Banner Schema Search

🌐 Live at bss.peopleworksservices.com — free, no signup, no tracking.

A smart, self-contained search · query-builder · audit · SQL-dialect translator tool for Ellucian Banner. Search 6,900+ tables and 133,000+ columns instantly, build Banner SQL by snapping visual blocks together, run 14 ready-made security audits, validate any SQL against the live schema with cross-dialect hints (SQL Server ↔ Oracle ↔ PostgreSQL for Banner SaaS), and optionally hook your own AI key for explanations and rewrites — the validator works 100% offline; AI is opt-in BYOK, just open one HTML file in any browser.

"How does it work so well without AI?" — Because for 95% of the job, AI isn't needed. It uses BM25, trigram fuzzy match, and a curated rules engine — the same algorithms that powered Google and Elasticsearch for decades. Pure math, zero magic. AI is reserved for the 5% where it really helps: open-ended explanations and SQL rewrites.

Banner Lego — snap blocks together to build Banner SQL. Each block teaches you the canonical join pattern (PIDM, TERM_CODE+CRN, effective-dated MAX() subquery) while the SQL generates live.

What's Inside

This is one HTML file with six distinct tools glued together:

Tool	What it does	How to reach it
🔎 Schema Search	Type English, find Banner tables/columns. BM25 + synonym expansion + trigram fuzzy matching. Instant results across 133K+ columns.	`/` or start typing
🧱 Banner Lego	Visual query builder. Snap source/join/filter blocks, see the SQL generate live, learn why each join is structured the way it is. 12 pre-built recipes covering Student, HR, Finance, Security.	`#/builder` · Dashboard card · `banner lego`
🛡️ Security Posture Scorecard	Paste one-row output of SR014 into a textarea — 8 traffic-light gauges light up: dormant accounts, terminated-with-access, super users, PII breadth, violations, unused classes, etc. Everything client-side, persists in localStorage.	`#/scorecard` · `security posture`
💡 Ask Banner	Natural-language → generated SQL. Ask "finance security classes", "access to SHAINST", "users in class SFAREGS", "Student user access inventory" — get Argos-ready SQL with step-by-step explanation.	Just type a question
☰ Ready-to-run Reports	14 vetted SR0XX reports: dormant accounts, terminated-with-access, direct-grant bypass, PII audit, violation dashboard, module overview, per-module user-form inventory, security posture scorecard.	`#/reports` · `~`
🔧 SQL Explainer	Paste any Banner SQL. Validates every table/column against the embedded schema, runs 15+ Banner pitfall checks, matches against a curated case library, shows "Did you mean…?" suggestions. Live syntax highlighting + multi-dialect translation hints.	`#/sql`
📋 Argos Cheatsheet	Parameter conventions (`:main_` / `:lcl_` / `:dbn_*`), property accessors, widget naming, 7 copy-ready SQL patterns, and the "Banner Cloud reality check" (Oracle → PostgreSQL migration table for Banner SaaS).	`#/argos`
🌐 Community Resources	Hand-curated index of free Argos webinars (Evisions), Ellucian training (Customer Center, Live conference), and 6 regional Banner User Groups (ALBUG / FLBUG / PABUG / BUGMI / UMBUG). Every link points to an official source.	`#/resources`
✨ AI Assistance (BYOK — optional)	Bring your own Anthropic / OpenAI / Ollama key. Three actions: "Explain in depth" per validation finding · "Suggest fix" with side-by-side LCS diff · "Translate from SQL Server" → Oracle. Client-side redaction of SSN/email/long-IDs before each request. OFF by default.	`#/settings`

Plus: dark/light theme toggle (sun/moon button — cycles Auto → Light → Dark — also Ctrl+Shift+L), command palette (Ctrl+Shift+F), Help panel with 50+ clickable example queries, target-dialect dropdown (Oracle on-prem vs PostgreSQL/SaaS), and a recent-SQL history that survives reloads.

