Technical requirements and supporting evidence
This reference preserves the detailed checks behind the task IDs. The roadmap sets current priorities; Next steps explains them in plain language.
Evidence behind each task
| ID | Supporting evidence | Required check or limit |
|---|---|---|
| F01 | Provenance ledger: 53 archive stretches form 27 conservative split groups; physical instrument/installation identity remains unknown. | Eleven metadata controls pass; preserve candidates and conservative grouping separately from verified flight identity. |
| F02 | Nine ESA files form eight groups: recording 187 for development and seven protected confirmation groups. Policy and ledger. | No protected model fitting or outcome inspection before method freeze. Future workers must enforce the reservation; integrity and receiver-only screening are permitted. |
| C01 | Independent force, receiver-derivative, approximation/averaging and local calibration controls. | Completed evidence; preserve failures and indistinguishable cases. Do not rerun. |
| C02 | Feasible scalar calibration: 12/12 known-answer identities and 72/72 matched comparisons pass; 147 certificates and 81 actual optimizer starts. | Earlier failed nonlinear study preserved. Handles near-zero scalar ambiguity without claiming unique calibration; observational uncertainty remains C03's gate. |
| C03 | 126 component checks and two-recording assessment: 144/144 scalar envelopes overlap, none unsupported. | Completed freeze; no retries. Physical/error envelopes remain declared assumptions, not calibrated coverage. |
| C04 | Whole-history necessary constraints: all 27,468 scalar cases overlap across 763 packet groups; one zero gain/offset correction survives all 24 history variants. Complete comparisons on both recordings favor globe. | No full gyro-free trajectory or quantitative gyro attribution is established; uncertainty and withheld prediction follow. Preserve conservative error assumptions and all excluded edges. |
| V01 | Prediction coverage, interface, increment bounds, local, geographic and 54 short-history controls pass. Shared-reference scale passes all eight families/8,192 assessments. The recording-scale study sealed three complete histories, one partial and two unstarted; broad bounds reach about 1,792 km, narrower sensitivity 477 m. | Tighter nonlinear propagation or supported training-feasible regions, adequate numerical treatment and observational covariance assessment remain open. Shared-scale qualification cannot repair nonlinear width. Earlier 10/12-fit evidence stays unresolved. |
| V02 | Saved prediction audit reconciles 18 prior analyses of one recording: 15 originally converged, three unresolved; three matched sets are complete. Specified-field globe retains a much smaller horizontal error, while calibration uncertainty and absolute adequacy remain open. | F01, V01 and qualified solver before new representative predictions. No independent replication or new winner follows from reusing old errors. |
| L01 | Structured covariance passes at 64,080 epochs/27 columns with a 24 MiB cache. Long streaming comparison passes all six controls: 480,000-packet values and all 27 derivatives match unchanged kernels exactly; traced stream working memory stays below 0.7 MiB. Native-control sharding passes all nineteen execution controls and all 24 native compatibility cases. One-thread native arrays match the four-thread reference bit for bit; complete bound reports match exactly. | Preserve earlier exact-pose failure. No active study changes allocation. Software equivalence does not qualify sampling phase, observational covariance, full-record accuracy or predictive adequacy. |
| E01 | Version 1 inventory admits no stretches. The motion diagnostic passes eleven identities. Rate/body controls pass all 48 identities under declared sampling hypotheses, preserving original engineering samples. | Receiver-error assessment and motion-model adequacy precede a revised policy. Actual phase/filtering, geographic/whole-record propagation, mounting and lever arms remain open; protected outcomes stay sealed. |
| E02 | ESA development comparison, then separately frozen confirmation on reserved groups. | C04, V02, L01, E01. Freeze model, uncertainty, adequacy, selection and numerical rules before protected model evaluation. |
| G01 | All 53 raw-gyro comparisons follow 48 independent controls. Median absolute rate differences: 0.88°/h specified-field globe, 2.84°/h zero-field globe, 5.87°/h level flat; all 243 fixed scenarios per stretch are preserved. | No fitted gyro correction or fused-attitude subtraction. Unknown aircraft motion/mounting/processing remain; 27 conservative groups are not independent-flight counts. Preserve the prior technical failure and both freezes; no rerun. |
| U01 | Immutable version 2 retains 53 stretches, 318 latest primaries, 791 history records, 27 conservative groups and gyro baselines; all 326 catalog entries and all history origins are reconciled. Publication and numerical-operation environments remain separate. | Qualified prediction errors, independent confirmation and calibrated uncertainty remain V02/E02/U02 requirements before the final seal. Unknown diagnostic-point ancestry remains explicit; version 1 is preserved. |
| U02 | Calibrate the shape-first decision and subsequent field comparison at the flight-group level. | V02, E02, U01 and separately frozen finite calibration design; include correlated errors, nuisance fitting and minimum over globe candidates. |
| R01 | Freely measured retained background vector and geographic patterns. | C04 and V02; audit geographic/installation rank before fitting. Field plan. |
