both angular plate scales at the centre and corners. 0 of eight endpoints passes the full-pupil selection screens. Six low-cost first-order fits leave a field with fewer than two surviving rays.

The selected measurable endpoint has 3,359.5 µm worst detector RMS versus 1,765.6 µm for RFS-113; the requirement is 24.2 µm. It loses 9,791 of 261,053 rays. Its two local angular scales are 82.27 and 77.79 mm per input slope, also outside 95–105 mm. This seed is excluded and does not replace RFS-113.

The 317-ray disk selects the endpoint; denser and independently rotated pupils test it afterwards. Increased detector containment cannot count as an improvement when rays are lost and focal length changes. The 139.2 mm package envelope and zero polynomial departure do not establish hardware feasibility.

The RFS-116 calculation above includes finite-pupil aperture and obstruction margins during fitting; clearance still fails. Independent production-ray differences verify the selected analytic matrices within the 1e-6 bound across 13 fields and three step sizes. Imaging, clearance and tolerance closure remain open; the historical shortlist still has zero passing designs.

Download all eight endpoints, ray-loss counts and dense checks · Read the calculation and dated reflection

With the RFS-113 mirrors and detector fixed, the on-axis X- and Y-input fans reach their local linear least-blur planes at offsets of 5.49 and -9.51 mm. Full-pupil on-axis RMS is 1,193.2 µm; even its best individual parallel plane retains 1,179.9 µm, against a 24.2 µm imaging limit.

An independent calculation differentiates exact surface sag, intersections and reflection. Production ray differences verify the result at three step sizes in all 13 fields. The two local angular-to-image scales are 105.68 and 89.35 mm per unit input slope. These describe an anisotropic chief-ray response at the detector; they do not replace the historical 96.857 mm finite-field centroid-fit EFL.

All 261,053 rays survive the finest disk; an independently rotated disk retains all 65,325 rays. Detector misses remain in the blur calculation. Linear and nonlinear components are correlated; their RMS values cannot be added in quadrature without the signed cross term supplied in the JSON. Individual refocused planes are diagnostics, with downstream clearance untested.

The RFS-115 experiment above uses both directions of the analytic pupil-response matrix to construct conic seeds and checks nonlinear blur, dense clearance and detector containment. Imaging still fails. No prescription is promoted; hardware clearance, tolerances and manufacturing remain unclosed.

Download matrices, all ray fans and dense-pupil results · Read the calculation and dated reflection

A 3,922-evaluation fit couples all 24 declared local even-X coefficients with 15 geometry, power and physical-plane focus parameters. Apertures and normalization radii stay fixed. Trials are screened for sampled ray loss, projected detector shadow, focal length, package size and 250 µm sampled freeform departure. Finite-detector extent and flat-plane focus spread contribute to the score.

At 20,081 pupil rays per field, worst RMS changes from 2,881.9 to 1,765.6 µm and detector containment from 129,207 to 145,120/261,053. The selected trial retains 261,053/261,053 rays. Both the 24.2 µm imaging limits and full detector containment remain unmet.

All selected independent clearance/EFL/package/sag screens pass: true. The fit uses only the 317-ray pupil; larger and rotated pupils do not select the result. RMS includes every surviving intercept, including detector misses. The mapped baseline loses aperture rays, so its blur uses a slightly different surviving population.

This preliminary fit is excluded from acceptance. Sampled freeform departure is 250.0 µm and maximum chief incidence 54.22°. Hardware obstructions, tolerances and manufacturing remain unclosed. The RFS-114 diagnosis above measures ray fans and relay power; repeated local fits around this seed risk small improvements without viable imaging.

Download fitted surfaces, detector frames and dense results · Read the calculation and dated reflection

The nonplanar builder applied eight stored freeform values to mixed odd/even-X terms through degree three. The declared convention assigns them to eight even-X terms through degree four. Independent monomial sag, slope, symmetry and reflection checks verify an explicit mapping for future fits. Even-X describes each mirror's local surface; it does not imply symmetry of the whole three-dimensional train.

The existing nonzero coefficients were fitted using the historical direct terms. Reassigning those numbers changes the physical surfaces. On the RFS-111 geometry, finest-pupil worst RMS rises from 2,623.7 to 2,881.9 µm. Detector containment falls from 131,850 to 129,207 intercepts, and 8 rays fail the aperture checks. The detector entrance shadow remains clear.

