Upload 5 files

.gitattributes

@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+Data-Remediation-Log.pdf filter=lfs diff=lfs merge=lfs -text
+Dataset-Datasheet_Becker-exemplar.pdf filter=lfs diff=lfs merge=lfs -text

DATA_REMEDIATION_LOG.md

@@ -0,0 +1,28 @@
+# LimitWise working file: data remediation log (Week 3)
+
+CORVANA BANK: MODELING TEAM
+
+LIMITWISE MODELING EXTRACT: DATA REMEDIATION LOG
+
+*Prepared by the modeling team for the LimitWise program file, in response to the Week 2 review. One row per change the team has now made to the extract before the first model is trained, and one row per finding the team chose to document rather than change. Two terms are used the same way in every row. The working file is every column the team holds after remediation: the fields as loaded, the fields added beside them, and the outcome. The model inputs are the subset of those columns the model will be trained on. A field can be in the working file and not a model input. Two rules apply to every row. The raw field is never overwritten. A field held out of the model inputs stays in the working file for measurement. The team takes the label off the account, not the account out of the file. Rows leave only under a written rule, and one row of this log (R-8) is the only place that happens.*
+
+**Extract:** 30,000 rows and 25 columns as loaded (23 fields, the account identifier, and the outcome). 29,995 rows and 40 columns in the working file after remediation (15 columns added, 5 rows removed under R-8). Those 40 columns are 27 model inputs, 12 fields held out of the inputs, and the outcome column. Outcome: default_payment_next_month, value of interest 1 ("default"), 22.1 percent of 29,995 (6,635 accounts, unchanged at one decimal from 22.1 percent of 30,000, and not the book rate).
+
+| Row | Field or fields | What the Week 2 review found | Operation performed | Reason | Options considered and not taken | Before and after | Where it lands | Notebook |
+|---|---|---|---|---|---|---|---|---|
+| R-1 | The extract as drawn (all rows) | 22.1 percent default, 6,636 of 30,000, against a book rate the dictionary puts near 3 percent, about 7.4 times. The dictionary says the sample was drawn to include enough defaults to learn from and does not say how | Leave as is and document. Reweight or recalibrate for enrichment: recorded as a modeling-stage decision, not performed | The enrichment is a fact the file can show. The method is open: no sampling memo or extraction query is held. If the draw was random within each outcome, a defaulting account entered the extract about 9.2 times as often as a non-defaulting one, and that is the figure a reweighting would use. Without the method, reweighting fixes the mix but not a biased draw. No threshold, loss criterion, or calibration claim is made against 22.1 percent | Reweight non-defaulters now, about nine to one. Calibrate now on a book-like holdout. Set a threshold on the extract's rate | No change to the data. 22.1 percent, 3 percent, 7.4 times, 9.2 times, printed | Not a field. A figure (the enrichment) and an open item (the method) recorded for the modeling and evaluation stages | 11.1 |
+| R-2 | PAY_0, PAY_2 to PAY_6 (six fields) | Values 0 and -2 appear in every repayment-status field, are not documented, and 0 is the most common value in every month. 25,939 records, 86.5 percent, hold one of them somewhere. Both default below the documented paid-duly code in PAY_0: 0 at 12.8 percent, -2 at 13.2, -1 at 16.8. The six fields are the six strongest predictors in the file | Recode undocumented values into a named level: six new fields, STATUS_0 and STATUS_2 to STATUS_6, beside the raw fields. -1 and 1 to 9 keep their documented meaning in words. 0 and -2 both become "no delinquency recorded, code undefined" | The meaning of 0 and -2 is open and stays open. The level name says what is known and nothing more: the documented delinquency codes are 1 to 9, and these accounts carry none. What the accounts did is visible in the documented amounts: in PAY_0, code 0 has a median prior statement of 48,261 and a median payment of 3,000, and code -2 a median prior statement of 1,099 with 34.5 percent having no prior balance at all. A reason for an account action must be writable in plain words, and "your status was 0" is not. The raw codes stay so the evaluation stage can test merged against separated under the Chief AI Officer's 5 percent tripwire | Define the codes from their behavior (asserts a meaning nobody confirmed). Drop the six fields from the inputs (the six strongest predictors would go). Leave the raw codes as the inputs (reasons could not be written) | Six raw fields with 11 values each, then six recoded fields beside them with 10 levels. "No delinquency recorded, code undefined" holds 17,496 records in STATUS_0 and 21,181 in STATUS_6. Rows 30,000 before and after | Raw in the working file, not model inputs. Recoded copies are the model inputs | 11.2 |
+| R-3 | BILL_AMT1 to BILL_AMT6, PAY_AMT1 to PAY_AMT6 | The status codes cannot be defined, but the statement balances and the payments are documented, most recent first | Derive a feature from documented fields: five new fields, PAID_1 to PAID_5, each payment set against the prior month's statement balance under one rule: no balance (prior statement zero or negative), paid in full, paid part, paid nothing | One rule for every month, so the evaluation stage can test whether the model leans on a status code whose rule changed (R-4). Every level is a plain word a reason could use. The earliest month's payment (PAY_AMT6) is set against a statement the extract does not hold, so there is no sixth field, and the team says so rather than inventing one. Nothing here defines a code: the table that sets PAID_1 beside PAY_0 shows counts only | A sixth field with an assumed prior statement. Same-month pairing (payment against the balance of the same month, which is not what the documentation describes). Derive nothing and rely on the codes | Five fields added. PAID_1: paid part 17,782, paid in full 6,823, no balance 3,175, paid nothing 2,220. Of the 14,737 accounts with code 0 in PAY_0, 13,956 paid part of the prior statement. Rows unchanged | PAID_1 to PAID_5: model inputs, added. The twelve raw amount fields: model inputs, unchanged | 11.3 |