Quick Start (3 steps)

# 1. Install the only dependency
pip install jinja2

# 2. Build the site (takes ~3 seconds)
python build.py

# 3. Open in your browser
#    Windows:
start docs\index.html
#    Mac/Linux:
open docs/index.html

That's it. You now have a fully searchable Banner data dictionary + query builder + security audit tool.

🧱 Banner Lego — The Visual Query Builder

Think Scratch for Banner SQL. New users building their first Banner report typically hit the same walls: "is the enrollment table SFRSTCR or SFRSTCA?", "what's the canonical join to SCBCRSE?", "why do I need MAX(effective_date)?". Banner Lego solves this by turning the query into visual blocks where each block carries its own canonical join pattern and a plain-English "why" tooltip.

Three block kinds, color-coded:

🟢 Source (green) — the FROM table. Seven sources: Student (SPRIDEN), Employee (PEBEMPL), Vendor (FTVVEND), Security User (GOBEACC), Term (STVTERM), etc.
🔵 Join (blue) — add a related table with its canonical join. 20+ joins: registered courses, section details, course catalog (with the MAX(eff_term) effective-dated pattern), primary advisor (with double-SPRIDEN trick), mailing address, preferred email, current jobs, invoices, security classes, etc.
🟡 Filter (yellow) — constrain the result. By term, active registrations only, by subject, approved invoices, active employees, aid year, graduation term, etc.

12 pre-built recipes (one-click load)

Student module:

📚 Student + Courses + Term (the classic registrar query)
🎯 Student + Academic Program (major/college/degree via SGBSTDN)
⛔ Student + Active Holds (SPRHOLD with date-window filter)
📧 Student + Contact Info (preferred email + primary phone + mailing address, all with correct status filters)
🧑‍🏫 Student + Primary Advisor (with double-SPRIDEN for advisor name)
📝 Student + Test Scores (SORTEST — ACT/SAT/TOEFL/placement)
🎖️ Student + Course Grades (SHRTCKN + SHRTCKG joined by tckn_seq_no)
📊 Student + Term GPAs (SHRGPAC — GPA + hours attempted/earned per term)
🏆 Student + Degrees Awarded (SHRDGMR filtered to DEGS_CODE = 'AW')

HR / Finance / Security:

💼 Employee + Current Jobs (PEBEMPL + NBRJOBS with MAX(effective_date) pattern + NBBPOSN for title)
🏪 Vendor + Invoices (FTVVEND + SPRIDEN for name + FABINVH approved only)
🛡️ Security User + Classes (GOBEACC + GURACLS with canonical audit pattern + GTVCLAS for class description)

Schema-validated at build time

Every column referenced in every block is verified against the real schema (field_info.txt) during python build.py. Exit code 2 if any block references a phantom column — no more invented columns ship to production.

[5/6] Validating Banner Lego catalog columns...
       Loaded schema: 6,750 tables, 134,567 columns
       Lego validator: 34 blocks scanned
       [OK] All referenced columns exist in the schema

When the SQL is ready, one click sends it to the SQL Explainer for pitfall checks before you paste it into Argos.

🛡️ Security Posture Scorecard

SR014 → 8 traffic-light gauges in one view. The scorecard is a one-row SELECT (runs in ~2-5 seconds at a typical site) whose output you paste into a textarea in the app. The dashboard paints itself with gauges for every major audit risk:

Metric	Green if ≤	Watch if ≤	Source report
💤 Dormant Accounts	0	20	SR001
🚨 Terminated w/ Access	0	5	SR002
⚡ Super Users	5	15	SR008
🛡️ Direct-Grant Bypass	10	30	SR004
🔒 PII Broad Access	10	30	SR011
📛 Violations (30d)	50	200	SR005
🗑️ Unused Classes	10	30	SR006
🏚️ Legacy GURUCLS	0	5,000	SR003

Everything runs client-side. Pasted values persist in localStorage — a new tab still shows your last scorecard. Click any gauge to jump to the full report.