Checks at 317 rays per field and a rotated 5,025-ray pupil show the same direction. Blur includes every surviving detector intercept, including active-area misses. Ray losses change the sampled population and disqualify the remapped trial. Imaging remains far outside the 24.2 µm limits; the historical shortlist stays at 34 measured runs and zero passes.

Correction to the earlier description: RFS-106/107 fitted mixed terms despite being described as plane-symmetric freeforms. The zero-freeform RFS-104/105 starts are unaffected. Historical prescriptions and acceptance records remain reproducible. The RFS-113 fit above uses explicit mapped coefficients with coupled geometry, dense clearance checks and finite detector containment.

This is a preliminary software and optical-model result. Substrates, mounts, detector housing, windows, baffles, package walls and tolerances remain unclosed.

Download polynomial checks, physical surfaces and dense results · Read the calculation and dated reflection

A 1,338-evaluation search couples mirror folds, spacings, optical powers, conics and physical-plane detector focus while preserving the implemented freeform surfaces and mirror apertures. The selected trial retains all 261,053 rays in the finest check and has zero projected detector-shadow overlap in all 13 fields. Larger and independently rotated pupils also retain every sampled ray.

On the same finest pupil, worst RMS blur falls from 3,094.1 to 2,623.7 µm. Containment rises from 128,625 to 131,850/261,053 intercepts on the 10.880 × 8.704 mm detector. EFL is 98.052 mm, the package spans 117.9 mm, and the optics mass proxy is 0.375 kg.

The 2,623.7 µm worst RMS and 7,909.0 µm parallel-plane focus spread still exceed the 24.2 µm limits. Roughly half the intercepts miss the active detector. Blur includes those misses; the image is not cropped to improve the result. This trial is excluded from optical acceptance. The historical shortlist remains 34 measured runs and zero passing designs.

Clearance covers sampled rays, flat mirror disks and the active detector rectangle. Mirror substrates, mounts, detector housing, windows, baffles and package walls remain outside this model. This is a preliminary geometry correction, with no tolerance or hardware validation.

Source-model finding: the nonplanar builder applies eight values directly to the full XY polynomial; the main builder maps them to declared indices. Those conventions produce different surfaces. This comparison preserves the implemented nonplanar arrays. The RFS-112 check above confirms the defect and measures the effect of remapping.

Download trial surfaces, detector frames and measured results · Read the calculation and dated reflection

All 38 paths missed by the reserved pupil cross the detector rectangle before reaching the primary mirror. The same source geometry, apertures, freeforms and physical detector plane were checked at six pupil samplings. The finest trace loses 2,657/261,053 rays (1.018%), all at this entrance-beam obstruction. Every sampled ray reaches the intended sequential apertures.

An independent projection of the detector rectangle onto the circular entrance pupil gives 1.027% mean overlap across the 13 fields. Its polygon membership matches the traced obstruction for every sampled ray. The 17-ray rings miss this shadow completely.

These are preliminary counts from zero-thickness mirror disks and the active detector rectangle. Mirror sag, backs, mounts, detector housing, window, baffle and package walls are absent from the obstruction model. This is not full-pupil clearance or a tolerance result. The imaging and detector-footprint failures remain; no candidate is added or promoted.

The RFS-111 correction above follows this finding with coupled geometry and power, dense search and independent pupil checks. Removing this shadow alone would not resolve the large measured blur.

Download intersections, sampling results and source hashes · Read the calculation and dated reflection

The current prescription retains all 221 reserved rays, but a denser 317-ray pupil at each of 13 fields retains only 4,083/4,121. All rays reach the sequential apertures; the existing obstruction model blocks 38 paths. The sparse-pupil result does not establish full-pupil clearance. Dense-pupil RMS describes surviving rays only.

At the corrected physical detector plane, worst RMS is 3661.9 µm. Even giving each field its own best parallel plane leaves 3617.8 µm worst residual RMS, far above the 24.2 µm limit. Parallel defocus alone cannot explain the imaging failure.

A 17-case scan perturbs three mirror curvatures, three conics and two spacings in both directions, with fixed apertures and freeforms. Its lowest-field-RMS screened trial reduces worst RMS to 3639.5 µm, retains only 105/221 detector intercepts, and worsens focus spread. Its denser pupil also loses rays. This is diagnostic evidence; no shortlist candidate is added.

The RFS-110 localization above identifies the detector in the incoming beam as the blocker and measures denser pupil samples. Repeating isolated focus or conic adjustments is unlikely to resolve the measured gap. Window, baffle, package-wall and manufacturing closure remain open.

Download all field measurements and source hashes · Read the response table and dated reflection