Qualitative & content coding
Code transcripts or documents consistently with a codebook.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- NVivo · ATLAS.ti · MAXQDA
- Class GUI-need label
G2· GUI-essential
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R08.10.01.02-T01 Code 8 burnout interviews in NVivo against a 6-code codebook
Original instruction
NVivo project burnout.nvp contains 8 imported interview transcripts (input/transcripts/P01-P08.docx). Apply the codebook in input/codebook_burnout.pdf (6 codes with inclusion/exclusion rules and anchor examples) at sentence-to-paragraph granularity; a passage may carry several codes. Create the codes as nodes exactly as named in the codebook. Save the project.
- Platforms
win· Windowsmac· macOS- Applications
- NVivo
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Checker opens the saved .nvp read-only and exports coding references (doc, char start/end, node). Per-code span F1 at character level vs gold, pass if macro-F1>=0.75 and Cohen's kappa>=0.70 on sentence units. NVivo has no scripting API; .qdpx imports appear in the project event log and fail the task.
Proposed partial credit
15% 6 nodes named exactly; 25% all 8 transcripts carry coding; 30% macro-F1>=0.60; 30% F1>=0.75 and kappa>=0.70
R08.10.01.02-T02 Build a hierarchical MAXQDA code system from a codebook spreadsheet
Original instruction
Open empty project caregiving.mx24 in MAXQDA. Recreate the code system in input/codebook.xlsx: 3 parent codes with 12 child codes, each with the listed color and the definition pasted into its code memo. Order children exactly as in the spreadsheet. Save the project.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsdrag· Drag & drop- Work product
WP08· App/system state- Conditions
licensed· Licensed pro app- Proposed difficulty
- 2 / 5
Proposed verifier
Checker reads the .mx24 SQLite code table: names, parent links, sibling order, RGB colors and memo text must match the spreadsheet exactly (whitespace-normalized). A codebook import via 'Import Code System' is allowed only if all colors and memos still match. MAXQDA has no scripting API and the VM has no shell or SQLite tool; .mx24 writes by any process other than MAXQDA are caught by file-access auditing and fail.
Proposed partial credit
30% 15 codes with correct hierarchy; 25% sibling order; 25% colors exact; 20% memos match
R08.10.01.02-T03 Autocode cost mentions in ATLAS.ti and prune false positives
Original instruction
ATLAS.ti project focus_groups.atlproj holds 12 focus-group transcripts. Code every passage where a participant talks about the financial cost of care with code 'Cost' (synonyms in input/cost_terms.txt). Use Text Search/Auto-coding to find candidates, then review each hit and delete quotations that are not about money (e.g. 'cost me my sleep'). Save.
- Platforms
win· Windowsmac· macOS- Applications
- ATLAS.ti
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicknav· Pan/zoom/scrollvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker exports quotations for 'Cost' from the saved project and matches them to 46 gold passages (overlap>=50% of characters). Pass if precision>=0.90 and recall>=0.90; 9 planted figurative uses must be uncoded. ATLAS.ti has no scripting API and the VM has no shell; .atlproj writes by any process other than ATLAS.ti are caught by file-access auditing and fail.
Proposed partial credit
20% code 'Cost' exists with quotations; 30% recall>=0.80; 30% precision>=0.80; 20% all 9 figurative decoys uncoded
R08.10.01.02-T04 Code time spans of job-insecurity talk on an NVivo audio timeline
Original instruction
NVivo project shiftwork.nvp contains audio file interview_03.mp3 (41 min) with no transcript. Listen and code the time ranges where the participant discusses job insecurity to node 'Job insecurity', and the ranges about scheduling to 'Scheduling', using the media timeline selection. Save the project.
