I use molecular dynamics and docking to compare molecular candidates, and study how data choices and computational conditions affect research findings. Each account connects the question and my decisions to the methods, code and results. The full archive also retains other studies and work I stopped.
83 recordsSelected research
Complete research archive
57 studies and data records, plus 26 hard-problem records. Progress notes remain distinct from published papers.
Studies & data 57
01 Revisiting the Keap1 shortlist with replicated MD Published
RESEARCH RECORD · 01 / 83
Revisiting my Keap1 shortlist changed which compounds I would test first
I revisited the compound that initially looked promising, compared five compounds under consistent conditions at human Keap1, and retained catalpol and geniposide for testing. This account explains the choices, measurements and modelling conditions behind that judgment.
I compared the compounds again using human Keap1, consistent charge models, longer simulations and independent runs, retaining catalpol and geniposide for experimental testing.
The findings compare computational candidates; they do not establish experimental binding or a treatment effect.
02 Designing a health guide for nonvisual reading Accepted
RESEARCH RECORD · 02 / 83
Designing a health guide for nonvisual reading
I wrote a research account of the heading structure and explanations in my Korean food-and-medicine guide. This note covers the design choices and verification limits; the accepted manuscript is awaiting production.
Read the full record03 MAIP-OH: designing information access around a dental visit Published
RESEARCH RECORD · 03 / 83
MAIP-OH: designing information access around a dental visit
A framework for information access around dental care, published in Oral Hygiene Journal in September 2026.
Read the full record04 Missing letters in herbal names: a provider correction and file comparison Public record
RESEARCH RECORD · 04 / 83
Tracing two missing letters through a public herbal dataset
The same two letters were missing from different herbal names. After the provider reported a correction, I compared both files and found 333 changed cells across 270 medicine records, along with remaining issues.
The same two letters were missing from several botanical names. I asked the source institution to review the data, then compared the replacement with the earlier file instead of silently reconstructing names.
05 Reporting a mismatch in the IBLT code Upstream fix · named acknowledgment
RESEARCH RECORD · 05 / 83
Checking how code estimates the difference between two lists
The estimate assumed three distinct cells per entry, but the position-selection code could leave fewer. I supplied reproduction evidence; the original project fixed the code and acknowledged the contribution.
Read the full record06 Questioning the “Korean Hyeam” catalogue attribution Public record
RESEARCH RECORD · 06 / 83
Questioning the “Korean Hyeam” catalogue attribution
I asked the holding library to check the author attribution, received confirmation of the KULINE deletion, and compared dated states across three public records.
Read the full record07 Stated interest and regular Korean-medicine use: a conference poster on older adults with disabilities Presented
RESEARCH RECORD · 07 / 83
Stated interest and regular Korean-medicine use: a conference poster on older adults with disabilities
Presented at the Korean Society of Health Informatics and Statistics Spring Conference on 12 June 2026. The poster examined stated interest, regular Korean-medicine use and unmet healthcare among 3,936 older adults with disabilities; its abstract appears in the conference proceedings.
Read the full record08 Information barriers and unmet care Active
RESEARCH RECORD · 08 / 83
Information barriers and unmet care
A survey analysis and manuscript on information access and unmet care among people with visual impairment. Findings remain private; status dated 3 September 2026.
Read the full record09 Information access as a health-equity issue Active
RESEARCH RECORD · 09 / 83
Information access as a health-equity issue
An invited commentary on understandable information and health equity. Its detailed argument remains private; editorial status dated 3 September 2026.
Read the full record10 Patient experience in Korean-medicine outpatient care Active
RESEARCH RECORD · 10 / 83
Patient experience in Korean-medicine outpatient care
Survey research on musculoskeletal outpatient experience, separating satisfaction from treatment effectiveness. Revised submission confirmed in the receipt preserved on 2 September 2026.
