ada-tracker

Builds an accessibility compliance checklist for a project in two passes: scoping, meaning how many of each element must be accessible, and technical, meaning what each accessible element must measure

SKILL.md

GitHub

Accessibility Tracker

Accessibility compliance is two questions asked in order. How many of these must be accessible, and what must each accessible one measure. This builds the checklist for both. The questions are the same everywhere. The answers are not, so the answers come from you.

Scope and limits — read first

What this does. Takes your program, works through the scoping question for each element type, then the technical question for each accessible element, and writes a CSV where every row carries the clause you cited. Rows you could not supply a value for come out as unverified and are counted, so a partial checklist never reads as a complete one.

What this does not do.

  • It does not know any accessibility requirement. No scoping tables, no dimensions, no ratios, no counts. Every value comes from you. That is the design, not a gap.
  • It does not calculate a required count from a rule it knows. You give the rule, in your standard's words; it applies that rule to your numbers and shows the arithmetic.
  • It does not read your drawings. Element counts come from you.
  • It is not a one-time review. code-compliance-review checks accessibility as one axis of a dated review document; this builds the checklist you carry through the project and update as elements land.
  • It does not check an existing building against what is readily achievable, and it does not do a barrier removal survey.
  • It does not layer a local amendment over a base standard on its own. If two documents apply, you say which governs each row.
  • It does not replace the Architect of Record and does not anticipate the authority.

What you need before starting. The governing accessibility standard and its edition, your program counts, and the clause text for whichever rows you want checked.

Workflow

Step 1. Bootstrap if needed

Check whether docs/ada/index.md exists. If not:

python scripts/init_workspace.py

If the project has no docs/ folder at all, confirm before creating one.

Step 2. Establish the standard

Ask, in one message:

  1. Which accessibility standard governs, and which edition?
  2. Is a second document layered on top, from a state, province, or city? If so, which one governs where they differ?
  3. Which element types are in scope: parking, entrances, routes, doors, ramps, toilet rooms, dwelling or sleeping units, seating, signage, others?
  4. How will you supply values? Type them, paste the clause, or name a file.

Record the standard and edition. They print on every row of the checklist, because a dimension without its standard is not a citation.

Step 3. Collect the program

For each element type in scope, ask for the totals the scoping question needs. Typically the total count provided, and the facility characteristic the standard keys the rule to.

Never guess a count and never infer one from the project type.

Step 4. Scope, one element type at a time

For each element type:

  1. Ask for the scoping rule, in your standard's words. Accept a typed rule with its clause, a pasted table to read it out of, or a file to read.

  2. Apply it to the total, and show the arithmetic:

    Total provided        <n>
    Rule, as supplied     <written out>   [cited as ...]
    Required accessible   <arithmetic>    = <n>
    
  3. If no rule arrives, the row is unverified and no count is produced.

Do not round on your own initiative. Ask whether the rule rounds up, down, or to the nearest, and cite the answer.

Step 5. Build the technical checklist

For each accessible element, ask for the dimensional requirements that apply. Same three routes, same rule: nothing arrives, the row is unverified.

Write a row per requirement:

Item Required Cited section Source Status Remarks
  • Required: the value as supplied, in the units supplied. Quote the clause wording where it was pasted rather than paraphrasing it.
  • Source: where it came from, as a file and line, a pasted clause, or the user directly.
  • Status: cited, computed (a scoping count, with the arithmetic above), or unverified.

Step 6. Write the checklist file

Read assets/ada-checklist-template.csv and fill it. Filename:

docs/ada/<project-slug>-accessibility-checklist.csv

Slug rule: lowercase, [^a-z0-9-] removed, max 40 characters. Write a markdown preview alongside it with the first fifty rows.

The head of the file carries the standard and edition. The foot carries the unverified count.

Step 7. Update the index

python scripts/append_to_index.py \
  --project <slug> \
  --jurisdiction "<as supplied>" \
  --standard "<standard and edition, as supplied>" \
  --row-count <n> \
  --csv <filename>.csv \
  --preview <filename>.md

Always pass --standard explicitly. The script has a default, and a checklist that inherits a default standard is a checklist nobody can trace.

Step 8. Report

Counts: rows cited, rows computed, rows unverified. The unverified list with what each one needs. Then:

Every requirement in this checklist came from you. I applied your rules to your numbers and supplied no accessibility values. Unverified rows were not checked at all.

Rules

  • Never supply an accessibility value. Not a count, not a ratio, not a dimension, not a clearance. There are none in this file and none bundled with it.
  • Show the arithmetic on every scoping count, so a wrong rule is visible rather than buried in a number.
  • Ask how the rule rounds rather than choosing.
  • Quote clause wording where it was pasted. A paraphrased accessibility requirement loses the exception that mattered.
  • Print the unverified count on the file, every time.
  • Where two documents apply, the user says which governs each row; never merge them.

Anti-patterns

  • Filling a clearance, a width, or a required count from recall because the figure is a familiar one. These are the most familiar numbers in the whole catalog and therefore the most dangerous to recall.
  • Searching the web for a scoping table and applying what comes back.
  • Producing a required count without showing the rule and the arithmetic.
  • Dropping unverified rows so the checklist reads clean.
  • Merging a base standard and a local amendment without being told which governs.
  • Bundling a scoping table or a dimensions file into this skill later. It had those once; they were removed on purpose, along with the script that computed counts from them.

Resources

  • assets/ada-checklist-template.csv — the checklist columns.
  • assets/ada-index-template.md — the project index structure.
  • scripts/init_workspace.py — creates docs/ada/ and an empty index.
  • scripts/append_to_index.py — appends a row to the index.

No reference files. The requirements come from your standard, not from here.