- Platforms
win· Windowsmac· macOS- Applications
- NVivo
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
av· Audio/video over timedrag· Drag & dropnav· Pan/zoom/scroll- Work product
WP10· Annotations- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Checker reads media coding references (start/end ms) from the .nvp. Per node temporal IoU vs gold segments; pass if mean IoU>=0.60 and every gold segment longer than 20 s overlaps a coded range. Coding on an added transcript instead of the media timeline scores 0. NVivo has no scripting API and the VM has no shell; .nvp writes by any process other than NVivo are caught by file auditing.
Proposed partial credit
20% both nodes exist with media coding; 40% each gold segment hit (proportional); 40% mean temporal IoU>=0.60
R08.10.01.02-T05 Code image regions in a MAXQDA photovoice project
Original instruction
MAXQDA project photovoice.mx24 has 10 participant photographs. For each photo draw rectangular image segments around objects that act as a 'Barrier' or 'Facilitator' to walking, following input/photovoice_guide.pdf (one segment per object). Save.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropgeom· Precise geometryvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker reads image segment rectangles and codes from the .mx24. Hungarian match to 31 gold boxes; a match needs IoU>=0.5 and the same code. Pass if F1>=0.80. MAXQDA has no scripting API and the VM has no shell or SQLite tool, so segments can only be drawn in MAXQDA; .mx24 writes by other processes are caught by file-access auditing and fail.
Proposed partial credit
20% segments exist on all 10 images; 40% box F1 at IoU>=0.5 regardless of code; 40% F1 with correct code>=0.80
R08.10.01.02-T06 Merge near-duplicate codes and group them in ATLAS.ti
Original instruction
ATLAS.ti project trust_study.atlproj has 47 codes, 9 of which are near-duplicates (e.g. 'Trust', 'trust ', 'Trusting'). Merge them following input/merge_map.csv so no quotation loses coding, then create the 4 code groups listed in input/groups.csv. Save.
- Platforms
win· Windowsmac· macOS- Applications
- ATLAS.ti
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickdrag· Drag & drop- Work product
WP08· App/system state- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker exports codes, quotation links and groups: exactly 38 codes with target names from merge_map; for each target the quotation set equals the union of its sources; 4 groups with exact membership. Deleting instead of merging is detected by lost quotation links. ATLAS.ti has no scripting API and the VM has no shell; .atlproj writes by any process other than ATLAS.ti are caught by file-access auditing and fail.
Proposed partial credit
30% code count 38 with exact names; 40% quotation unions preserved; 30% groups exact
R08.10.01.02-T07 Create NVivo cases with attributes and run a matrix coding query
Original instruction
Project nurses.nvp has 20 coded transcripts. Create one case per participant (classification 'Participant'), import attribute values from input/demographics.xlsx (site, role, years_band), then run a Matrix Coding Query of the 5 theme nodes by site and export the result as output/matrix.xlsx (coding-reference counts).
- Platforms
win· Windowsmac· macOS- Applications
- NVivo
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsmulti· Cross-app- Work product
WP02· Records/table- Conditions
licensed· Licensed pro app- Proposed difficulty
- 4 / 5
Proposed verifier
Checker reads case classifications and attribute values from the .nvp (all 60 values exact) and compares output/matrix.xlsx cell counts to gold computed from the saved coding. Matrix query must be saved in the project. NVivo has no scripting API and the VM has no shell or interpreter, so cases, attributes and the export must come from NVivo; .nvp writes by other processes are caught by file auditing.
Proposed partial credit
25% 20 cases with classification; 30% attributes exact; 20% saved matrix query; 25% exported counts exact
R08.10.01.02-T08 Run MAXQDA intercoder agreement and reconcile disagreements
Original instruction
Two coders coded the same 6 documents: input/coderA.mx24 and input/coderB.mx24. Merge them in MAXQDA, run Intercoder Agreement at segment level (90% overlap, code-occurrence off), export the kappa table to output/ica.xlsx, then produce consensus.mx24 that applies the 14 adjudication decisions in input/adjudication.pdf.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicknav· Pan/zoom/scrollmulti· Cross-app- Work product
WP10· Annotations- Conditions
licensed· Licensed pro appmultiapp· Multi-app- Proposed difficulty
- 5 / 5
Proposed verifier
output/ica.xlsx per-code kappa within 0.01 of gold run with the stated settings. consensus.mx24 segments must equal the gold adjudicated set (doc, code, span overlap>=90%) for all 14 decisions, and other segments unchanged. MAXQDA has no scripting API and the VM has no shell or interpreter, so the kappa table and merge must come from MAXQDA; .mx24 writes by other processes are caught by file-access auditing.