+| R-4 | PAY_0, PAY_2 to PAY_6 | Value 1 appears 3,688 times in PAY_0 and then 28, 4, 2, 0, and 0 in the older months, while value 2 holds between 2,626 and 3,927 in every month. The fields are named PAY_0 and PAY_2 to PAY_6. There is no PAY_1 | Leave as is and document, as a coding inconsistency across months | The most recent month's status was computed by a different rule from the older months, and the team cannot say what that rule was or why the names skip 1. Six fields sharing a name and not a rule is a risk the evaluation stage tests with the PAID fields beside them (R-3), which carry one rule for all months | Use only PAY_0 and discard the five older months (five months of documented amounts would go with them). Recompute the older months (no rule is known to recompute by) | No change to the data. The counts by month are printed | Not a field. An open item recorded for the evaluation stage (whether one rule computed all six months) | 11.4 |
+| R-5 | EDUCATION, MARRIAGE | EDUCATION holds three codes the dictionary does not document, 0 (14 records), 5 (280), and 6 (51), 345 in all. MARRIAGE holds one, 0 (54). Their default rates are a fraction of the file's, 0.0, 6.4, 15.7, and 9.3 percent, on groups too small to trust | Recode undocumented values into a named level: EDUCATION_MODEL and MARRIAGE_MODEL beside the raw fields, with 0, 5, and 6 in EDUCATION and 0 in MARRIAGE as one level, "unknown" | The dictionary states that all four codes appear in the data and are not documented in the codebook, and nothing else the team holds documents them. What any of the four records is open, and the recode does not close it. A named "unknown" is honest, and it stops the model from learning that one undocumented code is safe. The raw codes stay so the evaluation stage can test the merge against the codes kept separate under the Chief AI Officer's one-point tripwire | Treat the codes as missing and impute (nothing is missing, and imputing a demographic field is never done). Merge into "other" (asserts they belong with the documented "other" level). Drop the records | EDUCATION 7 codes, then EDUCATION_MODEL 5 levels, "unknown" 345 records at 7.5 percent. MARRIAGE 4 codes, then MARRIAGE_MODEL 4 levels, "unknown" 54 records at 9.3 percent. Rows unchanged | Raw EDUCATION and MARRIAGE: in the working file, not model inputs. EDUCATION_MODEL: model input. MARRIAGE_MODEL: in the working file, not a model input (R-10) | 11.5 |
+| R-6 | default_payment_next_month | The dictionary defines the label as default on payment in the following statement cycle, one month, and says nothing operational | Define the label window: leave as recorded and document. The documented window is one statement cycle. The operational definition is open | The label sets the base rate and the meaning of every error. A definition the team supplied would be a guess dressed as documentation. Whether the label's definition matches the loss forecast's definition of default is open, and the mismatch, if any, is stated rather than resolved | Redefine the label to a standard such as days past due (invents a definition). Relabel from the payment fields (replaces the recorded outcome with a derived one) | No change. 6,636 default (22.1 percent), 23,364 no default (77.9 percent) as loaded | The outcome column, unchanged. Not a model input | 11.6 |
+| R-7 | LIMIT_BAL, BILL_AMT1 to BILL_AMT6, PAY_AMT1 to PAY_AMT6 | The unit of the amounts is not stated. 1,930 accounts (6.4 percent) carry a negative balance in some month, smallest -339,603. The largest statement balance, 1,664,089, and the largest payment, 1,684,259, each exceed the largest credit limit, 1,000,000 | Leave as is and document. No cap, no trim, no sign correction | The unit is open, and the amounts are used as relative values. The dictionary says balances may be negative. 3,931 accounts (13.1 percent) carry a balance above their own limit in at least one month, so above-limit balances are common, not a handful of outliers. Capping real extremes would hide exactly the accounts the model most needs to see, and correcting a value nobody has confirmed is wrong would assert something | Cap or winsorize the amounts. Set negative balances to zero. Drop the accounts above the limit | No change to the data. The maxima, the counts, and the unit line are printed | All thirteen fields: model inputs, unchanged. The unit and the extremes: open items | 11.7 |
+| R-8 | All rows (ID used by the rule) | 35 patterns repeat in every field except ID, 70 rows. 30 pairs show no activity in any month. 5 pairs carry six months of identical non-zero balances or payments | Drop rows under a stated rule: identical in every field except ID, with a non-zero balance or payment in at least one month. Keep the lower ID, remove the other | Two accounts matching in every one of 24 fields including six months of activity is one account extracted twice under two identifiers. Two unused cards with the same limit and application details can coincide, so the 30 inactive pairs stay. Removal happens before the split rule (R-12) so no copy of an account can sit on both sides. This is the only row removal in the log. Which of the two identifiers is correct is open. ID is the account identifier, never a model input, kept to trace rows | Keep all 70 rows (five accounts would be doubled across the split). Drop all 70 (removes 60 rows the team cannot show are duplicates). Drop the pattern rather than one row of it | 30,000 rows before, 29,995 after, 5 removed. The 30 inactive pairs, 60 rows, stay. Rate of "default" 22.1 percent after (6,635 of 29,995) | Every other row stays. ID: in the working file, not a model input | 11.8 |
