Data catalog
Every table behind this site.
What each table is built from, who owns it, how fresh it is and what every column means, plus the metrics defined on them. It’s generated from the same definitions the nightly run tests against, so it can’t say something the tests don’t.
Lineage map
Drag a table to pull its lineage; select one to open it belowTables
16 tables, each queryable in any SQL sourceprojectsOne row per project card on /projects; Home's featured cards are the featured rows.
- Owner
- Logan Mansfield
- Built from
- Page markup → Parsed page rows
- Lives in
- data/console/site.sql
- Freshness
- Rebuilt every run; expected within 36 hours of the site pages
- Shown on
- Home: the featured projects; Projects: the featured projects; Projects: the more builds
slugVARCHARThe card's anchor id on /projects (#fraud, #owl-park...).titleVARCHARProject name as the card shows it.badgeVARCHARStatus badge on the card (Live, Running, Retired...), or null.eyebrowVARCHARThe small line above the title: kind of project.yearINTEGERYear shown on the card.featuredBOOLEANTrue for the three featured projects shown on Home too.statusVARCHARactive, or archived for retired cards.summaryVARCHARThe card's synopsis.tagsVARCHAR[]The card's tags, in order.linksSTRUCT("label" VARCHAR, url VARCHAR)[]The card's links as (label, url) pairs.sort_orderINTEGERPosition on /projects.
Tested for: at least 3 rows; no blanks in slug, title, eyebrow, year, summary; slug is unique; title is unique; status is one of active, archived; exactly three featured projects; every column is described. Latest results.
experienceOne row per role on the Home experience timeline.
- Owner
- Logan Mansfield
- Built from
- Page markup → Parsed page rows
- Lives in
- data/console/site.sql
- Freshness
- Rebuilt every run; expected within 36 hours of the site pages
- Shown on
- Home: the experience timeline
roleVARCHARJob title.noteVARCHARA short aside after the title (where a role started), or null.orgVARCHAREmployer.locationVARCHARWhere the role was based.startedDATEFirst month in the role.endedDATEMonth the role ended; null while it's current.is_currentBOOLEANTrue for the role held now.summaryVARCHAROne-line description of the work.sort_orderINTEGERPosition on the timeline.
Tested for: at least 1 row; no blanks in role, org, started, summary; a role ends after it starts; exactly one current role; every column is described. Latest results.
skillsThe Home toolkit, one row per skill.
- Owner
- Logan Mansfield
- Built from
- Page markup → Parsed page rows
- Lives in
- data/console/site.sql
- Freshness
- Rebuilt every run; expected within 36 hours of the site pages
- Shown on
- Home: the toolkit
skillVARCHARThe skill as its chip reads.categoryVARCHARThe toolkit group it sits in.sort_orderINTEGERPosition in the toolkit.
Tested for: at least 3 rows; no blanks in skill, category; skill is unique; category is one of Data stack, Forecasting & ML, Platforms I run at ZooTampa; every column is described. Latest results.
credentialsCertifications and workshops listed on Home.
- Owner
- Logan Mansfield
- Built from
- Page markup → Parsed page rows
- Lives in
- data/console/site.sql
- Freshness
- Rebuilt every run; expected within 36 hours of the site pages
- Shown on
- Home: the credentials
credentialVARCHARName of the certification or workshop.detailVARCHARExam code, issuer or kind.sort_orderINTEGERPosition in the list.
Tested for: at least 1 row; no blanks in credential; credential is unique; every column is described. Latest results.
resumeOne row: who I am, the Home "now" strip, and where to reach me.