💡 Ask Banner — Natural Language → SQL

Type a question in plain English. The tool recognizes 8 intent patterns and generates SQL inline with step-by-step explanation:

"finance security classes"            → GTVCLAS query scoped to sysi_code = 'F'
"Student user access inventory"       → SR012 SQL scoped to Student module
"access to SHAINST"                   → who can access this form (class + direct + group)
"users in class SFAREGS"              → class membership with audit_time pattern
"list banner modules"                 → SR013 + a pretty 8-card module grid
"security posture scorecard"          → jumps to the scorecard view
"banner lego" / "query builder"       → jumps to the visual builder
"dormant accounts" / "terminated…"    → opens the matching SR0XX directly

Every intent answer includes:

The generated SQL with syntax highlighting
An Explanation block explaining each clause in plain English
Caveats — Banner-specific gotchas (LISTAGG overflow, role-suffix edge cases, GUBOBJS schema location, etc.)
"Validate in SQL Explainer →" button — one click runs the built-in pitfall checker

14 Ready-to-Run Security Reports

Every report is vetted SQL with header metadata (category, severity, tables, description, when-to-use, caveats):

ID	Title	Category	Severity
SR001	Dormant Banner Accounts	Access Review	MEDIUM
SR002	Terminated Employees with Active Access	Access Review	HIGH
SR003	Class Membership Report (canonical)	Access Review	INFO
SR004	Direct Object Grants — Bypassing CLASS	Privilege Escalation	HIGH
SR005	Security Violation Dashboard (30d)	Monitoring	INFO
SR006	Unused Security Classes (cleanup)	Access Review	INFO
SR007	Most Populated Classes	Access Review	MEDIUM
SR008	Super Users — Top _M Grant Holders	Privilege Escalation	HIGH
SR009	Recent Access Changes (30d audit trail)	Monitoring	INFO
SR010	Object Access Concentration	Inventory	MEDIUM
SR011	PII / Sensitive Data Access Audit	Privilege Escalation	HIGH
SR012	Banner User Access Inventory — Finance	Inventory	INFO
SR013	Banner Module Access Overview	Inventory	INFO
SR014	Security Posture Scorecard feed	Monitoring	INFO

All use canonical Banner patterns: GURACLS with MAX(audit_time) + audit_action <> 'D', LISTAGG ... ON OVERFLOW TRUNCATE for long lists, GOBEACC in GENERAL schema, role-suffix decoding with edge-case handling for custom roles (BAN_DEFAULT_NO_ACCESS etc.).

🔧 SQL Explainer & Validator

Paste any Banner SQL and the tool inspects it against the live schema and a curated knowledge base of real Banner integration cases.

Schema validation — Every table and column referenced is checked against the embedded Banner schema. Unknown identifiers flagged with "Did you mean…?" suggestions (Levenshtein distance).
Banner pitfall detection — Catches real-world gotchas:
- PHRHIST used without PHRHIST_DISP >= '60' filter (in-progress vs posted)
- Payroll tables without a PICT_CODE filter
- SPRIDEN without CHANGE_IND IS NULL (current vs historical)
- SPRADDR with student address types instead of HR-specific
- PEBEMPL_CURRENT_HIRE_DATE vs PEBEMPL_FIRST_HIRE_DATE for tax reporting
- PHRHIST_GROSS > 0 filter that excludes negative reversal records
- PERETOT for SUI/unemployment without student-employment exclusion
- Security-aware rules: GURACLS without audit pattern, role-suffix traps, direct-grant bypass detection, GOBEACC schema mismatches
Function-aware parser — Correctly handles EXTRACT(MONTH FROM col), TRIM(LEADING ' ' FROM col), SUBSTRING(col FROM n FOR m) — no false-positive table references.
Business-case matching — Compares your query's tables against data/business_cases.txt (W-2 reporting, SURS exemptions, payroll adjustments, Federal/Medicare/SS tax) and data/security_cases.txt (12 security-specific cases) and surfaces matching scenarios with lessons learned.
Live syntax highlighting — As you type.
All offline — Everything runs in the browser. Your SQL never leaves your machine.