+| R-9 | BILL_AMT1, LIMIT_BAL, PAY_AMT1, BILL_AMT2 | The six statement-balance fields correlate at 0.80 to 0.95 (section 8a), so the top reason for an account can shift between near-identical fields at every retrain | Derive a feature from documented fields: UTIL_RECENT = BILL_AMT1 / LIMIT_BAL, and PAY_RATIO_RECENT = PAY_AMT1 / BILL_AMT2 where BILL_AMT2 is above zero. Both from the most recent month. Raw fields kept as inputs | Both are quantities a person would reason with, from documented fields, in the same unstated unit as their sources, so both are relative. Where there was no prior balance to pay, the ratio is left empty rather than invented, and PAID_1 (R-3) already records "no balance" for every such account. The evaluation stage tests reason stability across retrains under the 10 percent tripwire | Drop five of the six balance fields (loses month-to-month information). Fill the empty ratio with a value (asserts a payment behavior where there was nothing to pay). Derive nothing | Two fields added. UTIL_RECENT median 0.314, 2,113 accounts above 1. PAY_RATIO_RECENT median 0.078, 6,820 accounts at 1 or more, empty for 3,175 accounts (10.6 percent), the only empty values in the working file | UTIL_RECENT and PAY_RATIO_RECENT: model inputs, added. The raw fields: model inputs, unchanged | 11.9 |
+| R-10 | SEX, AGE, MARRIAGE_MODEL (and the raw MARRIAGE, already out under R-5). EDUCATION_MODEL stays | Sex, age, and marital status are prohibited bases under the Equal Credit Opportunity Act and Regulation B. The four demographic fields are among the weakest predictors in the file, 0.034 to 0.074 (section 8b). The documentation never says whose demographics a supplementary-card account carries | Hold protected attributes out of the model inputs, keep them in the working file for measurement. EDUCATION_MODEL stays a model input, with monitoring, pending the evaluation-stage test | You cannot measure what you deleted. The Chief AI Officer's watch list measures false alarm and miss rates by sex, age band, marital status, and education, and tests whether the remaining inputs recover each field. The Chief AI Officer declined an untested drop in Week 2 and ruled that the decision is made on the test (with and without, worth more than half a point of area under the curve, AUC). This row records the recommendation and the test. The team takes the label off the account, not the account out of the file: every cardholder stays with all of their other fields, and the model is simply never told their sex, age, or marital status. One limitation sits on all four fields: whose demographics a multi-card account carries is open | Delete the four fields from the working file. Keep all four as inputs and test later. Drop EDUCATION now, ahead of the ruling | 30 model inputs, then 27. All four demographic fields and their recoded copies still in the working file. Rows 29,995 before and after | SEX, AGE, MARRIAGE, MARRIAGE_MODEL: in the working file, not model inputs. EDUCATION_MODEL: model input, monitored. The supplementary-card question: an open item on every measurement by these fields | 11.10 |
+| R-11 | Credit bureau data (no field) | Nothing in the extract or its documentation refers to bureau data | Leave as is and document: no join has been made | No bureau field is present and the dictionary refers to none. Whether bureau data could join these accounts is open. Any later join is a new extract and a new datasheet | Join bureau data now (nothing to join). Assume it is absent from the book (unknown) | No change. The field list as loaded is printed | Not a field. An open item recorded in the datasheet | 11.11 |
+| R-12 | Train/test split | The six months are columns, not rows, so there is no time index across rows. After R-8 each account appears once | Fix the train/test split: random by row, stratified on the outcome, 80 percent training and 20 percent held back as the test set, fixed seed, after the R-8 deduplication. Recorded now, executed at the modeling stage | The split decides whether the evaluation numbers are honest. With no time index and each account once, random stratified is the defensible rule, and the datasheet says so. The held-back test set is never used for training | Split by time (no time index across rows). Split by account (each account appears once after R-8, so a random split of rows already keeps every account on one side). Resample before the split (leaks) | Expected sizes: 23,996 training rows (about 5,308 defaults), 5,999 in the held-back test set (about 1,327 defaults). For the smallest measurement groups: age 65 and over, 110 records, about 22 in the held-back set. MARRIAGE_MODEL "unknown," 54 records, about 11. EDUCATION_MODEL "other," 123 records, about 25. Not executed this week | Not a field. A rule recorded for the modeling stage | 11.12 |
+
+**Open items carried in the datasheet, not resolved by any row above:** how the 30,000 accounts were drawn and the book default rate for the period they cover, what repayment-status values 0 and -2 record in each of the six fields, whether one rule computed all six months and why there is no PAY_1, what EDUCATION 0, 5, and 6 and MARRIAGE 0 record, what "defaulted on payment" means operationally and whether it matches the loss forecast's definition, the unit of the amounts and how a balance or a payment can exceed the largest credit limit, whether the 30 inactive repeated patterns are one account or two and which identifier is correct for the five pairs treated as one, whose demographics a supplementary-card account carries, and whether credit bureau data can be joined. Section 12 of the notebook prints the same nine.