Read the full record11 OPTN–TBK1: comparing docking across protein assemblies Published
RESEARCH RECORD · 11 / 83
OPTN–TBK1: comparing docking across protein assemblies
I matched the OPTN–TBK1 docking conditions and compared two-chain and four-chain inputs. The paper was published in Perspectives on Integrative Medicine on 20 October 2026. The computational data are also public.
Read the full record12 Auditing generative-AI content about Korean medicine Revision received
RESEARCH RECORD · 12 / 83
Auditing generative-AI content about Korean medicine
An audit separating textual readability from warning preservation in AI health-information rewrites. A major revision was resubmitted and acknowledged on October 6, 2026; model-level findings remain unpublished.
Read the full record13 Making East Asian eye-symptom terms easier to navigate With editor
RESEARCH RECORD · 13 / 83
Making East Asian eye-symptom terms easier to navigate
Source-traceable eye terminology research separating word matching from passage classification. Major revision submitted on 5 September 2026; With Editor when checked that day.
Read the full record14 Patient education for people with blindness or low vision Active
RESEARCH RECORD · 14 / 83
Patient education for people with blindness or low vision
A progress record on a scoping review of patient education and health communication for people who are blind or have low vision. It explains the question and preparation while keeping unpublished findings private.
Read the full record15 Turning Donguibogam processing narratives into event data Active
RESEARCH RECORD · 15 / 83
Turning Donguibogam processing narratives into event data
A progress record on representing Donguibogam processing narratives as events while retaining their link to the source. The model, metadata and validation records are in a manuscript; detailed findings and source examples remain private.
Read the full record16 Re-reading the “village doctor” in a 1911 source Technical check
RESEARCH RECORD · 16 / 83
Re-reading the “village doctor” in a 1911 source
A progress note on the label village doctor and the medical labour described beneath it, explaining source criticism, manuscript preparation and the technical-check receipt dated 5 September 2026.
Read the full record17 Reading clinical-trial records by state, time, and anchor Active
RESEARCH RECORD · 17 / 83
Reading clinical-trial records by state, time, and anchor
The state–clock–anchor study and TrialTrace, a locally runnable explorer for 5,629 reconstructed transitions across 400 trial records, with reproducibility materials and an OpenSSF Passing badge.
Read the full record18 Digital collation of a Joseon emergency-medicine text Active
RESEARCH RECORD · 18 / 83
Digital collation of a Joseon emergency-medicine text
A source-traceable collation of Joseon emergency-handbook copies, preserving uncertain and excluded comparisons. In review as of 3 September 2026.
Read the full record19 Reproducibility of a peptide–HLA shortlist Active
RESEARCH RECORD · 19 / 83
Reproducibility of a peptide–HLA shortlist
Comparing variant agreement with exact peptide–HLA combinations in public cell-line data. Software v1.0.0 is public; the manuscript was in editorial processing as of 3 September 2026.
Read the full record20 Direction and magnitude in combination prediction Active
RESEARCH RECORD · 20 / 83
Direction and magnitude in combination prediction
Evaluating response patterns, scale and data splits together. Package public; the tracker updated on 5 September records the manuscript as being with an editor.
Read the full record21 Validation design for an AMR benchmark Active
RESEARCH RECORD · 21 / 83
Validation design for an AMR benchmark
Checking evaluation data and task conditions before interpreting model rankings. The tracker updated on 5 September records the submission; the next editorial step is pending.
Read the full record22 Preference and use of Korean medicine among older adults with disabilities Active
RESEARCH RECORD · 22 / 83
Preference and use of Korean medicine among older adults with disabilities
A within-respondent comparison of preference definitions and main-provider use, kept separate from unmet healthcare. The dated record lists the manuscript as in review on 3 September 2026.
Read the full record23 Can pattern and constitution categories help choose a treatment? Active
RESEARCH RECORD · 23 / 83
Can pattern and constitution categories help choose a treatment?
A literature-based framework for examining personalization claims. The public preprint and later manuscript are different versions; status dated 3 September 2026.