🔎 Schema Search — BM25 + Banner Synonyms

Type English, get Banner tables. "employee hire date" expands into empl, emp, pebempl, hired, employment, current_hire, dt, activity_date, effective_date — all visible as toggleable chips above the results so you can see exactly what the ranker is looking at. Typos are handled via trigram similarity (addres still finds SPRADDR). 905 matches across 379 tables for this query — ranked by BM25 in <50 ms.

🗺️ Discover Every Module on Your Site

Ask "list banner modules" and the tool shows the canonical Ellucian module catalog (Finance, Student, HR, Advancement, Financial Aid, Accounts Receivable, General, Position Control) plus the SQL (SR013) that lists whatever modules actually exist at your site. Click any card and it auto-fires the per-module user-access inventory for that module.

How Does It Work Without AI?

This is the question everyone asks. The answer: the same techniques that powered search engines for 30 years before LLMs existed.

BM25 Ranking (core algorithm)

BM25 (Best Matching 25) is from the 1990s and is still the default in Elasticsearch, Apache Lucene, and Apache Solr.

score(Document, Query) = Σ for each query term of:
    IDF(term) · (tf · (k1 + 1)) / (tf + k1 · (1 - b + b · docLength / avgDocLength))

IDF = how rare a word is. "PIDM" is rare and important; "the" is common and worthless.
tf = how many times the word appears in this document.
k1=1.5, b=0.75 = standard tuning parameters.

A document scores high when it contains rare, important words from your query, especially if those words appear frequently relative to document length.

Banner Synonym Expansion (the secret sauce)

Banner names are like PEBEMPL_ECLS_CODE. Humans search for "employee class code". A hand-curated synonym map bridges the gap:

'employee': ['empl', 'emp', 'pebempl']
'address':  ['addr', 'spraddr', 'street', 'city', 'state', 'zip']
'salary':   ['wage', 'pay', 'compensation', 'earnings']

When you search "employee hire date", the engine tokenizes, expands with synonyms at 40% weight, and BM25-ranks the combined set.

Fuzzy Matching (typo tolerance)

Trigram similarity — the same technique used by PostgreSQL's pg_trgm. employy still finds employee via character-level overlap.

Build-Time Indexing (why it's fast)

All the heavy computation happens once during python build.py:

table_info.txt + field_info.txt
    |
    v
[Python: Parse → Tokenize → Stem → Calculate BM25 → Build Inverted Index]
    |
    v
[Lego catalog validator — verifies every block's column refs vs schema]
    |
    v
index.html (data + index + 12 Lego recipes + 14 reports embedded as JSON)
    |
    v
[Browser: hash lookups + simple math → instant results]

Project Structure

BannerSemanticSearch/
├── build.py                              # Main build (6 steps: parse → categorize → index →
│                                         #   relationships → validate Lego catalog → generate)
├── requirements.txt                      # jinja2 only
├── src/
│   ├── parser.py                         # Reads table_info.txt, field_info.txt
│   ├── categorizer.py                    # Maps prefixes to 14 Banner modules (incl. Security)
│   ├── indexer.py                        # BM25 + synonym map + trigram index
│   ├── relationships.py                  # FK parsing + inferred relationships
│   └── generator.py                      # Jinja2 rendering + report/case loading
├── scripts/
│   ├── convert_darkmode.py               # One-shot transformer: @media dark → html.dark class
│   └── validate_lego_catalog.py          # Build-time gate: verifies every Lego block's columns
├── templates/
│   └── index.html                        # The SPA (HTML + CSS + JS in one file, ~6K lines)
├── data/                                 # Input (pipe-delimited)
│   ├── table_info.txt                    #   TABLE|TYPE|DESCRIPTION
│   ├── field_info.txt                    #   TABLE|COLUMN|DESCRIPTION
│   ├── relationships.txt                 #   FK constraints (optional)
│   ├── business_cases.txt                #   33 payroll/tax/integration cases
│   ├── security_cases.txt                #   12 security-specific cases
│   └── reports/                          #   14 SR*.sql files with header metadata
│       ├── SR001_dormant_accounts.sql
│       ├── …
│       └── SR014_security_posture_scorecard.sql
└── docs/                                 # Generated output
    ├── index.html                        # THE OUTPUT — open this in any browser (~20 MB)
    └── img/
        └── banner-lego.png               # Screenshot for README

Data Extraction Guide

The tool needs 3 data files from your Banner Oracle database.

