More control. Provenance intact.

1. Open the project or batch and review every folder, sample identity, experiment list, gradients, sequence, preparation and temperature. Explicitly enable the intended ramps; another sample’s indices are not transferable signal selections.
2. Distinguish calculating the current ramp, one resin’s enabled ramps and the enabled batch. Preflight identifies each blocker; it neither excludes a questionable ramp nor copies another calibration to start calculation.
3. In results, use the selector to visit every sample and ramp. The global summary remains; the plots and lower signal table belong only to the selected result. A ramp without a result is labelled accordingly rather than displaying the previous ramp’s result.
4. Sort columns to inspect numerical values, including sigma; sorting does not recalculate. Save and reopen the batch to check identities, states and revisions. Retain failed, cancelled and unexecuted entries: completed does not mean scientifically accepted.

Scientific boundary: An average of different ramps is descriptive. Different diffusion times or conditions can reflect evolution or time dependence; they do not make the ramps independent replicates.

1. Open the assistant in the source ramp. Distinguish the selected peak from the suggested one and inspect every acquired point with its sign; a 23-experiment ramp must display all 23, including negative and excluded points. No valid logarithmic points does not mean no signal.
2. Use Propose points and, where appropriate, change the selection or number of levels. Compare full and selected fits, b coverage, residuals, R², conditional error and exclusion reasons. Do not select solely for the highest R² or lower requirements to accept a reference.
3. Confirm identity, Rh and conditions, then explicitly archive the reference. The TTMS archive is an independent review: it runs no inversion and creates no molecular map. The reviewed KK2-52 79–101 case retained 15 of 23 points; this is not a universal mask for other ramps.
4. If another ramp of the same sample, folder and processing has present but unfit TTMS, review the suggestion to inspect a previous ramp or a compatible archived source. A copy in another folder is not the same source. Open Reuse fTTMS between ramps, choose that source and explicitly select destinations. If no verifiable accepted source exists, return to the source ramp: an admissible but unarchived diagnostic is insufficient.
5. Permission reuses only fTTMS, with provenance and compatibility checks. It copies neither signal, point mask, temperature nor acquisition timing. For the empirical KK2 recipe, each destination uses its own TE: fT = 293/T and total factor = fTTMS × fT, once. Other recipes retain their declared thermal law.
6. Save the decision in a new revision before recalculation. You can withdraw transferred permission, manually review the local reference or explicitly choose to work without a standard; no alternative is selected silently. Check each destination’s status when running a batch as well.

Scientific boundary: Transfer requires a human decision and compatible preparation and protocol. High R², recovered Rh and file integrity establish neither identity, purity, absence of convection nor experimental validity of the factor.

1. Inspect distribution endpoints and open support review/proposal. Distinguish displayed axis limits from the D bounds actually used in inversion.
2. The proposal uses measured D before TTMS and temperature and multiple observations of each selected signal, not just the standard. Review regression of ln(I) against b, its useful points and at least four distinct b levels, and its comparison with the signed nonlinear fit. Check estimated D against configured minimum and maximum D; a single value or the TTMS value does not establish the support needed for every signal.
3. Explicitly confirm new bounds and calculate another review if justified. Changing signals, b values or extraction invalidates an earlier proposal. Retain the previous inversion and compare predictions, residuals, tails and endpoint maxima between reviews.
4. If signal still accumulates at an endpoint, retain the warning. Support and boundary review can show endpoint contributions without another fit; it neither identifies a finite D nor authorizes invented components.

Scientific boundary: Proposed support is an operational diagnostic, not a guarantee of complete tails, population count, noise compatibility or identifiability. Do not relax criteria to close a distribution.