| R02 | Solar/sidereal precision comparison with freely fitted rate. | R01 plus supported clock/gain/bias precision and longer records; no inference from a fixed-rate candidate. |
| S01 | Wuhan independent positions/eligibility and stationary-consumer physical validation. | Separate dataset-specific qualification; no automatic transfer of archive thresholds. |
| D01 | Exact instrument/processing documentation and defensible calibration alternatives. | Supporting work, not a blanket reason to suspend analysis. |
Earlier plans and their task IDs
| Existing design or unresolved requirement | Backlog home |
|---|---|
| Channel-removal requirements, calibration ambiguity and failed nonlinear matrix | C02's mathematical controls and C03/C04's descriptive assessments are complete. Stronger gyro attribution and uncertainty remain V01/V02 work. Preserve every earlier failure. |
| Error-budget next gate: timing, quantization, averaging, support and joint uncertainty | C03; deterministic count/period and direction-cone bounds are checked, the combined observational model is not. |
| Shared error-model design, bounded uncertainty and earlier incomplete prediction studies | V01/V02, L01 and U02; reuse their controls, not completed start allowances. |
| ESA selection proposal and rate/axis/lever-arm qualification | E01, with F02 reservation enforced before any development or confirmation fit. |
| Raw gyro geographic-prediction companion, residual trends and shared-recorder comparisons | G01 and V02; no fused-attitude substitution or independent-vote counting. |
| Free vector/latitude/rate/altitude follow-ups and solar/sidereal precision | R01/R02 after contribution and prediction qualification. |
| Wuhan independent GNSS processing, consumer physical controls and exact device documentation | S01/D01; retained as separate secondary/supporting work. |
Measurement-contribution and prediction rules
The gyro-free magnitude comparison is a necessary constraint, not a complete motion reconstruction. Agreement does not identify orientation or establish that a full history fits without gyros. Never insert fused attitude or assumed aircraft heading to replace omitted observations. Channel costs need not divide into percentages of gyro and acceleration evidence.
Keep coordinate metric, gravity, calibration, receiver differentiation and timing dependencies explicit. Select support from receiver motion and completeness before inspecting model-dependent residuals. Use every nuisance column in rank checks and retain nonlinear ambiguity.
NASA prediction splits are retrospective development validation: all 53 recordings have informed development. Withhold prediction observations from calibration, covariance selection and initialization corrections. Distinguish within-record prediction from calibration transfer across a verified installation. Report all position components, especially height, rather than rewarding a better horizontal score with an implausible altitude.
Adequacy thresholds and predictive uncertainty must be specified before evaluation. A smaller relative error does not establish an adequate model; uncalibrated criteria remain descriptive. Preserve convergence, physical adequacy, predictive performance and calibrated detection as separate result fields.
ESA development and confirmation boundary
Recording 187 is the development recording. All other currently cataloged ESA flight groups are reserved for confirmation. Inventory, integrity checks and receiver-only eligibility screening are allowed; model fitting and outcome inspection are not allowed until the method and confirmation identities are frozen. Shared files and overlapping flights remain in one group.
The policy is at the flight-group level, not a filename-based claim of independence. Different dates do not prove a different physical instrument or installation. If no interval qualifies, publish that result; do not silently relax the policy or consume confirmation groups for development.
ESA has its own slower-flight eligibility, units, processing assumptions and calibration bounds. Existing Applanix limits do not automatically transfer. Long-record infrastructure must avoid dense whole-record covariance/projection matrices and validate any block/state-space treatment against the dense reference.
Publication, deferred work and execution safeguards
Preserve the raw gyro plot. Its companion must separate precomputed predictions from reconstructions using fitted orientation; neither the common sign nor Group-1 attitude is independent curvature evidence. Report unsupported mounting or aircraft-motion corrections as unknown.
The paper dataset must preserve source/group/selection versions, failed fits, all prediction outcomes and exclusions. Calibrate the complete hierarchy, including selection of the better globe model, field comparison and recording dependence, before attaching significance or promoting production thresholds.
Further NASA recovery, broad horizontal timing searches, large Z-axis grids and passenger-flight geometry optimization are retired or deferred. Precise device identification remains supporting work. Free-field precision, Wuhan expansion and stationary-consumer expansion follow the higher-priority gates.
Each new numerical study needs a separate source/input/environment freeze, finite starts/evaluations, at most four total numerical threads and no adaptive retries. Completed studies never restart or donate unused allowance. Failed qualification blocks dependent observed fits. No automatic corpus-wide campaign or production promotion follows from this roadmap.