- Owner
- Logan Mansfield
- Built from
- Page markup → Parsed page rows
- Lives in
- data/console/site.sql
- Freshness
- Rebuilt every run; expected within 36 hours of the site pages
- Shown on
- Home: the now strip
nameVARCHARFull name.headlineVARCHARThe hero's first clause.nowVARCHARThe "Now" row of the strip.buildingVARCHARThe "Building" row of the strip.latestVARCHARThe "Latest" row of the strip.locationVARCHARHome base, from the footer.emailVARCHARContact email from the Email button.linkedinVARCHARLinkedIn profile URL.githubVARCHARGitHub profile URL.resume_pdfVARCHARRésumé PDF URL.siteVARCHARThis site's URL.
Tested for: exactly 1 row; no blanks in name, headline, now, building, latest, email; every column is described. Latest results.
fraud_totalsOne row: the fraud pipeline's last 30 days in total.
- Owner
- Logan Mansfield
- Lives in
- data/fraud-snapshot.json → totals
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the last 30 days totals
checkoutsINTEGERCheckouts scored.flaggedINTEGERCheckouts sent to review or declined.declinedINTEGERCheckouts declined.reviewedINTEGERCheckouts sent to manual review.fraudINTEGERCheckouts the answer key marks as fraud.fraud_caughtINTEGERFraudulent checkouts that were flagged.false_declinesINTEGERLegitimate checkouts that were declined.legitINTEGERCheckouts the answer key marks as legitimate.fraud_dollars_stoppedDOUBLEDollars of fraud that were flagged.fraud_dollars_approvedDOUBLEDollars of fraud that got through.p50_scoring_msDOUBLEMedian time to score a checkout, in milliseconds.p95_scoring_msDOUBLE95th-percentile scoring time, in milliseconds.p95_checkout_msDOUBLE95th-percentile time for the whole checkout, in milliseconds.scoring_timeoutsINTEGERCheckouts approved because scoring timed out (fail open).active_daysINTEGERDays in the window with at least one checkout.
Tested for: exactly 1 row; no blanks in checkouts, fraud, fraud_caught, legit; fraud plus legit equals checkouts; flagged equals reviewed plus declined; caught fraud is no more than all fraud; updated within 36 hours; every column is described. Latest results.
fraud_dailyOne row per America/New_York day with checkouts in the last 30 days.
- Owner
- Logan Mansfield
- Lives in
- data/fraud-snapshot.json → daily
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the fraud per day chart
dayDATECalendar day, America/New_York.checkoutsINTEGERCheckouts scored that day.approvedINTEGERCheckouts approved.reviewedINTEGERCheckouts sent to manual review.declinedINTEGERCheckouts declined.fraudINTEGERCheckouts the answer key marks as fraud.fraud_caughtINTEGERFraudulent checkouts that were flagged.false_declinesINTEGERLegitimate checkouts that were declined.fraud_dollars_stoppedDOUBLEDollars of fraud that were flagged.fraud_dollars_approvedDOUBLEDollars of fraud that got through.p95_scoring_msDOUBLE95th-percentile scoring time that day, in milliseconds.
Tested for: at least 1 row; no blanks in day, checkouts, fraud, fraud_caught; day is unique; approved, reviewed and declined add up to checkouts; the days add up to the 30-day totals; updated within 36 hours; every column is described. Latest results.
fraud_score_binsThe model's score histogram: checkouts per 0.05-wide score bin, split by the answer key.
- Owner
- Logan Mansfield
- Lives in
- data/fraud-snapshot.json → score_bins
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the score distribution chart
score_fromDOUBLEBottom of the bin (inclusive).score_toDOUBLETop of the bin.legitINTEGERLegitimate checkouts scored in the bin.fraudINTEGERFraudulent checkouts scored in the bin.
Tested for: 1 to 20 rows; score_from is unique; bins sit between 0 and 1; the bins add up to the 30-day totals; updated within 36 hours; every column is described. Latest results.
fraud_patternsFraud and the share caught, by the generator's fraud pattern.
- Owner
- Logan Mansfield
- Lives in
- data/fraud-snapshot.json → patterns
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the catch rate by fraud pattern table
patternVARCHARThe kind of fraud the generator played out.fraudINTEGERFraudulent checkouts of that pattern.caughtINTEGEROf those, how many were flagged.
