GH Film Hub v2: Year Two of the Grading System

Last October I wrote about building a film review pipeline — CSV in, grades out, PDFs and dashboards for the Godwin Heights wide receivers. In April I wrote about hosting that site for $0 on GitHub Pages. Both posts were about getting the system to exist.

This post is about using it for a second season.

Week 1 vs Holland is already on the hub. The 2026–2027 season card sits next to last year's Top 5. The grading engine is the same family of codes I started with. What changed is everything around it: how a week file gets in, how pages get out, and whether a player can open the site and understand how to score well without asking me what ER means.

Open the Film Review Hub

Live hub (behind the Team Tools PWA): ghfb.360web.cloud/film. Public mirror: GitHub Pages. Source: GH-Flim-Review. Project page: GH Film Review Pipeline.

The Problem I'm Solving

Year one proved the rubric. Year two exposed the ceremony:

Sunday-night Python. Prep the CSV, run film_grade.py, generate PDFs and HTML, commit out/, wait for Pages.
One season, one index. Last year's grades lived in a folder. There was no "this is year two, that is year one" landing page.
Codes without a legend in front of players. I knew LF from the sheet. They saw a letter grade and asked what they did wrong.
Sheet and site drifted. New codes on the film tab (LF, BBL, EP) did not show up in Code Counts until the pipeline knew them.
Group film that collapsed ++ into --. Key-play positives and negatives looked like the same column if you stripped the + and - while matching headers.

I did not want a new grading system. I wanted the one I already trust to show up on the hub a few minutes after I finish the week tab.

Where Year One Left Us

The v1 pipeline was a local ritual that produced a static site:

CSV Input → Data Preparation → Film Grading → Report Generation → Static Site

That still happens. The difference is who runs it, and from where.

Side-by-side of year-one local scripts versus Film Hub v2: sheet or upload into process-week, then grade, commit, and deploy to Pages and the VPS /film/ hub

v1 strengths I kept:

Custom rubric, snap-normalized metrics, player PDFs, weekly snapshot, group film
Static HTML that coaches can open on a phone
GitHub Pages as a public mirror, VPS /film/ behind the Team Tools hub

v1 ceiling I hit by week one of year two:

I was the only person who could publish a week
The hub did not explain the rubric
A CSV sitting in Google Sheets was not enough — I still had to babysit the pipeline

What's Live in v2

I walked the live hub today. This is what coaches and players actually get.

Two seasons on one landing page

The Film Review Hub is no longer a single-year index. The season selector shows 2026–2027 and 2025–2026 as cards, each with a Top 5 wide receivers list by season-average grade. Year two is already ahead of last year's opening pace on the leaderboard — which is the point of keeping last season one click away. The same landing page links Upload week CSV and Team Tools.

How to score well, on the season page

Every season index now has a How to score well panel above the weeks table. It is the rubric in coaching language, not a code dump:

Relentless effort every rep (earn Relentless Effort, avoid Loaf / Not Full Speed)
Master your assignment (avoid Missed Assignment)
Elite, crisp routes (earn Elite / Good Route, avoid Bad Route)
Win the ball (earn Spectacular Catch, avoid Dropped Pass)
Add yards and chains (C+ / R+ at +0.5 per yard, broken tackles at +0.5, First Downs +5)
Dominate the run game (Good Block +2, Pancake +10)
Finish drives (Touchdown +15, Extra Point Conversion +2)
Stay locked in (avoid Lack of Focus and Bad Body Language)

Players can open the hub and see the same legend the sheet uses. That is the usability change I cared about most.

Week 1 vs Holland, generated this afternoon

The 2026–2027 weeks table is live: dashboards, snapshot, summary PDF, group film PDF, and CSVs. The Week 1 dashboard has a player filter, letter grades, average scores, and a PDF link per player. Footer timestamp: 2026-08-29 17:19. That is the ingest pipeline, not a hand-built page.

Three Ways a Week File Becomes Pages

The new Process week CSV Action (process-week.yml) is the product. Finish the film tab, get it into csv/{season}/WkN_Opponent.csv, and tools/run_week.py grades, writes out/, then dispatches Pages and VPS deploys so a CSV-only push does not publish a stale site.

Three ingest paths — published Google Sheet, git push, and hub upload helper — converging on process-week.yml and run_week.py, then grade, commit, and deploy

Path	When I use it
Published Google Sheet	Usual path. Actions → Process week CSV → source `published_sheet`, season, week, opponent. `tools/fetch_published_csv.py` pulls the tab.
git push	I already have the file. `csv/2026-2027/Wk1_Holland.csv` on `main` triggers the path filter.
Hub upload helper	`/film/upload.html` builds the required filename from season / week / opponent and opens GitHub's upload UI.

The upload page is a filename helper, not a mystery form. Season defaults to 2026–2027, week 1, opponent Holland, required path csv/2026-2027/Wk1_Holland.csv, buttons for Upload on GitHub and View Actions. After the file lands, pages appear on the hub in a few minutes. Same result from git add csv/… && git push.

That workflow landed in PR #2. The rest of today was making the grades match the sheet once the pages existed.