Read the full record24 Before choosing treatment candidates from gene activity and docking scores Active
RESEARCH RECORD · 24 / 83
Before choosing treatment candidates from gene activity and docking scores
A manuscript rebuilding reversal and docking decisions around data and controls, with a bounded reanalysis scope. Publication status dated 3 September 2026.
Read the full record25 When numbers in a trial report do not agree Active
RESEARCH RECORD · 25 / 83
When numbers in a trial report do not agree
Source-linked numerical checks on published randomized-trial reports. Manuscript received 28 August; follow-up external validation paused 5 September. Paper-level judgments remain private.
Read the full record26 The same depression score can contain different symptoms Active
RESEARCH RECORD · 26 / 83
The same depression score can contain different symptoms
A manuscript comparing symptom responses across separate public survey samples. Results and mental-health interpretations remain private; review status dated 3 September 2026.
Read the full record27 Does an ageing clock mean the same thing in later data? Submitted
RESEARCH RECORD · 27 / 83
Does an ageing clock mean the same thing in later data?
A study separating an ageing clock’s development fit from its stability in later data. GeroScience acknowledged the revised submission on 13 September 2026; results remain unpublished.
Read the full record28 Stopping uncertain gene-identifier conversions before they reach analysis On hold
RESEARCH RECORD · 28 / 83
Stopping uncertain gene-identifier conversions before they reach analysis
Rules and code for explicit failure handling in biomedical identifier conversion. Submission ended 4 September 2026; follow-up validation work paused 5 September.
Read the full record29 Checking public-health website accessibility one layer at a time Submission withdrawal confirmed
RESEARCH RECORD · 29 / 83
Checking public-health website accessibility one layer at a time
The audit separates rule verdicts, exposed control states and a keyboard search route, including a correction to the collector’s focus handling. Submission withdrawal was confirmed on 4 September 2026; site-level numbers and conclusions remain private.
Read the full record30 Checking database releases before comparing food composition Under review
RESEARCH RECORD · 30 / 83
Checking database releases before comparing food composition
A manuscript on releases, record matching and comparison sets in two food-composition systems. JSFA technical corrections submitted 4 September; under review as recorded 5 September 2026.
Read the full record31 Has a model learned the ingredients, or recognized the formula? Active
RESEARCH RECORD · 31 / 83
Has a model learned the ingredients, or recognized the formula?
A manuscript comparing formula identity and herb composition in Kampo reports. After resubmission on 1 September, I received a rejection decision on 6 September 2026 (Korea Standard Time). Signals remain private.
Read the full record32 Thinking about olfactory cerebrospinal-fluid outflow in separate segments On hold
RESEARCH RECORD · 32 / 83
Thinking about olfactory cerebrospinal-fluid outflow in separate segments
Hypothesis, manuscript and figures on segment-specific olfactory outflow. Submission ended 4 September; same-hypothesis resubmission stopped 5 September 2026. No mechanism or treatment effect established.
Read the full record33 Examining a TBXAS1 signal alongside the cells that contribute to it Under review
RESEARCH RECORD · 33 / 83
Examining a TBXAS1 signal alongside the cells that contribute to it
TBXAS1 reanalysis and patient-level validation manuscript. Under review at Molecular Vision as recorded 5 September 2026, with a Technical Brief route and upload guidance pending.
Read the full record34 Testing outcome-informed robustness scores with a paired sham-label control Manuscript and code available
RESEARCH RECORD · 34 / 83
Testing outcome-informed robustness scores with a paired sham-label control
A stress test can use held-out answers to choose which microbial features it leaves untouched. The paired sham-label audit holds the model, rows and corruption dose fixed to isolate that label-pairing contribution, with null controls on two public matrices.
Read the full record35 An AI-accessibility review registered and left on hold Paused
RESEARCH RECORD · 35 / 83
An AI-accessibility review registered and left on hold
Registered with OSF on 23 May; public availability confirmed on 7 September. Searching, screening and synthesis remained unstarted in the 5 September ledger.
Read the full record36 PF-02: rebuilding the rules after corpus screening Paused
RESEARCH RECORD · 36 / 83
PF-02: rebuilding the rules after corpus screening
Registered 23 July, amended 26 July. A 60-article corpus screened; outcome coding had not begun at the public amendment.