Tested for: at least 1 row; pattern is unique; pattern is one of card_testing, stolen_card_resale, gift_card_cashout, account_takeover, fraud_ring; the patterns add up to all fraud; updated within 36 hours; every column is described. Latest results.
fraud_compareThe rules and the new model side by side on the same checkouts: one row per scorer.
- Owner
- Logan Mansfield
- Built from
- Supabase checkout tables → Supabase model scores → Scored checkouts, last 30 days → Checkouts both scorers saw
- Lives in
- data/fraud-snapshot.json → compare
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the new model vs rules table
scorerVARCHARrules or model.checkoutsINTEGERCheckouts both scorers saw.fraudINTEGEROf those, how many the answer key marks as fraud.legitINTEGEROf those, how many are legitimate.flaggedINTEGERCheckouts this scorer flagged (sent to review or declined).caughtINTEGERFraudulent checkouts this scorer flagged.cutoffDOUBLEThe lowest score this scorer flagged at: the rules' risk score, or the model's fraud probability.p50_msDOUBLEMedian scoring time, in milliseconds.p95_msDOUBLE95th-percentile scoring time, in milliseconds.
Tested for: exactly 2 rows; scorer is unique; scorer is one of rules, model; no blanks in scorer, checkouts, fraud, legit, flagged, caught; fraud plus legit equals checkouts; caught fraud is no more than flagged or all fraud; both scorers saw the same checkouts; updated within 36 hours; every column is described. Latest results.
fraud_compare_patternsFraud caught by each scorer, by the generator's fraud pattern, on the checkouts both scorers saw.
- Owner
- Logan Mansfield
- Built from
- Supabase checkout tables → Supabase model scores → Scored checkouts, last 30 days → Checkouts both scorers saw
- Lives in
- data/fraud-snapshot.json → compare_patterns
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the new model vs rules by fraud pattern chart
patternVARCHARThe kind of fraud the generator played out.scorerVARCHARrules or model.fraudINTEGERFraudulent checkouts of that pattern.caughtINTEGEROf those, how many this scorer flagged.
Tested for: at least 2 rows; pattern and scorer are unique together; scorer is one of rules, model; pattern is one of card_testing, stolen_card_resale, gift_card_cashout, account_takeover, fraud_ring; caught fraud is no more than all fraud; the patterns add up to each scorer's catches; updated within 36 hours; every column is described. Latest results.
fraud_compare_sweepWhat each scorer would flag at a few other cutoffs, on the checkouts both scorers saw.
- Owner
- Logan Mansfield
- Built from
- Supabase checkout tables → Supabase model scores → Scored checkouts, last 30 days → Checkouts both scorers saw
- Lives in
- data/fraud-snapshot.json → compare_sweep
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
- Shown on
- Live data: the other cutoffs table
scorerVARCHARrules or model.cutoffDOUBLEFlag every checkout scored at or above this.flaggedINTEGERCheckouts that cutoff would flag.caughtINTEGERFraudulent checkouts that cutoff would flag.
Tested for: at least 2 rows; scorer and cutoff are unique together; scorer is one of rules, model; cutoffs sit between 0 and 1; caught fraud is no more than flagged; updated within 36 hours; every column is described. Latest results.
fraud_compare_day_oneThe rules and the new model over the model's first 24 hours live, a fixed window (orders placed from Nov 14, 22:27:41 UTC, to 24 hours later), counting only rows loaded and model scores recorded before Nov 15, 22:57:41 UTC (the window end plus 30 minutes). The site's copy quotes these counts, so both the order window and the load/score cutoff are fixed; claims in metrics.json check the copy against them.