+
+*End of log. Questions about any row go to the modeling team through the AI Governance Office.*

Data-Remediation-Log.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02755df49d778530f838b6e2f55541844be2ccc4187ee497de3fc40e206290df
+size 133430

Dataset-Datasheet_Becker-exemplar.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2da9847e2732720f0fd291f058866966babb39e7cf2221be2a7aac9a6525dc3
+size 105795

README.md

@@ -1,3 +1,202 @@
 ---
 license: cc-by-4.0
+pretty_name: Corvana Bank LimitWise working file (remediated)
+language:
+  - en
+task_categories:
+  - tabular-classification
+tags:
+  - ai-governance
+  - datasheet
+  - datasheets-for-datasets
+  - credit-risk
+  - fairness
+  - calibration
+size_categories:
+  - 10K<n<100K
 ---
+*The working file behind LimitWise v2: the extract after the Week 3 remediation, as the AI Governance Practicum (Track 1) uses it. The organization, the system, and the people are fictional. The data is real and public: Default of Credit Card Clients, UCI Machine Learning Repository, id 350 (Yeh, 2016), CC BY 4.0.*
+
+**Where the data is.** This repository holds the documents, not a copy of the data. The file itself, `corvana_modeling_extract_remediated.csv` (29,995 rows, 40 columns, MD5 `f79f13a5119f78572e9f6cc3243ea6ae`), is in [drdavidprivacy/practicum-case-packs](https://huggingface.co/datasets/drdavidprivacy/practicum-case-packs), beside its own README with the row and column arithmetic. Every notebook in the course loads it from there by URL: [corvana_modeling_extract_remediated.csv](https://huggingface.co/datasets/drdavidprivacy/practicum-case-packs/resolve/main/corvana_modeling_extract_remediated.csv).
+
+**What is in this repository**
+
+- `README.md`: this page, the dataset datasheet.
+- `DATA_REMEDIATION_LOG.md` and `Data-Remediation-Log.pdf`: the modeling team's log of every change made to the extract before the first model was trained (Week 3), the evidence behind sections 2 and 4 of the datasheet.
+- `Dataset-Datasheet_Becker-exemplar.pdf`: the datasheet below as a PDF.
+
+**Used by:** the model [drdavidprivacy/limitwise-model](https://huggingface.co/drdavidprivacy/limitwise-model) and the Space [drdavidprivacy/limitwise](https://huggingface.co/spaces/drdavidprivacy/limitwise).
+
+**Datasheet below:** the AI Governance Lead's exemplar (Tom Becker), the answer key to the Week 3 Dataset Datasheet, in the structure of Gebru et al., "Datasheets for Datasets." Students replace it with their own; see "Make this your own" at the end.
+
+---
+
+# Dataset Datasheet: the LimitWise working file (Corvana Bank)
+
+**Analyst:** Tom Becker, AI Governance Lead (exemplar)
+**Organization:** Corvana Bank
+**System:** LimitWise
+**Dataset:** LimitWise modeling extract (30,000 open card accounts with six months of history)
+
+
+## Section 1: Motivation
+
+**Objective:** Record why the dataset exists. A dataset assembled for one purpose carries that purpose's assumptions into every later use, so the reason it was created is the first thing a reader needs.
+
+*Completed by the AI Governance Lead. Read it, but do not rewrite it.*
+
+### For what purpose was this dataset assembled, and what gap was it meant to fill?
+
+The extract was assembled to build and test LimitWise, the early warning model the bank is considering for its card book: a monthly score of each open account's probability of payment default in the next statement cycle. The gap it was meant to fill is the one the program charter names. The bureau scores and the incumbent behavioral scorecard are backward-looking at the moment that matters most, the months before a customer stops paying, when bills climb and payments shrink, and the credit cycle two years ago ran charge-offs well past plan before anyone called.
+
+### Who assembled it, and on whose behalf?
+
+The modeling team drew the extract from the bank's card servicing records: the application fields, the credit limit, and six months of statement, payment, and repayment-status history for each account. They did so on behalf of the LimitWise program, whose executive sponsor is the Chief Credit Officer, who owns the loss forecast and wants the model feeding it, and whose governance is owned by the Chief AI Officer, who is the approver of record.