The Technology Stack

Film log:     Google Sheet (published CSV) or csv/{season}/WkN_Opponent.csv
Grading:      Python, pandas, film_grade.py, tools/run_week.py
Reports:      ReportLab PDFs, HTML dashboards, snapshot tables
Hub:          Static site on VPS /film/ (Team Tools proxy) + GitHub Pages mirror
Ingest CI:    process-week.yml → commit out/ → dispatch deploy.yml + deploy-vps.yml
Coach UX:     Season selector, upload helper, How to score well, player filter

No new database. No new app framework. The hub is still generated HTML. v2 is the path into that HTML.

Rubric Updates From a Year of Film

The original post listed the year-one legend. After a full season of coding plays, three codes were on the sheet and missing from the pipeline — so Code Counts lied. Today they are first-class:

Year-two rubric codes: Extra Point Conversion +2, Lack of Focus -0.5, Bad Body Language -0.5, with the coaching language that appears on reports and the hub

Code	Meaning	Points
EP	Extra Point Conversion	+2
LF	Lack of Focus	-0.5
BBL	Bad Body Language	-0.5

Broken tackles are +0.5 per tackle (not +1.0). Dashboards always list the full code legend. Player reports now emit coaching lines for LF and BBL instead of swallowing those tags.

I also had to stop normalizing headers by stripping every non-alphanumeric character. Key play ++ and Key play -- were collapsing into the same field, so the group film PDF mixed positives and negatives. The matcher now keeps + and -. That is a one-line usability bug that only shows up when you actually run group film for a real week.

Custom Scoring System (Year Two)

The pipeline still scores from codes, then maps a 0–100 score to A–F. The live season page prints this legend:

Positive plays

Touchdown: +15
Relentless Effort: +5
Elite Route: +7
Good Route: +2
Catch/Rush yardage: +0.5 per yard
Broken Tackles: +0.5 per tackle
Good Block: +2
Pancake Block: +10
First Down: +5
Spectacular Catch: +10
Extra Point Conversion: +2

Negative plays

Missed Assignment: -10
Dropped Pass: -15
Bad Route: -2
Loaf: -2
Not Full Speed: -3
Whiffed Block: -1
Holding: 0
Lack of Focus: -0.5
Bad Body Language: -0.5

Letter bands are unchanged: A 90+, B 80+, C 70+, D 60+, else F. Outstanding still means 95+.

Implementation Highlights

One Action, three sources

process-week.yml either fetches the published sheet, takes a csv_path from workflow_dispatch, or diffs csv/**/*.csv on push. Then it runs run_week.py per file, regenerates upload.html, commits out/ (and csv/ when the sheet was fetched), and dispatches both deploys.

on:
  push:
    branches: [main]
    paths:
      - "csv/**/*.csv"
  workflow_dispatch:
    inputs:
      source:
        type: choice
        options: [published_sheet, git]
        default: published_sheet

Shared hub chrome

Season selector and upload page import the same HUB_LINK_SCRIPT so Team Tools and hub navigation stay consistent instead of forking copy in two generators.

Batch is still there

I can still rebuild a season locally. I should not have to, for a normal Friday night.

Technical Challenges Solved

Sheet-first ingest without a new app

The film log already lived in Google Sheets. v2 treats the published CSV as the source of truth and GitHub as the build worker — same idea as building around coach spreadsheets, applied to film instead of attendance.

Deploy after a CSV-only commit

Generating out/ in CI is useless if Pages still serves last week's HTML. Process-week commits artifacts, then dispatches Pages and VPS workflows so the hub and the mirror move together.

Rubric fidelity

If the sheet has a code the pipeline ignores, Code Counts and the group film PDF are fiction. Year two started by making LF, BBL, and EP count, and by not treating ++ and -- as the same column.

Project Impact

Time: A week file can go sheet → hub without a local venv session.
Clarity: Players see How to score well on the same page as their week links.
Continuity: Last season stays on the hub instead of disappearing into an old folder.
Trust: Code Counts match the sheet; group film keeps positives and negatives apart.

Human Reflections

Year one was "can I grade film in a way that survives a Friday night and still makes sense on Monday?" Year two is "can the receivers open the same site I open, see last year next to this year, and know what the codes mean without a meeting?"

The grading system did not need a rewrite. It needed to stop being a private script. Putting How to score well on the season page sounds small until you remember that the whole point of the codes is player development — not a dashboard I am proud of.

The ingest work is the other half of that. If publishing a week takes twenty minutes of Python, I will skip a week when the game ends late. If publishing a week is "finish the tab, run Process week CSV," the hub stays current. Week 1 vs Holland going live this afternoon is the test I actually care about.

I am entering the second year with the same philosophy I started with: watch the film, log the truth, let the pipeline do the arithmetic. v2 just means the arithmetic shows up where the kids already look.

What's Next

This is still the GH Film Review Pipeline. v2 is the hub growing up around a rubric that already worked.

Next weeks of 2026–2027 should be boring in the best way: finish the tab, publish, check Top 5, send the group film PDF. If a new code appears on the sheet, it has to land in LEGEND_POINTS the same day or Code Counts will lie again.

If you are building a coach tool, the lesson I keep relearning is the same one as the attendance hub: do not replace the spreadsheet. Make the spreadsheet publish.

GH Film Hub v2: Year Two of the Grading System

Human Reflections

Building AI with AI Series