Prerequisites

Access to Banner's Oracle database (SQL Developer, TOAD, SQL*Plus, or any Oracle client)
SELECT privileges on schemas: SATURN, PAYROLL, POSNCTL, BANINST1, GENERAL, FIMSMGR

File 1: `table_info.txt`

SELECT trim(TABLE_NAME) || '|' || TABLE_TYPE || '|' ||
       trim(NVL(COMMENTS, '(no comments)'))
FROM   all_tab_comments
WHERE  OWNER IN ('SATURN','PAYROLL','POSNCTL','BANINST1','GENERAL','FIMSMGR')
ORDER BY TABLE_NAME;

File 2: `field_info.txt`

SELECT trim(c.TABLE_NAME) || '|' ||
       c.COLUMN_NAME || '|' ||
       trim(NVL(c.COMMENTS, '(no comments)'))
FROM   ALL_COL_COMMENTS c
WHERE  c.OWNER IN ('SATURN','PAYROLL','POSNCTL','BANINST1','GENERAL','FIMSMGR')
ORDER BY c.TABLE_NAME, c.COLUMN_NAME;

Largest file (~11 MB, 130K+ rows). Takes 10-30 seconds to extract.

File 3: `relationships.txt` (optional)

SELECT cc.TABLE_NAME || '|' || cc.COLUMN_NAME || '|' ||
       rc.TABLE_NAME || '|' || rc.COLUMN_NAME || '|' ||
       c.CONSTRAINT_NAME
FROM   all_constraints c
JOIN   all_cons_columns cc ON cc.CONSTRAINT_NAME = c.CONSTRAINT_NAME AND cc.OWNER = c.OWNER
JOIN   all_cons_columns rc ON rc.CONSTRAINT_NAME = c.R_CONSTRAINT_NAME
                          AND rc.OWNER = c.R_OWNER
                          AND rc.POSITION = cc.POSITION
WHERE  c.CONSTRAINT_TYPE = 'R'
       AND c.OWNER IN ('SATURN','PAYROLL','POSNCTL','BANINST1','GENERAL','FIMSMGR')
ORDER BY cc.TABLE_NAME, cc.COLUMN_NAME;

Without this file, the tool still infers ~10,000 relationships from naming conventions. With it, you get ~15,000.

SQL*Plus spooling

SET LINESIZE 2000 PAGESIZE 0 FEEDBACK OFF HEADING OFF TRIMSPOOL ON LONG 4000
SPOOL table_info.txt
-- (paste query here)
SPOOL OFF

SQL Developer: Run → right-click → Export → Format: Text, Delimiter: | → Save.

Place all files in data/, then run python build.py.

Sharing with Your Team

Option	How
Send the file	Mail `docs/index.html` to anyone. They open in any browser. Done.
GitHub Pages	Push the repo, enable Pages on `/docs`. Share the URL. ~4 MB gzipped.
Shared drive	Copy `docs/index.html` anywhere. No installation for users.
Run from source	`git clone` → `pip install jinja2` → `python build.py` → open

Contributing

Adding a new Banner Lego block or recipe? The build-time validator has your back:

python build.py
# …
[5/6] Validating Banner Lego catalog columns...
       [FAIL] sd.shrdgmr_phantasm — SHRDGMR.SHRDGMR_PHANTASM not found
       [ERROR] Banner Lego catalog has invalid column references.

Add your block to templates/index.html in the BLOCK_CATALOG literal and/or a new recipe in BUILDER_RECIPES. The validator will catch phantom columns before the build completes.

Adding a new security report? Drop a .sql file in data/reports/ with the required header comments (REPORT_ID, TITLE, CATEGORY, TABLES, SEVERITY, DESCRIPTION, WHEN_TO_USE, CAVEATS). The build will pick it up and wire it into the reports index, command palette, and Ask Banner's P4 shortcut mechanism.