1. Read the sample, experiments and review shown beside the detail. Compare the global mean with that ramp’s mean, not an individual signal or M of a central D.
2. Distinguish the saved numerical threshold for moments/sigma from the map’s visual threshold. Unfiltered sigma can use different support from global-table sigma. Between-signal standard deviation is neither measurement uncertainty nor a confidence interval.
3. C1, C2… are numbered within each signal and ramp; they are not the second sample or second global-group row. Review support, fit, convergence, chemical role and declared response before interpreting each component’s moments.
4. For —, read the retained reason: no admissible component, censored support, an unarchived result type and undocumented molecular response are different states. Available D descriptors can be retained without forcing Mn, Mw or dispersity when their physical basis is unsupported.

Scientific boundary: Do not change an unknown response declaration to mass weighting merely to fill C2. A descriptive or Gaussian component is not an established species; linear-polymer calibration does not validate absolute masses of stars or brushes.

1. Open Calculate ramps and 2D average and review included ramps and the algorithm before execution. Retain per-ramp results and the average review; do not replace individual b values with mean b.
2. Check Original, Model and unmodelled signal, the side projection and diagnostics separate from the 2D field. The lateral Gaussian count is neither the dictionary-atom count nor a molecular count.
3. Adjust framing, acquired spectra/attenuation series, gain and visual threshold. Fine threshold control changes visibility only, not inversion, projection or saved moments. A small standard is not silently converted to polymer mass.
4. Select Add this average to report and wait for confirmation that the figure is ready. This snapshot preserves the view in Spanish/English. Export inserts the retained PNG without reinversion or another field fit; later display changes do not replace it unless you explicitly add the new view.

Scientific boundary: If the capture is missing, belongs to another review or fails integrity checks, the problem is reported: another ramp’s map is not borrowed, and no replacement is recalculated to fill the report.

1. Review each signal’s state on completion and open its diagnosis: compatible with the convection model, not detected in the check and inconclusive are different conclusions.
2. Check signed observations, phase, noise, coverage, sequence moments and stability between compatible ramps. If compatible extraction or necessary information is missing, retain inconclusive; do not interpret it as no convection.
3. Physical correction or a compatible control acquisition requires explicit review and creates another revision. Automatic diagnosis does not itself rescale D, Rh, masses or the complete map.

Scientific boundary: High R² and fit-capable TTMS do not rule out convection. A software check does not establish absolute absence or replace an experimental control acquisition.

1. Explicitly select archived results and the sample/ramp/batch scope. For an average, first add its compatible exact figure. Review the displayed revision, recipe, standards and factors before export.
2. Choose Español or English and generate summary and full reports from that review. Check identical values in their shared tables; signals are not refitted and maps are not recalculated to complete a document.
3. Check consecutively numbered tables and figures and their references in the text. The added average belongs to its sample and follows the first table; batches retain each sample’s figure identity and order.
4. Read reasons for unavailable values and scientific limitations. Retain project/batch, documents and receipts together. For different settings create a new review; do not relabel an older report or edit its data to force agreement.

Scientific boundary: Professional formatting, file integrity and identical tables are documentary checks. They do not certify convergence, molecular response, identifiability or experimental validity.

1. From DOSY/MwOSY open DRECT / DREAM · experimental AI. The original route and New · DRECT / DREAM B23 are different options; B23 uses exactly 23 actual b values and signed observations without interpolation.
2. Review local installation, model paths/hashes and preparation before execution. Weights are not downloaded automatically. Calculate, save and reopen research reviews separately; accepted maps and molecular results are not replaced.
3. B23 + physical fit is another explicit variant, not a new published native DRECT/DREAM output. Compare with TRAIn using the same observations, grid, extraction and factors; inspect endpoint maxima, false components and weak recovery, not residuals alone.
4. B23 does not pass every independent check and is not the default method. TabPFN V4 support remains a separate research study: it did not establish the required incremental advantage over the no-TabPFN control and is not a production-UI feature.

Scientific boundary: Cross-application numerical parity is not experimental validation. These variants do not automatically establish species, absolute masses, Mn/Mw/dispersity or calibrated uncertainty; failed checks and previous reviews remain preserved.