+
+### Under what program or budget was it assembled?
+
+Under the LimitWise program, management's response to the board's direction for earlier warning after the credit cycle two years ago. The extract and the prototype were produced within that program. No separate funding was attached to the data, and none is recorded in the program file.
+
+## Section 2: Composition
+
+**Objective:** Describe what is in the dataset as facts with numbers: what one record is, how many there are, what the outcome is, what is missing, and which groups of people the data identifies. This is the section your Week 2 EDA Observation Log was written to feed.
+
+### What does one record represent, and how many records and fields are there?
+
+One record is one open card account: its credit limit, four fields carried over from the application record (sex, education, marital status, age), six months of repayment status, statement balance, and payment amount, most recent first, and a label for default in the following statement cycle. There are 30,000 records and 25 columns as loaded, 23 fields, the account identifier, and the outcome (notebook sections 1 and 2). After the team's remediation the working file has 29,995 rows and 40 columns, 15 of them added beside their sources: 27 model inputs, 12 fields held out of the inputs, and the outcome column (section 12).
+
+### Is the dataset everything, or a sample? If a sample, what is it a sample of, and how was it drawn?
+
+A sample. The data dictionary states that the sample was deliberately drawn to include enough defaults to learn from and that the book default rate runs near 3 percent per cycle, so the extract's rate is about 7.4 times the book's (section 9, check 5). How the 30,000 accounts were drawn is not documented, no sampling memo or extraction query is held, and whether the draw was random within each outcome is open (modeling team response, answer 1). If it was, a defaulting account entered the extract about 9.2 times as often as a non-defaulting one (section 11.1). It is a sample of the bank's open card accounts, and nothing in the program file says in what proportion it stands to the book.
+
+### What is the outcome field, how is it recorded, and what share of records carry the value of interest?
+
+default_payment_next_month, recorded as 1 (default) and 0 (no default), shown in the notebook in words as default and no default. Default, the value of interest, means the account defaulted on payment in the following statement cycle, which the dictionary states is one month. What "defaulted on payment" means operationally, and whether it matches the loss forecast's definition, is not documented. 6,636 of 30,000 records are default, 22.1 percent (section 3), and 6,635 of 29,995 after the team's one row removal, still 22.1 percent (section 12). That is the extract's rate, not the bank's.
+
+### Is any information missing or unrecorded, and where?
+
+No field as loaded has a missing or unrecorded value in any of the 30,000 records (section 4). Two things read like gaps and are not. Coded fields hold values the dictionary does not document, EDUCATION 0, 5, and 6 (345 records), MARRIAGE 0 (54), and repayment-status values 0 and -2 in 25,939 records (section 9, checks 2 and 3), and those are recorded values, not gaps. And one derived field in the working file, PAY_RATIO_RECENT, is empty by construction for 3,175 accounts whose prior statement balance was zero or negative, with PAID_1 recording "no balance" for every one of them (section 11.9). Those are the only empty values in the working file.
+
+### Does the dataset identify groups of people by a protected characteristic, directly or through other fields?
+
+Yes, directly. SEX (1 = male, 11,888 records at 24.2 percent default, and 2 = female, 18,112 at 20.8), AGE (21 to 79, with under 25 at 27.2 percent on 2,685 records and 65 and over at 25.5 on 110, too small to trust), MARRIAGE, and EDUCATION sit in the file as four complete, separate fields (section 7). Sex, age, and marital status are prohibited bases under the Equal Credit Opportunity Act. Through other fields, weakly: the strongest links are between the demographic fields themselves, AGE with MARRIAGE at 0.277, and among the model inputs LIMIT_BAL carries the most information about a group, 0.167 for age, 0.142 for education, 0.116 for sex (section 8c). The notebook tests single fields only, so combinations are untested. One limitation sits on all four fields: the credit limit includes supplementary cards, and the documentation does not say whose demographics a multi-card account carries.
+
+### Are there known errors, oddities, or redundancies in the data?
+
+Five. The six repayment-status fields are the six strongest predictors, 0.423 down to 0.251 (section 8b), and they hold two values the dictionary does not document, 0 and -2, in 86.5 percent of records, with 0 the most common value in every month (section 9, check 3). Value 1 appears 3,688 times in the most recent month and then 28, 4, 2, 0, and 0 in the older months, and there is no field named PAY_1 (section 9, check 3), so the months were not computed by one rule. The six statement-balance fields correlate at 0.80 to 0.95, fifteen pairs (section 8a). And the amounts: 1,930 accounts (6.4 percent) carry a negative balance in some month (section 9, check 4), the largest balance, 1,664,089, and the largest payment, 1,684,259, each exceed the largest credit limit, 1,000,000 (section 5), and 3,931 accounts carry a balance above their own limit in at least one month (section 11.7). Thirty-five patterns repeat in every field except ID, 70 rows, 30 of them inactive in every month and 5 with six months of identical activity (section 9, check 1).
+
+### Does the dataset contain anything that could identify a person, or anything confidential?
