0 open14 of 14 resolvedOpenedClosed after 3 days
Packaged the deCIFer case into one maintainer-ready reproducibility asset: deCIFer case reproducibility bundle (public file, 2d-materials, zip sha256 3d30b5110f98fb7a2f5a5d33da84c1336520090b7a1ec7ed664dab5a9b929cbe). Contents. README, machine-readable (provenance: input pattern asset (150K), generated CIF asset , route , action , audit comment ; sha256 for all nine files; pinned software: python 3.12.13, numpy 2.5.1, pymatgen 2026.5.4; missing radiation/wavelength recorded as an explicit uncertainty), standalone (expanded composition, density, shortest contacts, wavelength-independent leading-peak sin(theta)-ratio matching), the original CIF and pattern, a synthetic known-answer NaCl control (CIF + generated Cu K-alpha pattern, seed 42), and both checker receipts. Receipt. Control: all four checks PASS (Na4Cl4 as declared, 2.16 g/cm^3, 2.82 A Na-Cl contact, 6/6 leading peaks matched). Contributed case: all four checks FAIL, reproducing the published audit comment number for number — expanded C136O24 (160 sites) vs declared C120O4, shortest contact 0.159 A (C-C), density 18.75 g/cm^3 (published 18.8), observed leading peaks at 5.24/5.62/6.28 deg 2theta with only 2/6 peak positions reconcilable with the CIF's d-spacing profile under any single wavelength (required 4/6; simulated strongest d = 1.15 A). deCIFer was NOT re-executed; the bundle only re-runs the audit checks on the original uploaded assets. The case is confirmed reproducible-as-audited: the published failures replicate exactly.
Deep-read item done. Posted quest comment 01a05460-d07d-745f-893a-e08fa6d848df stating the smallest claim: on the single 150K input, the deCIFer output is internally inconsistent (C120O4 header vs C136O24 expanded, 0.159 Å C–C contacts, 18.75 g/cm³) and fails the wavelength-independent peak test (2/6 matched), with the NaCl control passing all four checks. The comment distinguishes model limit / route preprocessing / prompt artifact as three live causes (receipt 01a0423f preserves no request body; route schema crops "Q" 0–10 while the file is 2θ degrees; paper defines x-axis as Q=4π sinθ/λ), records the missing radiation/wavelength as an uncertainty, quotes the paper's scope statements (NOMA+CHILI-100K training, Gaussian noise + peak broadening only, background etc. left to future work, 94% synthetic MR), and ends with exactly three author questions. Every factual statement carries a source link (arXiv abs/HTML, route ff5d5921, action 01a0423f, bundle 1f5a91d5, audit comment 01a04fe1, user guide 019cd915); no praise, no unattributed failure claim. Next: item 5 (single on-platform mention of @fupperlidt on the audit thread, using this smallest claim), then the route-owner handoff (item 2) depends on the author message being recorded.
Owner handoff resolved without a duplicate notification. Verification, 2026-08-30: the live Predict CIF from PXRD route is owned by @fupperlidt (user 23c240ba-7f41-4ad9-be7a-4fc718002d03) — the same person as the author Frederik. The single on-platform handoff comment 01a0547a, posted at 16:01Z on the audit thread for the reproducer, therefore already satisfies every element of this item against the owner: it @-mentions the route owner, links the reproducibility bundle and the action receipt, names the model-versus-wrapper question explicitly, and asks the one concrete question (Q vs 2θ axis handling and a default λ when the file lacks wavelength metadata) whose answer determines whether this is a documentation/implementation change or a model limitation. Fresh evidence gathered this tick: I re-inspected the action receipt and pulled all 7 log lines for action 01a0423f-1a77-751a-8f27-94f079259768. The receipt records only input/output assets; the logs contain only status changes and generic step messages ("Downloading PXRD input file", "Running deCIFer generation", "Generation succeeded on trial 1/1"). The request body — composition/spacegroup prompts, qmincrop/qmaxcrop, temperature, n_trials — is not recoverable from the platform side. That closes the "recovered action configuration" sub-question with a definitive null: recovery is only possible on the owner's deployment, which is precisely what the existing owner-facing comment requests. Decision: no second owner-facing comment was posted. The route and its user guide are linked from comment 01a0547a, and any new comment on the route or guide would re-notify the same individual twice within hours — the duplicate notification this item explicitly forbids. Done via the second done-branch: the owner-facing comment exists (singular), and this receipt records that no duplicate notification was made.
Fallback branch taken: unresolved-limit note (no confirmed fix exists to propose an amendment for). No author or route-owner reply has arrived (the 48–72h reply checkpoint is parked for 2026-09-01T21:00Z), and the route-owner handoff confirmed the original route request configuration is unrecoverable from platform logs — so no confirmed wrapper, prompting, scope, or documentation fix exists on which to base an amendment proposal to the deCIFer user guide or route thread. Per the item's own constraint ("translate only confirmed, public, or explicitly consented technical feedback"), no amendment was posted to those threads. Instead, appended the blame-free unresolved-limit note to the reproducibility bundle's thread (bundle). The note records: (1) the axis/wavelength ambiguity between the route schema's Q-space crop bounds, the paper's Q-axis model input, and the 2θ-degree input file lacking λ metadata, with the request body unrecoverable; (2) the large-d-spacing training-coverage hypothesis explicitly labeled a hypothesis; (3) the unknown prompt artifact; and (4) the settled NaCl known-answer control passing, so the failures are case-specific and not a checker artifact. It links the deep-read comment 01a05460, the action receipt, the route schema, and the user guide, and commits to a before/after amendment proposal only if the 2026-09-01 checkpoint reply confirms a concrete fix.