- Owner
- Logan Mansfield
- Built from
- Supabase model scores → Supabase checkout tables
- Lives in
- data/fraud-snapshot.json → compare_day_one
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
scorerVARCHARrules or model.window_startVARCHARStart of the window (UTC, inclusive).window_endVARCHAREnd of the window (UTC, exclusive).checkoutsINTEGERCheckouts in the window both scorers saw.fraudINTEGEROf those, how many the answer key marks as fraud.legitINTEGEROf those, how many are legitimate.flaggedINTEGERCheckouts this scorer flagged.caughtINTEGERFraudulent checkouts this scorer flagged.
Tested for: exactly 2 rows; scorer is unique; scorer is one of rules, model; no blanks in scorer, window_start, window_end, checkouts, fraud, legit, flagged, caught; fraud plus legit equals checkouts; caught fraud is no more than flagged or all fraud; the window is the fixed first 24 hours; updated within 36 hours; every column is described. Latest results.
fraud_metricsOne row per site metric over fraud_totals: the definitions every page reads, with today's value. Each metric is also a macro, like catch_rate(fraud_caught, fraud).
- Owner
- Logan Mansfield
- Lives in
- data/pipeline/metrics.sql, from data/pipeline/metrics.json
- Freshness
- Rebuilt every run; expected within 36 hours of the fraud snapshot
metricVARCHARThe metric's name, as metrics.json and its macro call it.labelVARCHARWhat the site calls it.definitionVARCHARNumerator over denominator, in fraud_totals columns.numeratorDOUBLEThe numerator's value.denominatorDOUBLEThe denominator's value.valueDOUBLENumerator divided by denominator (blank when there's nothing to divide by).
Tested for: metric is unique; no blanks in metric, label, definition, value; every metric is a share between 0 and 1; catch rate and fraud approved add up to 1; updated within 36 hours; every column is described. Latest results.
github_activityPublic GitHub activity per day and repository over the last 30 days, as far back as the 300 most recent public events the API serves.
- Owner
- Logan Mansfield
- Built from
- GitHub public API → Public GitHub events
- Lives in
- data/pipeline/github.json → activity
- Freshness
- Rebuilt every run; expected within 30 hours of the GitHub pull
dayDATECalendar day, America/New_York.repoVARCHARPublic repository (owner/name).eventsINTEGERPublic events that day (pushes, pull requests, reviews...).pushesINTEGEROf those, pushes.
Tested for: any number of rows; no blanks in day, repo, events; day and repo are unique together; pushes are no more than events; updated within 30 hours; every column is described. Latest results.
github_reposPublic repositories on github.com/lmansf (forks left out).
- Owner
- Logan Mansfield
- Built from
- GitHub public API
- Lives in
- data/pipeline/github.json → repos
- Freshness
- Rebuilt every run; expected within 30 hours of the GitHub pull
repoVARCHARRepository (owner/name).languageVARCHARGitHub's main language for it, or null.starsINTEGERStargazers.pushed_onDATEDay of the last push, UTC.archivedBOOLEANTrue if the repository is archived.
Tested for: at least 1 row; no blanks in repo, pushed_on; repo is unique; updated within 30 hours; every column is described. Latest results.
Metrics
each is also a SQL macro-
Catch rate
catch_rate(fraud_caught, fraud)Share of fraudulent orders flagged: sent to review or declined by the rules, or marked as fraud by the new model.
fraud_caught / fraud, onfraud_totals,fraud_daily,fraud_patternsascaught / fraud,fraud_compareascaught / fraud,fraud_compare_patternsascaught / fraud,fraud_compare_day_oneascaught / fraud.Home and Projects quote it as “flagged about half the fraud”; the nightly run checks it stays 40% to 60%.