Read the full record37 What does agreement between language models actually measure? Paused
RESEARCH RECORD · 37 / 83
What does agreement between language models actually measure?
Representation, panel and missingness sensitivity in agreement on synthetic cases. Rework and resubmission paused on 21 July 2026 after five rejections.
Read the full record38 I stopped the accessibility audit when the instrument would not hold still Stopped
RESEARCH RECORD · 38 / 83
I stopped the accessibility audit when the instrument would not hold still
While testing a collector across eight health-information portals, I traced false keyboard-trap findings, incorrect page labels and shifting task classifications. I retained the excluded runs and stopped the project without publishing portal-level judgments.
Read the full record39 Stopping public-data anomaly leads Stopped
RESEARCH RECORD · 39 / 83
Stopping public-data anomaly leads
Prior work, identifiability and falsifying tests guided the stops. No lead was promoted in the ledger preserved on 1 August 2026.
Read the full record40 Why I stopped looking for an extension to a Paeonia–Glycyrrhiza reanalysis Stopped
RESEARCH RECORD · 40 / 83
Why I stopped looking for an extension to a Paeonia–Glycyrrhiza reanalysis
Rework stopped on 3 September after checking study origins and comparable contrasts. The 13 July public package remains; it is not independent validation or efficacy evidence.
Read the full record41IRIS baseline correction and a bounded transfer testUpstream correction verified · bounded transfer test
Research record · 41 / 83
When a sum becomes zero: checking the IRIS response-gene baseline
A discrepancy in IRIS’s released response-gene baseline led to an upstream correction. I checked the function and separately tested a fixed model under explicit external-input assumptions.
Reported a discrepancy in the released response-gene function and checked its upstream correction, then examined missing-input sensitivity in a fixed 1,536-cell transfer test.
42Halofantrine retinal transcriptomic reanalysisPublic code and manuscript · submitted to ReScience C
Research record · 42 / 83
Recomputing the shared-gene intersection in a retinal study
Nine explicit reconstructions left the source gene-overlap figure unresolved. I released code and a manuscript separating those recovery gaps from additional sensitivity analyses.
Examined recovery of S7A and ITPR3 statistics, quantified cross-model sensitivity, and released an executable reanalysis with a manuscript submitted to ReScience C.
43OpenADMET: predicting CYP inhibition from structureMethod report · submission record
Research record · 43 / 83
Predicting CYP inhibition from molecular structure—and documenting the submitted model
I prepared CYP inhibition predictions for 750 blinded compounds and documented the actual submitted models: missing-target handling, structure-aware validation, and the fitted combinations.
Direct-inhibition regression and TDI classification, with a public method report for the submitted models. Submission acknowledgements are distinguished from official evaluation results.
44EvalEval: checking a model release date against its sourceCorrection requested · awaiting response
Research record · 44 / 83
A model card’s release year did not match the official announcement
I compared the Grok 4 release date on an EvalEval card with official sources and filed a correction request. This is a metadata check, not a finding that the model’s benchmark scores are wrong.
Compared the Grok 4 card’s release year with official sources and filed a reproducible correction request. The public issue is awaiting a response.
45Three English words in newspapers decades before public dictionary datesSource comparison
RESEARCH RECORD · 45 / 83
Three English words in newspapers decades before public dictionary dates
Dated newspaper citations for self-enforcing, old-shoe and undercount precede public Merriam-Webster dates; source provenance, senses and the limits of a twelve-case review.
These are dated citation comparisons; novelty and dictionary adoption remain unconfirmed.
4639.1% and 34.2%: comparing two reports of the same trialSource comparison
RESEARCH RECORD · 46 / 83
39.1% and 34.2%: comparing two reports of the same trial
A source comparison of the 39.1% published and 34.2% posted estimates in NCT05212948, with population counts, estimator descriptions and reproducible aggregate arithmetic.