Proposed partial credit
15% projects merged; 30% kappa table matches; 40% 14 adjudications applied (proportional); 15% no collateral changes
R08.10.01.02-T09 Code 400 open-ended survey answers in ATLAS.ti
Original instruction
Import input/survey_q7.xlsx (400 responses to 'What would improve your clinic visit?') into ATLAS.ti as survey documents with respondent IDs. Code each response with exactly one of the 7 codes in input/codebook_q7.pdf ('Other' when none fits). Save project q7.atlproj.
- Platforms
win· Windowsmac· macOS- Applications
- ATLAS.ti
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickkey· Shortcutsvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Checker exports response-ID to code. Accuracy vs gold (two-expert consensus) >=0.80 and Cohen's kappa>=0.70; responses with 0 or 2+ codes count as wrong. Survey import must preserve respondent IDs. ATLAS.ti has no scripting API and the VM has no shell or interpreter; .atlproj writes by any process other than ATLAS.ti are caught by file-access auditing and fail.
Proposed partial credit
15% 400 documents with IDs; 15% every response has one code; 35% accuracy>=0.70; 35% accuracy>=0.80 and kappa>=0.70
R08.10.01.02-T10 Frame-code 30 news articles at article level in NVivo
Original instruction
NVivo project heatwave_news.nvp contains 30 news articles. Code each article (headline and lead paragraph only) to the frames in input/frames.pdf: Conflict, Human interest, Economic, Morality, Responsibility. An article can carry 1-3 frames. Save.
- Platforms
win· Windowsmac· macOS- Applications
- NVivo
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker derives the frame set per article from coding references inside headline+lead spans (coding outside counts as an error). Mean per-article Jaccard vs gold >=0.75. NVivo has no scripting API and the VM has no shell; the .nvp must be saved by NVivo, and writes to it by any other process are caught by file-access auditing and fail.
Proposed partial credit
20% 5 frame nodes; 20% all 30 articles coded; 30% mean Jaccard>=0.6; 30% >=0.75 and no coding outside lead
R08.10.01.02-T11 Attach and link analytic memos in MAXQDA
Original instruction
In MAXQDA project eldercare.mx24, attach in-document memos with the texts in input/memos.txt to the 5 segments identified there by document and quoted first words, then link memos M3, M4 and M5 to the codes named in the file. Save.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsnav· Pan/zoom/scroll- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 2 / 5
Proposed verifier
Checker reads memo table: 5 memos with exact text, anchored within 1 paragraph of the target segment, and memo-code links for M3-M5 exactly as listed. MAXQDA has no scripting API and the VM has no shell or SQLite tool; .mx24 writes by any process other than MAXQDA are caught by file-access auditing and fail.
Proposed partial credit
40% 5 memos with exact text; 30% anchored at right positions; 30% 3 code links
R08.10.01.02-T12 Draw an ATLAS.ti network view of code relations from a theory diagram
Original instruction
ATLAS.ti project selfcare.atlproj has 8 codes. Recreate the model in input/theory.png as a network view named 'Model v1': add the 8 codes and draw the 10 relations with the right relation types ('is cause of', 'contradicts', 'is associated with') and directions. Save.