FAQ

Q: How big is the output file? ~20 MB (schema data, search index, all 14 reports, Lego catalog, Scorecard, etc.). Opens instantly in any modern browser. On GitHub Pages with gzip, ~4 MB.

Q: Does it need internet? No. Everything is in one HTML file. Open it from desktop, USB, network drive, anywhere.

Q: Does it use AI or machine learning? No. Zero AI. BM25 + synonym expansion + trigram matching + hand-curated case libraries.

Q: What Python version? Python 3.8+. Only dependency: jinja2.

Q: Can I add my own synonyms? Yes. Edit src/indexer.py, add entries to BANNER_SYNONYMS, rebuild. At search time, use the interactive term builder to add session-only terms.

Q: What Banner versions does this support? Any version. Reads generic Oracle metadata (ALL_TAB_COMMENTS, ALL_COL_COMMENTS). Works with Banner 8, Banner 9, and anything in between.

Q: My site has a custom sysi_code not in the canonical list (F/S/P/A/R/T/G/N). Fine — SR013 / list banner modules lists whatever codes actually exist in your GURAOBJ. The 8-card reference grid in Ask Banner shows the canonical Ellucian codes as guidance, but the SQL reports real data.

Q: Does the Scorecard call my database? No — the Scorecard is 100% client-side. You run SR014 in your own Oracle tool, paste the one-row result into a textarea, and gauges populate via JavaScript. Nothing leaves your browser.

License

MIT. Share freely. If your institution finds this useful, we'd love to hear about it — open an issue or send a pull request.

Name		Name	Last commit message	Last commit date
Latest commit History 38 Commits
data		data
docs		docs
scripts		scripts
src		src
templates		templates
.gitattributes		.gitattributes
.gitignore		.gitignore
ARGOS_INTEGRATION_PILOT.md		ARGOS_INTEGRATION_PILOT.md
LICENSE		LICENSE
README.md		README.md
build.py		build.py
requirements.txt		requirements.txt
sql_explainer_prototype.py		sql_explainer_prototype.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Banner Schema Search

What's Inside

Quick Start (3 steps)

🧱 Banner Lego — The Visual Query Builder

12 pre-built recipes (one-click load)

Schema-validated at build time

🛡️ Security Posture Scorecard

💡 Ask Banner — Natural Language → SQL

14 Ready-to-Run Security Reports

🔧 SQL Explainer & Validator

🔎 Schema Search — BM25 + Banner Synonyms

🗺️ Discover Every Module on Your Site

How Does It Work Without AI?

BM25 Ranking (core algorithm)

Banner Synonym Expansion (the secret sauce)

Fuzzy Matching (typo tolerance)

Build-Time Indexing (why it's fast)

Project Structure

Data Extraction Guide

Prerequisites

File 1: `table_info.txt`

File 2: `field_info.txt`

File 3: `relationships.txt` (optional)

SQL*Plus spooling

Sharing with Your Team

Contributing

FAQ

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

Banner Schema Search

What's Inside

Quick Start (3 steps)

🧱 Banner Lego — The Visual Query Builder

12 pre-built recipes (one-click load)

Schema-validated at build time

🛡️ Security Posture Scorecard

💡 Ask Banner — Natural Language → SQL

14 Ready-to-Run Security Reports

🔧 SQL Explainer & Validator

🔎 Schema Search — BM25 + Banner Synonyms

🗺️ Discover Every Module on Your Site

How Does It Work Without AI?

BM25 Ranking (core algorithm)

Banner Synonym Expansion (the secret sauce)

Fuzzy Matching (typo tolerance)

Build-Time Indexing (why it's fast)

Project Structure

Data Extraction Guide

Prerequisites

File 1: table_info.txt

File 2: field_info.txt

File 3: relationships.txt (optional)

SQL*Plus spooling

Sharing with Your Team

Contributing

FAQ

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

File 1: `table_info.txt`

File 2: `field_info.txt`

File 3: `relationships.txt` (optional)

Packages