The estimate difference is documented; its cause and the registry analysis cutoff remain unresolved.
47Why 37 earthquake records all say 10 kilometres deepCatalogue analysis
RESEARCH RECORD · 47 / 83
Why 37 earthquake records all say 10 kilometres deep
The 37 exact 10 km depths in a selected Indian Ocean sequence have operator-assigned metadata. Origin records and radius checks explain the scope of this catalogue case.
Metadata provenance is reproducible; a new geological mechanism or statistically established anomaly is not demonstrated.
48Three discovery leads, checked against their sourcesSource audit
RESEARCH RECORD · 48 / 83
Three discovery leads, checked against their sources
Two plant records refer to one sighting. Historical word dates and protein references need different source checks. Three discovery leads are traced through their evidence and unresolved claims.
Source identities, counting units and prior literature are documented; national first records, new Korean first uses and protein classifications remain unresolved.
49An average-flux bug report for EXOFASTv2 and EXOZIPPyCorrections merged in both tools · named independent reporter credit verified
Research record · 49 / 83
Reporting an average-starlight error in EXOFASTv2
Reported the difference between flux at an average distance and the time average of flux, with a reproducible check. Both tools were corrected, and their public pull requests name Sam Yee and me as independent reporters with an ORCID link.
Read the full record50UNK-g: narrowing the identity of an unassigned proteinCandidate hypothesis · non-peer-reviewed summary
Research record · 50 / 83
UNK-g: narrowing the identity of an unassigned protein
A comparison of 7,266 predicted models nominated Q38DP6 for UNK-g. The non-peer-reviewed summary weighs structural and proteomic support against shared observations and uncalibrated scores.
Read the full record51Pneumocystis: a candidate explanation for genetic-code mismatchesPublic research summary · unreviewed hypothesis
Research record · 51 / 83
Pneumocystis: could its genetic instructions be read differently?
A public-data summary of a mitochondrial genetic-code hypothesis, distinguishing recurring record patterns from independent samples and RNA observations from protein-level evidence.
Read the full record52BigWonder: reviewing five KRAS-inhibitor resistance candidatesOngoing research · unvalidated hypotheses
Research record · 52 / 83
KRAS inhibitor resistance: before calling five genes therapeutic targets
I compared public evidence without treating RNA changes as target validation. USP36 was excluded from promotion as a selective new target; four genes remain unvalidated hypotheses. Validated new therapeutic targets: zero.
Read the full record53PyEt: two differences between the equations and their codeReported · awaiting response
Research record · 53 / 83
PyEt: two differences between the equations and their code
I compared two optional PyEt calculations with their cited equations and original sources, reproduced a constant and wind-square discrepancy, and reported both. The effects depend on the input; maintainer confirmation remains pending.
Read the full record54pvlib: one site turned every output into NaNBug · v0.16.2 milestone
Research record · 54 / 83
pvlib: one site turned every output into NaN
I reported NaNs from a single-site WVM input. A contributor reproduced the issue and a project member agreed on the expected behaviour; these replies do not establish an upstream correction or merge.
Read the full record55Medical fragment F-lmfj: comparison and MMFC identification creditIdentification credited by MMFC
Research record · 55 / 83
Identifying a medical fragment against a 1483 printed edition
Comparison of surviving wording and text/commentary sequence identified a Hippocratic passage; MMFC records my independent identification by name.
Read the full record56Witness: a missing-file report and the v2 correctionv2 correction released
Research record · 56 / 83
Witness: from a missing-file report to a corrected package
I reported 24 missing files and outlier and output-path problems. The original project released a separate v2 package; the article distinguishes partial score reproduction from the follow-up correction.
Read the full record57A sideways force that changed direction with speedData and code public
Research record · 57 / 83
Poroelastic contact: speed-induced transverse-force reversal
At fixed material properties and indentation, low and high speeds produced opposite transverse forces. Inputs, pressure fields, and code for 38 numerical cases are public; these are not specimen experiments.