- Platforms
win· Windowsmac· macOS- Applications
- ATLAS.ti
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropvis· Visual judgmentclick· Point & click- Work product
WP03· Visual layout- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker exports code-code links: the set of (source, relation, target) triples must equal the 10 gold triples (symmetric relations order-free), and network 'Model v1' must contain all 8 nodes. ATLAS.ti has no scripting API and the VM has no shell, so links must be drawn in the network editor; .atlproj writes by other processes are caught by file-access auditing and fail.
Proposed partial credit
20% network exists with 8 nodes; 50% triples correct (proportional); 30% no extra links
R08.10.01.02-T13 Split a broad MAXQDA code into three subcodes by reviewing segments
Original instruction
In project peer_support.mx24, code 'Support' has 64 segments. Create subcodes 'Emotional', 'Instrumental' and 'Informational' under it, read every segment and move it to the right subcode per input/support_defs.pdf (a segment may need two). Leave 'Support' with no direct segments. Save.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Checker maps each original segment (by span) to its final subcode set; accuracy vs gold >=0.80 (set match), no segment lost, parent has 0 direct segments. MAXQDA has no scripting API and the VM has no shell or SQLite tool; .mx24 writes by any process other than MAXQDA are caught by file-access auditing and fail.
Proposed partial credit
15% 3 subcodes under 'Support'; 15% parent emptied without losing spans; 70% per-segment accuracy scaled to 0.80
R08.10.01.02-T14 Correct NVivo sentiment autocoding on 60 audited responses
Original instruction
Project patient_feedback.nvp holds survey.xlsx as a dataset. Run sentiment autocoding on column 'comment', then review the 60 rows listed in input/audit_rows.txt and fix their coding so each sits in exactly one of the four sentiment nodes (very/moderately negative/positive) or none if neutral. Save.
- Platforms
win· Windowsmac· macOS- Applications
- NVivo
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker confirms the sentiment node tree exists, then compares the 60 audited rows' node membership with gold labels; pass if accuracy>=0.90 and no audited row in two sentiment nodes. NVivo has no scripting API and the VM has no shell; the .nvp must be saved by NVivo, and writes to it by any other process are caught by file-access auditing and fail.
Proposed partial credit
20% sentiment autocoding run; 20% no double membership; 60% accuracy on audited rows scaled to 0.90
R08.10.01.02-T15 Export an ATLAS.ti code co-occurrence table and report the top pair
Original instruction
In coded project housing.atlproj, run the Code Co-Occurrence Table for the 5 codes in group 'Barriers', show c-coefficients, export it to output/cooc.xlsx and write the code pair with the highest c-coefficient to output/answer.txt as 'CodeA|CodeB'.
- Platforms
win· Windowsmac· macOS- Applications
- ATLAS.ti
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicknav· Pan/zoom/scroll- Work product
WP12· Answer/decision- Conditions
licensed· Licensed pro app- Proposed difficulty
- 2 / 5
Proposed verifier
output/cooc.xlsx counts and c-coefficients equal gold computed from the project (tolerance 0.001); answer.txt pair exact (order-free). ATLAS.ti has no scripting API and the VM has no shell or interpreter, so the table must be exported by ATLAS.ti; .atlproj writes by other processes are caught by file-access auditing.
Proposed partial credit
50% exported table matches; 50% correct pair
R08.10.01.02-T16 Transcribe and code a 4-minute interview clip in MAXQDA
Original instruction
MAXQDA project rural_health.mx24 has audio clip int07_part2.mp3 (4 min, two speakers). Use MAXQDA's transcription mode to transcribe verbatim with speaker labels 'I:' and 'R:' and a timestamp at each speaker turn, then code the transcript with the 4 codes in input/codebook.pdf. Save.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
av· Audio/video over timetype· Text & formskey· Shortcuts- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 5 / 5
Proposed verifier
Checker exports the transcript: WER<=15% vs gold, speaker labels correct on >=90% of turns, timestamps within 2 s of gold turn starts and linked to the media file; code span F1>=0.70. Pasting externally generated ASR text is flagged by missing per-turn timestamp links. MAXQDA has no scripting API and the VM has no shell, so ASR tools are unavailable; .mx24 writes by other processes are caught by file auditing.