+
+One identifier field, ID, which is an account identifier within the extract only, 1 to 30,000 (section 2). No name, card number, or date. The extract is cardholder account data and is confidential to the bank, held under the bank's information security controls. Whether a record could be re-identified by combining the credit limit, the demographic fields, and six months of amounts has not been assessed, and this datasheet records that as a limitation.
+
+## Section 3: Collection process
+
+**Objective:** Record how the data came to exist, and be honest about what is not known. Most of the answers here come from the data science team's written response. Where the team said a question is open, this section says so in the same words.
+
+### How was the data recorded, and by what system or process?
+
+The records were recorded by the bank's card servicing systems in the ordinary running of the accounts: the four demographic fields as recorded at application, the credit limit as the total line including supplementary cards, and the statement, payment, and repayment-status fields for six consecutive statement months, most recent first (data dictionary). The extract was drawn from those records by the modeling team. The repayment-status fields were not computed by one rule in all six months, and the team cannot say what the rules were (modeling team response, answer 3).
+
+### Over what period, and by whom?
+
+Not documented. The extract's documentation does not state the calendar months the six statements cover or when the extract was drawn, and the file has no date field (section 2). The team's written response does not supply a period. Whether the book rate the dictionary gives, near 3 percent, was measured over the same period on the same definition is open.
+
+### Were the people the records describe told their information would be used this way?
+
+Not established. The Use Case Assessment listed the permissibility of using account history for model development among the assumptions to be checked against the cardholder agreement and the bank's privacy notices, and that check is not in the program file. This datasheet records it as a limitation, not a finding either way.
+
+### Which questions about how the data was gathered cannot be answered from the documentation the program holds?
+
+Nine, as the team's written response and section 12 of the notebook list them: how the 30,000 accounts were drawn and the book default rate for the period they cover, what repayment-status values 0 and -2 record in each of the six fields, whether one rule computed all six months and why there is no PAY_1, what EDUCATION 0, 5, and 6 and MARRIAGE 0 record, what "defaulted on payment" means operationally and whether it matches the loss forecast's definition, the unit of the amounts and how a balance or a payment can exceed the largest credit limit, whether the 30 inactive repeated patterns are one account or two and which identifier is correct for the five pairs treated as one, whose demographics a supplementary-card account carries, and whether credit bureau data can be joined.
+
+## Section 4: Preprocessing, cleaning, and labeling
+
+**Objective:** Record what the data science team has now performed on the dataset, before the first model is trained, precisely enough that a reviewer could reproduce it. The data science team's remediation log is the source, and sections 11 and 12 of the notebook are where every number in this section is printed.
+
+### Was any cleaning, recoding, or removal performed? Summarize the changes.
+
+Yes, twelve logged rows, printed before and after in notebook section 11. The enrichment was recorded and the sampling method left open, with reweighting and calibration recorded as modeling-stage decisions and no threshold set on 22.1 percent (R-1). Repayment-status values 0 and -2 were recoded into one named level, "no delinquency recorded, code undefined," in six new fields beside the raw fields (R-2). Five per-month payment-behavior fields were derived from the documented bills and payments under one rule (R-3). The month-rule inconsistency was documented, not fixed (R-4). EDUCATION 0, 5, 6 and MARRIAGE 0 were recoded into one "unknown" level in two new fields (R-5). The label was left as recorded with its operational definition open (R-6). The amounts were left as recorded, no cap and no sign correction (R-7). One row of each of five repeated pairs with identical six-month activity was removed under a written rule, the only row removal (R-8). Utilization and payment ratio were derived from the most recent month (R-9). SEX, AGE, and MARRIAGE were held out of the model inputs and kept in the working file, with EDUCATION kept as an input under monitoring (R-10). No row left the file in that step: the team took the label off the account, not the account out of the file. No bureau data was joined (R-11). The train/test split rule was recorded for the modeling stage, random by row, stratified on the outcome, 80 percent training and 20 percent held back as the test set, fixed seed, after the deduplication: 23,996 training rows and 5,999 in the held-back test set expected, not executed (R-12, section 11.12).
+
+### Which fields were held out of the model inputs, and why? Are they still in the file?
+
+Twelve fields are in the working file but out of the model inputs, and all twelve are still in the working file (section 12). SEX, AGE, and the recoded MARRIAGE_MODEL, because sex, age, and marital status are prohibited bases under the Equal Credit Opportunity Act and Regulation B, 30 candidate inputs to 27 (R-10, section 11.10). The raw MARRIAGE and EDUCATION codes, because their recoded copies carry the model's view and the raw codes are kept for the evaluation-stage test of merged against separate (R-5). The six raw repayment-status fields, PAY_0 and PAY_2 to PAY_6, because "your status was 0" is not a reason a notice could carry, and the raw codes are kept for the merged-against-separated test (R-2). And ID, the account identifier, never a model input (R-8). EDUCATION_MODEL stays a model input with monitoring, pending the test the Chief AI Officer ruled on. Every one of the 29,995 cardholders stays in the training data with all of their other fields. The Chief AI Officer's watch list measures false alarm and miss rates by sex, age band, marital status, and education, and tests whether the remaining inputs recover each field, which is why none of them was deleted.