Read the full recordHard problems 26
58 A four-dimensional open convex realization of the Fano plane code Submitted
RESEARCH RECORD · 58 / 83
A four-dimensional open convex realization of the Fano plane code
A four-dimensional open convex realization of the Fano plane code. The 13 September revision of the universal 23-dimensional follow-up adds finite-certificate verification conditions and a bound on representation size.
Read the full record59 A rational fibre for unidentifiability in leak-free mammillary models Submitted
RESEARCH RECORD · 59 / 83
A rational fibre for unidentifiability in leak-free mammillary models
An explicit rational curve preserves the original input–output equation of a leak-free mammillary model. The proof covers all n ≥ 5 and every parameter named in Conjecture 3.18, with exact computational checks alongside it.
Read the full record60 Fragmentation rules and finite-time explosion Submitted
RESEARCH RECORD · 60 / 83
Fragmentation rules and finite-time explosion
Allocation rules change whether the same compartment model explodes. The 13 September revision adds conditions under which explosion persists when alternative allocations occur at a uniformly bounded rate per substrate-containing compartment.
Read the full record61 A 10×10 magic square made from 100 consecutive primes Changes requested
RESEARCH RECORD · 61 / 83
A 10×10 magic square made from 100 consecutive primes
An exact square uses the 100 consecutive primes from 797 through 1493, with all 22 line sums equal to 11,426. The construction finishes a near-solution with exact swaps and a column rearrangement.
Read the full record62 Exactly counting the starting patterns that fill an 8×8 grid Public materials
RESEARCH RECORD · 62 / 83
Exactly counting the starting patterns that fill an 8×8 grid
Exact count, 65 coefficients and reproducibility materials released 5 September. TheoremDB problem still Open when checked 7 September.
Read the full record63 The reflection-invariant cyclic packing optimum Public materials
RESEARCH RECORD · 63 / 83
The reflection-invariant cyclic packing optimum
A checked construction meets an exhaustive upper bound of twelve; unrestricted existence of thirteen remains unresolved.
Read the full record64 Iterates and complete inverse images for OEIS A006165 Public materials
RESEARCH RECORD · 64 / 83
Iterates and complete inverse images for OEIS A006165
A formula for every iterate of OEIS A006165 gives the repeated values, their exact multiplicities and every inverse image. It also explains the 33-input plateau behind a discrepant numerical example.
Read the full record65 Rebuilding P2624’s public proof trees Public record
RESEARCH RECORD · 65 / 83
Rebuilding P2624’s public proof trees
Seven proof trees, 299,903,736 nodes and published component hashes matched in the 20 August 2026 reconstruction.
Read the full record66 Distinct-slope search on the 10×10 grid Public materials
RESEARCH RECORD · 66 / 83
Rechecking how many grid points can be placed without parallel segments
I checked an existing ten-point arrangement and covered every eleven-point candidate through 450 starting cases. The record explains the implementations, replay conditions, prior credit and current review status.
Read the full record67 Fourteen quartets and the seven-taxon minimum Public materials
RESEARCH RECORD · 67 / 83
Fourteen quartets and the seven-taxon minimum
An existing lower bound meets a known construction, with exact partition and tree checks.
Read the full record68 A026124 recurrence and exact boundary terms Novelty disputed
RESEARCH RECORD · 68 / 83
A026124 recurrence and exact boundary terms
A six-term recurrence certificate with its starting range, retaining Mathar’s prior-work assessment.
Read the full record69 An exact certificate for the order-eight recurrence of OEIS A026771 Novelty disputed
RESEARCH RECORD · 69 / 83
An exact certificate for the order-eight recurrence of OEIS A026771
An exact certificate verifies the nine-term recurrence of OEIS A026771 and its starting range. The public note is linked from OEIS and preserves credit for the established generating function and methods.
Read the full record70 Checking a 13-element example, then looking for a larger one Unverified
RESEARCH RECORD · 70 / 83
Checking a 13-element example, then looking for a larger one
Checked a 13-element example and reviewed the original search coverage and proof route. The dated partial checkpoint links to the later public-certificate reconstruction.