Proposed partial credit
30% WER<=0.15; 20% speakers and timestamps; 50% code F1>=0.70
R08.10.01.02-T17 Build an NVivo framework matrix with linked cell summaries
Original instruction
Project discharge.nvp has 12 coded participant cases and 5 theme nodes. Create framework matrix 'FM1' with cases as rows and themes as columns. In each cell write a 1-3 sentence summary of that participant's coded content and add a see-also link to at least one coded reference supporting it. Save.
- Platforms
win· Windowsmac· macOS- Applications
- NVivo
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsmulti· Cross-app- Work product
WP01· Document- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Structure checked from the .nvp: 12x5 matrix with correct row/column members, every cell with coded content non-empty, each with a see-also link to a reference in that case x theme intersection. Summary faithfulness graded by an LLM rubric against the coded text (stated; 30% weight). NVivo has no scripting API and the VM has no shell; .nvp writes by any process other than NVivo are caught by file-access auditing and fail.
Proposed partial credit
20% matrix structure; 25% all applicable cells filled; 25% valid see-also links; 30% LLM rubric faithfulness
R08.10.01.02-T18 Control pair of T01: produce the burnout coding as REFI-QDA via script
Original instruction
Control pair of T01. With the same 8 transcripts (input/transcripts/*.docx) and codebook_burnout.pdf, produce output/burnout.qdpx (REFI-QDA project) containing the 6 codes and coded selections, using Python or any non-GUI tool. No QDA desktop app is available.
- Platforms
lin· Linuxmac· macOS- Applications
- Python · REFI-QDA
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Checker validates the .qdpx against the REFI-QDA schema, maps selections to character spans and scores with the T01 gold: macro-F1>=0.75 and kappa>=0.70. Same thresholds as T01 so GUI vs CLI results are comparable.
Proposed partial credit
20% schema-valid qdpx; 15% 6 codes; 30% macro-F1>=0.60; 35% F1>=0.75 and kappa>=0.70
R08.10.01.02-T19 Code Spanish-language interviews with an English codebook in ATLAS.ti
Original instruction
ATLAS.ti project migrant_health.atlproj contains 6 Spanish interview transcripts. Apply the 8-code English codebook in input/codebook_en.pdf at sentence level, writing a short English comment on each quotation coded 'Discrimination'. Save.
- Platforms
win· Windowsmac· macOS- Applications
- ATLAS.ti
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker scores sentence-level code assignment vs bilingual gold: macro-F1>=0.70. Every 'Discrimination' quotation must have a non-empty comment. ATLAS.ti has no scripting API and the VM has no shell or interpreter; .atlproj writes by any process other than ATLAS.ti are caught by file-access auditing and fail.
Proposed partial credit
15% 8 codes present; 55% macro-F1 scaled to 0.70; 30% comments on all Discrimination quotations
R08.10.01.02-T20 Quantitize codes and crosstab them by site in MAXQDA
Original instruction
In coded project clinics.mx24 (18 documents), create document variables from code frequencies for codes 'Wait time', 'Staff attitude' and 'Cost' (counts), set document variable 'site' from input/sites.csv, then run Crosstabs of these codes by site and export output/crosstab.xlsx.
- Platforms
win· Windowsmac· macOS- Applications
- MAXQDA
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP02· Records/table- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Checker reads document variables from the .mx24 (54 counts and 18 site values exact) and compares crosstab.xlsx cells with gold computed from saved coding. MAXQDA has no scripting API and the VM has no shell or interpreter, so variables and crosstab must come from MAXQDA; .mx24 writes by other processes are caught by file-access auditing.
Proposed partial credit
35% count variables exact; 25% site variable exact; 40% crosstab export matches
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.