+
+### Which fields were derived or recoded, and is the raw field still available beside each?
+
+Fifteen fields were added, each beside its source, and the source is available beside every one of them (section 12: 25 columns as loaded, 40 after). STATUS_0 and STATUS_2 to STATUS_6, the six repayment-status fields with documented values in words and 0 and -2 as "no delinquency recorded, code undefined," 17,496 records in STATUS_0 (R-2, section 11.2). PAID_1 to PAID_5, each payment set against the prior month's statement balance as no balance, paid in full, paid part, or paid nothing, five fields because the earliest month's prior statement is not in the extract (R-3, section 11.3). EDUCATION_MODEL and MARRIAGE_MODEL, with 0, 5, 6 and 0 as "unknown," 345 and 54 records (R-5, section 11.5). UTIL_RECENT, the most recent balance over the limit, median 0.314, and PAY_RATIO_RECENT, the most recent payment over the prior statement where that statement was above zero, median 0.078 and empty for 3,175 accounts (R-9, section 11.9). Both are in the same unstated unit as their sources.
+
+### How many records were removed, and under what rule?
+
+Five. 30,000 rows before and 29,995 after (R-8, section 11.8). The rule: identical in every field except ID, with a non-zero balance or payment in at least one of the six months, keep the lower ID and remove the other. Five pairs met it. The 30 inactive pairs, 60 rows, did not and stay in the file, because two unused cards with the same limit and application details can coincide. Nothing else was removed, and the rate of default is 22.1 percent before and after.
+
+### Is every change recorded somewhere a reviewer can reproduce it?
+
+Yes. Every change is a numbered row in the modeling team's remediation log, and each row names the notebook subsection where the before and the after are printed (11.1 to 11.12). Section 12 prints the file after remediation: 27 model inputs, the twelve held-out fields with their reasons, the fifteen fields added, the 3,175 empty values in PAY_RATIO_RECENT, and the nine open items. The notebook runs from the top and reproduces every number.
+
+## Section 5: Uses
+
+**Objective:** Fix what the dataset is for and what it is not for. This section inherits the red lines from the Use Case Assessment and the watch list from the Chief AI Officer's Week 2 memo. It is where the dataset's limits become instructions to whoever uses it next.
+
+### What has the dataset been used for so far?
+
+To build the LimitWise prototype the modeling team produced before any governance file existed, on the extract as it came, which back-tested well against the incumbent scorecard. This week the team has prepared the extract for the next version of the model, which has not yet been trained, and this datasheet documents that preparation. The extract has not been used in any account action, and the Chief AI Officer's condition two, no retraining until the repayment-status codes are defined in writing, has not been lifted.
+
+### What is it intended to be used for?
+
+To train and evaluate LimitWise as an early warning score for existing card accounts, under the posture the Chief AI Officer directed in Week 1: preventive first, outreach and hardship offers with a human reviewing every contact, and nothing that touches an account's terms in the first phase. Evaluation will carry the six watch-list items from the Week 2 memo, each with the tripwire, owner, and review point the memo states: calibration by score band against a holdout that reflects the book, not the extract. The repayment-status codes, defined in writing for each field before any retraining, then model stability with 0 and -2 merged against separated, and a plain-language reason for every value in every one of the six fields before any phase touches an account's terms. The education and marital status codes merged into one level and again treated as missing, with per-group metrics compared. False alarm and miss rates by sex, age band, marital status, and education, and monthly outreach contact rates by group, with groups too small to evaluate flagged and counted, cardholders 65 and over among them (110 records, about 22 in the held-back test set, section 11.12). The four demographic fields trained with and without, and each predicted from the remaining inputs to name any proxy. And reason stability, the top reason for every flagged account compared across two random halves.
+
+### What must it never be used for?
+
+Anything outside the account-level early warning use this program assessed. Per the Use Case Assessment's red lines: the score never triggers an account action without human review, and in the first phase never triggers an account intervention at all. Never used for marketing, new-account underwriting, or collections prioritization. No action on the score during a customer's declared hardship. SEX, AGE, and MARRIAGE are never model inputs, and no reason built on a repayment-status code reaches a cardholder until the code has a plain-language meaning. Because the extract's rate is not the bank's, it must not be used to set a threshold, a loss criterion, or a calibration claim against 22.1 percent until the sampling method and the book rate for the period are documented.
+
+### What about its composition or collection could affect future uses, including fairness and legal obligations?
+
+Five things. The sampling method is unknown, so scores are relative rankings until a book-like holdout or a documented rate arrives, and at the scale of a book scored monthly, a small error rate is a large number of cardholders, which is the arithmetic the Chief AI Officer's memo carries into the validation report. The six strongest inputs carry a level whose meaning is open, so every reason the model gives that rests on "no delinquency recorded, code undefined" is a reason the bank cannot yet put in a notice under Regulation B. The most recent month's status was computed by a different rule from the older months, which is a training-serving risk the derived payment-behavior fields exist to test. Whose demographics a supplementary-card account carries is open, so every fairness measurement by sex, age, marital status, or education is a measurement of the recorded cardholder, not necessarily of the person whose payments produced the outcome. And the label's operational definition is open, so any comparison of LimitWise's default rate with the loss forecast's is a comparison of two definitions until the Chief Credit Officer's office confirms they match.