Home and Projects quote it as “caught 110 of 111 fraudulent orders in its first 24 hours live”; the nightly run checks it is still 110 of 111 on
fraud_compare_day_one(scorer model).Projects quote it as “the new model caught 110 of 111 fraudulent orders”; the nightly run checks it is still 110 of 111 on
fraud_compare_day_one(scorer model).Projects quote it as “where the rules caught 52”; the nightly run checks it is still 52 of 111 on
fraud_compare_day_one(scorer rules).Projects quote it as “110 of 111 caught”; the nightly run checks it is still 110 of 111 on
fraud_compare_day_one(scorer model). -
Fraud approved
miss_rate(fraud, fraud_caught)Share of fraudulent orders the rules approved: the fraud that got through.
(fraud - fraud_caught) / fraud, onfraud_totals,fraud_daily,fraud_patternsas(fraud - caught) / fraud. -
Good customers declined
false_decline_rate(false_declines, legit)Share of legitimate orders the rules declined outright.
false_declines / legit, onfraud_totals. -
Good customers flagged
legit_flag_rate(flagged, fraud_caught, legit)Share of legitimate orders flagged: sent to review or declined by the rules, or marked as fraud by the new model.
(flagged - fraud_caught) / legit, onfraud_totals,fraud_compareas(flagged - caught) / legit,fraud_compare_day_oneas(flagged - caught) / legit.Home and Projects quote it as “and no legitimate orders”; the nightly run checks it stays 0%.
Projects quote it as “25 good orders flagged out of 1,964”; the nightly run checks it is still 25 of 1,964 on
fraud_compare_day_one(scorer model). -
Flags that were fraud
flag_precision(fraud_caught, flagged)Share of flagged orders that were really fraud (precision).
fraud_caught / flagged, onfraud_totals,fraud_compareascaught / flagged,fraud_compare_sweepascaught / flagged. -
Orders flagged
flag_rate(flagged, checkouts)Share of all orders flagged: sent to review or declined by the rules, or marked as fraud by the new model.
flagged / checkouts, onfraud_totals,fraud_compare. -
Fraud rate
fraud_rate(fraud, checkouts)Share of all checkouts that were fraudulent.
fraud / checkouts, onfraud_totals,fraud_daily. -
Scoring timeouts
timeout_rate(scoring_timeouts, checkouts)Share of checkouts whose fraud score timed out and failed open (approved unscored).
scoring_timeouts / checkouts, onfraud_totals.
Sources and staging
raw Page markup
The Home and Projects pages as published. They are the source of truth for the site's own tables, so a table can never say something the pages don't.
index.html, projects.html
staging Parsed page rows
Each card, role, skill, credential and the hero, read out of the page markup into typed rows (dates parsed, badges split from titles).
scripts/build_console_data.py
raw Supabase checkout tables
The fraud pipeline's four raw Postgres tables of simulated webstore checkouts. They stay in Supabase: nothing row-level ever reaches this site.
raw_transactions, raw_fraud_decisions, raw_fraud_truth, raw_checkout_responses
raw Supabase model scores
The new model's scoring log: one row each time it scored a checkout, with its fraud probability, its decision and how long it took. It stays in Supabase: nothing row-level ever reaches this site.
realtime_fraud_audit
staging Scored checkouts, last 30 days
One row per scored checkout in the last 30 America/New_York days, joined to the generator's answer key. Built inside a read-only transaction on Supabase; only the aggregates below leave it, as data/fraud-snapshot.json.
the base CTE in scripts/fraud-snapshot.sql
staging Checkouts both scorers saw
Scored checkouts the new model also scored live, one at a time, joined to its latest score. A one-off batch rescore of older orders on Nov 14 is left out: it ran in bulk and gave every order nearly the same score.
the model_scores and compare_base CTEs in scripts/fraud-snapshot.sql
raw GitHub public API
Public activity and public repositories for github.com/lmansf. Private work never appears here.
api.github.com /users/lmansf/events/public and /users/lmansf/repos
staging Public GitHub events
One row per public event in the last 30 days, keeping only its id, type, repository and time (no commit messages, names or emails).
data/pipeline/stg_github_events.json, written by scripts/data_run.py