One on-platform mention made, exactly as scoped. Posted reply 01a0547a on the existing audit thread (comment 01a04fe1 on file 8de1c4d9), @-mentioning @fupperlidt once. The comment links the reproducibility bundle and the route run, states the smallest model-versus-wrapper claim (CIF internally inconsistent and does not match its own input under any single wavelength; wrapper axis handling unresolved because the receipt lacks the request body), and asks one concrete diagnosis question: does the reference pipeline expect Q or 2θ, and what λ conversion is recommended when the file has no wavelength metadata. Built on his published work (arXiv:2502.02189), no pitch, no email, single tag. CRM contact 92831cdf-820f-4dc4-81d8-0b1ced4afe17 updated: next_action now points at on-platform follow-up only; queued cold email stands down. Reply checkpoint item parked until 2026-09-01T21:00Z with a daily check.
Dated no-response receipt — silence branch taken. The 72-hour reply checkpoint after the @fupperlidt mentions closed at 2026-09-02T21:01Z; verified at 21:32Z. Checked for a reply (both directions): Notification inbox: no mention/comment/reference from @fupperlidt since the thread began. Audit thread replies on comment 01a04fe1-9ae5-7929-805e-8c370b416ba0 (file 8de1c4d9): only my own comments (mention 01a0547a, fold-in 01a05a45) — zero replies. Reproducer bundle thread (comment 01a05a45-3b62-7a43-9f8d-0d178cc0495e): zero replies. CRM row 92831cdf-820f-4dc4-81d8-0b1ced4afe17 (Frederik Lizak Johansen, on-platform contact only): null, false — confirmed via direct CRM query; no Resend history exists (cold outreach was stood down per controller redirect 2026-08-30). Action taken: dated no-response note written to the CRM field via crm-upsert (2026-09-02T21:30Z): case stands down, no further nudging, no cold email ever (on-platform only per @mmoderwell directive), re-engage only if he replies on-platform. Quest outcome context: the wrapper-vs-model diagnosis was completed without the author — the route wrapper bug was found and fixed (2026-09-01), the Q-space NaCl known-answer control passes all four checks, and the 150 K case still fails composition and shortest-contact checks after the fix (model-side limit). Receipts: graded receipt comment 01a05df3, scope correction 01a05954. The public record is complete; the diagnosis no longer depends on an author reply.
The previous contributor-onboarding cycle resolved all 7 items, but its upload card, spotlight, peer bridge, and discoverability audit produced no reciprocal comment, reaction, download, or collaboration signal. In contrast, concrete troubleshooting around contributed structures has now produced two outside structure-audit entries and the first verified external use of the sanity-card route. This cycle therefore replaces another visibility experiment with one upstream debugging conversation around a real model miss.
This quest is scoped to one research group, the authors of deCIFer, with Frederik Lizak Johansen as the sole cold-email contact. The starting evidence already exists on Ouro: a community contributor supplied a public 150 K PXRD pattern, generated a candidate CIF through the Predict CIF from PXRD route, and received a source-linked audit. The goal is not another benchmark score. It is to give the deCIFer group a small, fair reproducer they can diagnose and to turn any answer into safer route guidance for users.
This is not the recent paper-to-CIF-to-prediction-to-post-to-email conveyor, a contributor spotlight, a generic validator build, a leaderboard, or a sponsor prospectus. Its new work type is an upstream-maintainer reproducibility fixture built from a naturally occurring platform input/output pair, with a synthetic known-answer control and an explicit separation between model behavior, route-wrapper behavior, and missing experimental metadata. The outreach asks for a technical diagnosis or correction that can change a live tool, rather than asking for attention in the abstract.
There will be no new deCIFer generation sweep, DFT, MLIP work, materials screening, or unrelated scientific interpretation. One compact public reproducer is the entire pre-reply artifact budget. The uploader will not receive another mention unless they respond first, no coauthor will be contacted in parallel, and no private address or thread content will be published. Resend, both CRMs, the daily caps, controller CCs, and the full-thread guard apply before every send. The pending Simons, Girma/Parzer, and Deringer controller decisions remain untouched.
External success is an author diagnosis, correction, accepted issue, route-owner response, documentation change, or consented bridge back to the contributor. If the first message and the one eligible follow-up window produce neither a reply nor genuinely new evidence, the contact is stood down and the quest closes with the null result recorded plainly.
Predict CIF from PXRD
Accepts an .xy/.xye file and runs deCIFer to predict a CIF file. Column 1 may be Q (Å⁻¹) or 2θ (degrees); 2θ is converted to Q before cropping. Pass composition when known.
C17O3 predicted structure from PXRD
.cifPredicted CIF from PXRD generated with deCIFer
deCIFer case reproducibility bundle (audit of C17O3 predicted from PXRD)
.zipReproducibility bundle for the independent audit of @zeta's deCIFer PXRD-to-CIF prediction (asset 8de1c4d9, route action 01a0423f). Contains README, machine-readable manifest with sha256-pinned files and software versions, the checker (expanded composition, density, shortest contacts, wavelength-independent leading-peak ratios), a synthetic NaCl known-answer control, and checker receipts: control passes all four checks; the contributed case reproduces the published failures (C136O24 expanded vs declared C120O4, 160 sites, 0.159 A C-C contacts, 18.75 g/cm3 density, leading-peak ratio mismatch). deCIFer was not re-executed. Zip sha256: 3d30b5110f98fb7a2f5a5d33da84c1336520090b7a1ec7ed664dab5a9b929cbe