+
+## Section 6: Distribution
+
+**Objective:** Record whether and how the dataset leaves the organization, and under what terms.
+
+*Completed by the AI Governance Lead. Read it, but do not rewrite it.*
+
+### Will the dataset be shared outside the organization, and how?
+
+No. The extract stays inside the bank's environment. It is held in the LimitWise program file and is available to the program team only. The bank's model inventory classifies LimitWise as high-materiality, and the extract is part of the model package that independent validation will review under that classification, inside the bank.
+
+### Under what terms or restrictions?
+
+Internal use for the LimitWise program, under the bank's information security controls. There is no license because there is no distribution. Any request to use the extract outside the program, including for the loss forecast directly, for marketing, for new-account underwriting, or for collections prioritization, is a new use and requires a new Use Case Assessment.
+
+### Are there legal or contractual limits on sharing it?
+
+Yes. The extract is cardholder account data under the Office of the Comptroller of the Currency's supervision and the bank's own privacy commitments. The cardholder agreement and the bank's privacy notices govern whether account history may be used for model development at all, and the Use Case Assessment listed that permission among the assumptions to be checked. Nothing in that check would permit sharing outside the bank.
+
+## Section 7: Maintenance
+
+**Objective:** Record who looks after the dataset, how errors are reported and corrected, and what happens to it over time.
+
+*Completed by the AI Governance Lead. Read it, but do not rewrite it.*
+
+### Who maintains the dataset, and how are errors reported and corrected?
+
+The modeling team maintains the extract and the remediated working file. Errors, questions about a field's meaning, and requests for corrections go to the team through the AI Governance Office, which records each one in the program file. The written questions the Chief AI Officer forwarded after Week 2, and the team's answers, are the first entries.
+
+### Will it be updated, and will earlier versions be kept?
+
+The extract as loaded is not changed. Every change is made to the working file beside it, and the remediation log records each change, so an earlier state can always be recovered. If the sampling memo, the definitions of repayment-status values 0 and -2, or the operational definition of default is documented later, this datasheet is the document that changes. The open items in sections 2 and 3 are where. A later join of credit bureau data would be a new extract and a new datasheet, not a revision of this one.
+
+### For how long will it be retained, and what happens to it after that?
+
+Retention follows the bank's existing record retention schedule for card account records and model documentation. The specific period is not stated in the program file, and this datasheet records that as an open item for the Chief AI Officer's office rather than inventing one.
+
+---
+
+## Make this your own
+
+This page carries the AI Governance Lead's exemplar so that the repository is complete on day one. Your portfolio should carry yours. Three steps, about ten minutes:
+
+1. On Hugging Face, the **+** menu, then **New Dataset**. Owner: your account. Name: `limitwise-data`. License: cc-by-4.0. Public. Create it.
+2. Export your Dataset Datasheet (your Week 3 deliverable) as a PDF and upload it to the new repository under the name `Dataset-Datasheet.pdf` (Files, then Contribute, then Upload files).
+3. Upload a `README.md`: copy `README_TEMPLATE.md` from this repository, put your name in it, and replace YOUR-ACCOUNT with your Hugging Face account name everywhere it appears (including inside the metadata block at the top, which otherwise stays as it is). The template links to your PDF. If you would rather have your card render as the page itself, paste the text of your Dataset Datasheet below the metadata block instead.
+
+Then add the new repository to your collection beside your Space and your model repository. Do not upload the Lead's exemplar, the memos, the audit briefings, the notebooks, or the data.

README_TEMPLATE.md

@@ -0,0 +1,23 @@
+---
+license: cc-by-4.0
+pretty_name: Corvana Bank LimitWise working file (remediated)
+language:
+  - en
+task_categories:
+  - tabular-classification
+tags:
+  - ai-governance
+  - datasheet
+---
+
+# Dataset Datasheet: the LimitWise working file (Corvana Bank)
+
+**Governance analyst:** [YOUR NAME]
+
+**Program:** AI Governance Practicum, Track 1, cohort 1
+
+The datasheet is the PDF in this repository: [Dataset-Datasheet.pdf](Dataset-Datasheet.pdf).
+
+The data itself is not copied here. It is the remediated working file in [drdavidprivacy/practicum-case-packs](https://huggingface.co/datasets/drdavidprivacy/practicum-case-packs). The model trained on it is [YOUR-ACCOUNT/limitwise-model](https://huggingface.co/YOUR-ACCOUNT/limitwise-model).
+
+*The organization, the system, and the people are fictional. The data is real and public: Default of Credit Card Clients, UCI Machine Learning Repository, id 350 (Yeh, 2016), CC BY 4.0.*