Read the full record71 A proposed bound for Pierce remainder chains Preprint · unreviewed
RESEARCH RECORD · 71 / 83
How long can a Pierce remainder chain last? A proposed upper bound
A public preprint proposes an n^(3/10+ε) worst-case upper bound using an integer relation among four remainders. External peer review and formal proof-assistant verification remain pending.
Read the full record72 A frozen protein-structure prediction Models received · abstract submitted
RESEARCH RECORD · 72 / 83
A frozen protein-structure prediction
CASP17 received five T2461 models on 20 August 2026; the methods abstract was submitted on 20 September. Official scores and rankings remain unconfirmed.
Read the full record73 Finite checks in two Erdős attempts Unverified
RESEARCH RECORD · 73 / 83
Finite checks in two Erdős attempts
Triple-sum checks, a separate exhaustive-test design, and an unproved interval-sieve step.
Read the full record74 LANTERN–CAMI: fixing sample correspondence and leaving package validation unfinished Paused
RESEARCH RECORD · 74 / 83
LANTERN–CAMI: fixing sample correspondence and leaving package validation unfinished
Mapping, frozen-code and synthetic-input checks passed. As checked on 7 September, PR 60 remains a draft; container failure left packaging and official submission incomplete.
Read the full record75 KORE Gate-7: assembling the final output without rerunning completed work Paused
RESEARCH RECORD · 75 / 83
KORE Gate-7: assembling the final output without rerunning completed work
A checked-cache finalization procedure: internal assembly completed on 14 July 2026; the scientific result remains sealed.
Read the full record76 Rechecking 47 medical-text candidates exposed a source-boundary error Paused
RESEARCH RECORD · 76 / 83
Rechecking 47 medical-text candidates exposed a source-boundary error
Rechecking 47 candidates from Uibang Yuchwi left 20 inside the stricter source boundaries. An attribution error and a catalogued surviving manuscript changed the research question.
Read the full record77 Reading the existing notes behind clay-tablet candidate scores Paused
RESEARCH RECORD · 77 / 83
Reading the existing notes behind clay-tablet candidate scores
A 75-pair candidate list brought scores together with existing object notes. Line overlap, whole-text similarity and the model score measured different things, while known relationships needed separating from physical joins.
Read the full record78 Keeping image order separate from manuscript foliation Paused
RESEARCH RECORD · 78 / 83
Keeping image order separate from manuscript foliation
An acquisition row kept image index 265 separate from manuscript-side label 131r. The work distinguished Gallica image inventories from selected files and preserved locations for later textual comparison.
Read the full record79 Ulrich u4: verification and earlier work Stopped
RESEARCH RECORD · 79 / 83
Ulrich u4: verification and earlier work
Strengthened v2 published on 21 August. Overlap with a March public result closed the journal path on 2 September 2026.
Read the full record80 A CPU surface pipeline, with no text recovered Stopped
RESEARCH RECORD · 80 / 83
A CPU surface pipeline, with no text recovered
Surface-location and flattening code retained. Zero new ink or characters on 18 August; no comparative performance result because zero pairs were evaluated.
Read the full record81 Dijiang: rebuilding the records behind a progress count Stopped
RESEARCH RECORD · 81 / 83
Dijiang: rebuilding the records behind a progress count
Old progress excluded on 16 July. New input/output and resume controls fixed; a separate job archive retained on 19 July.
Read the full record82 NEAT/NEO-GUARD34: stopping without an independent validation dataset Stopped
RESEARCH RECORD · 82 / 83
NEAT/NEO-GUARD34: stopping without an independent validation dataset
Independent patients, a matching endpoint and a prespecified holdout were required. The routes examined on 21 August 2026 did not supply them; the decision was NO-GO.
Read the full record83 The unfinished final Panax comparison Stopped
RESEARCH RECORD · 83 / 83
The unfinished final Panax comparison
Known-sequence controls and read/assembly checks followed by a reported technical publication NO-GO.
Read the full record