# Evidence-Only Capability Statement: every line traced to proof

> A one-page federal capability statement your agent can draft without inventing anything. Every competency, past job and status sits in a claim ledger with the file that proves it. A checker traces each line back, refuses invented figures, stale SAM data, uncertified set-aside statuses and praise words, and your signature is tied to the exact text.

- URL: https://paperchasewebb.ai/bots/marketplace/items/evidence-only-capability-statement-every-line-traced-to-proof/
- Type: Prompt pack
- Category: Solo business
- By: PaperChaseWebb
- Works with: Claude Code, Codex, Grok Bot, Hermes (named in the listing); Three Python scripts, a claim ledger and company record template, a worked example, AGENTS.md and CLAUDE.md rules, a skill file, a Grok Bot standing order and Codex and Hermes notes
- Price: $24 USD
- Difficulty: Beginner
- Setup time: 20 min
- Time to result: A first checked draft in about an hour, most of it gathering proof files
- Format: Markdown (.md) file, evidence-only-capability-statement-every-line-traced-to-proof.md
- Version: 1.0, updated 2026-11-03
- License: Use inside your own company, including when you prepare statements for clients. Do not resell or republish the kit itself.

## What you end up with

A capability statement where every line has a proof file behind it, the held-back claims are listed, and you signed the exact version that goes out.

## What you get

- cap_check.py: 26 error checks and 4 warnings on the statement, ledger and company record, read-only
- cap_draft.py: a first statement drafted only from evidenced and partial claims, every line tagged
- A claim ledger template with evidenced, partial and not-evidenced statuses
- A company record template for SAM.gov data, statuses and the owner's signature
- A fictional worked example that checks CLEAN, with two claims held back
- The six-step method and the list of what the checker refuses
- AGENTS.md, CLAUDE.md, a capability-statement skill, a Grok Bot standing order, Codex and Hermes notes
- 47 tests, one planted fault per check

## Who it is for

Small federal contractors who want a capability statement that says exactly what their records show, drafted fast and signed by a person.

## Tools needed

- Python 3.8 or newer: the tracer; standard library only
- Any text editor or spreadsheet: fill in claims.csv and company.json; no account or login
- Grok Bot, Claude Code, Codex or Hermes (optional): interviews you for claims and drafts under AGENTS.md; never fetches identifiers, signs or sends
- You need: Your SAM.gov entity record, read by you
- You need: Your certification letters, if any
- You need: The contracts, invoices or letters behind each claim
- You need: Python 3.8 or newer

Cost to run: The kit is the only purchase. SAM.gov registration is free; the kit never logs in anywhere.

## How it works

1. **Run the tests.** Unpack the kit and run python3 -m unittest discover -s tests. Expected: 47 tests pass and the example checks CLEAN.
2. **Fill the record from SAM.gov.** Open your SAM.gov entity record and each certification registry yourself and copy what you read into company.json with today's date. Expected: Identifiers, statuses and dates your agent did not guess.
3. **Build the ledger.** One row per claim in claims.csv with its proof file, marked evidenced, partial with a qualifier, or not-evidenced. Expected: Every claim you want to make, graded against a file.
4. **Draft, check, sign.** Run cap_draft.py, edit the wording, run cap_check.py until only the approval checks remain, then sign the --hash value in owner_approval. Expected: CLEAN, with your signature on the exact text.

## Sample prompt: Build the first statement (Grok Bot, Claude Code, Codex, Hermes)

```
Follow AGENTS.md in this kit. Interview me for claims one at a time, add each to claims.csv with the proof file I give you, then run cap_draft.py and cap_check.py. Stop with the checker output, the held-back list, and the hash for me to sign.
```

Prompt 1 of 2.

## Sample prompt: Refresh before a submission (Grok Bot, Claude Code, Codex, Hermes)

```
Run cap_check.py on this folder with today's date. List every check that is stale or failing, and the exact proof or recheck each one needs. Change nothing.
```

Prompt 2 of 2.

## Free preview of evidence-only-capability-statement-every-line-traced-to-proof.md

````markdown
# Evidence-Only Capability Statement

**A one-page capability statement where every line points to a file that proves it. Your agent drafts it, a script checks it, and nothing goes out until you sign the exact version.**

Version 1.0, 2026-11-03. By PaperChaseWebb Inc. Works with Grok Bot, Claude Code, Codex and Hermes, or by hand. Python 3.8 or newer, standard library only.

## The problem

A capability statement is the page a contracting officer reads in under a minute. Agents write them fast and write them generously: a rounded-up headcount, a customer you never got permission to name, "industry-leading", a set-aside status the company does not hold. Each one reads fine until someone checks.

Some of those are not just weak copy. Under 13 CFR 121.108, registering in a federal database as a small business counts as a deliberate certification of size, and misrepresenting it carries civil and criminal penalties. Since January 1, 2024, a firm must hold SBA VetCert certification to win an SDVOSB set-aside or sole-source award (13 CFR 128.200(c)). A statement that says more than your records do is a risk, not a draft.

## What you end with

1. **A claim ledger.** Every competency, past job and differentiator in one CSV row, with the file that proves it and a status: evidenced, partial or not-evidenced.
2. **A statement drafted from the ledger only.** Not-evidenced claims stay out and are listed for you. Partial claims carry their qualifier.
3. **A checker that traces every line.
````

## Test record

- Date: Nov 3, 2026
- What was run: tests/test_cap_check.py and tests/test_cap_mini.py (47 tests) on Linux with Python 3.8, 3.9, 3.11, 3.12 and 3.13; the full kit and the free sample unpacked from their files and run fresh.
- Result: 47 passed, 0 failed on all five versions; each check caught by its own planted fault; the unpacked example checks CLEAN.
- Not run: macOS or Windows. A live Grok Bot, Claude Code, Codex or Hermes session. A real company's SAM.gov record.

Copied from this item's dated test notes. What was not run stays in.

## Worth knowing

- The checker makes sure every line has a proof file and says no more than the ledger. A person still reads the files to confirm they say what the claim says.
- It does not check SAM.gov or SBA registries for you. You read them and enter the dates; the checker fails data over 30 days old.
- Not legal advice. For a size or status question on a specific bid, ask your local APEX Accelerator or counsel.
- The output is clean Markdown. Lay it out in your own one-page template.
- On 2026-11-03: 47 tests passed on Python 3.8, 3.9, 3.11, 3.12 and 3.13 on Linux. Agent rules guide; they are not a lock.

## Questions

**How is this different from the Veteran GovCon AI Kit?** The Veteran GovCon AI Kit helps you draft a capability statement and shape-check it: drafting prompts, a SAM.gov profile checklist and capstat_check.py for shape and wording. This kit ties every line to a proof file, runs 26 checks against the claim ledger and binds your signature to the exact text.

**Can my agent write the statement?** Yes. It interviews you, fills the ledger and drafts. It cannot fetch your identifiers, mark its own work approved or send anything; the checker fails a statement the drafter approved.

**What if I have no federal past performance yet?** Use commercial work with a contract, invoice or reference letter, and title the section Relevant experience. The checker warns you instead of letting you imply federal work.

**Does it work on Windows or a Mac?** Plain Python with no installs; tested on Linux so far.

## How to buy

$24 one time at https://paperchasewebb.ai/bots/marketplace/items/evidence-only-capability-statement-every-line-traced-to-proof/. Stripe checkout, then an instant download page; the download link is also in your emailed receipt.

Seller: PaperChaseWebb Inc., mgmt@paperchasewebb.com