claudetools/session-logs/2026-06/2026-06-21-howard-security-assessment-scoring.md

# Session — security.azcomputerguru.com: posture scoring + gap/upsell findings engine

## User
- **User:** Howard Enos (howard)
- **Machine:** Howard-Home
- **Role:** tech

## Session Summary

Built out the ACG Security Assessment tool (`security.azcomputerguru.com`, the `security-assessment`
submodule) per Howard's brief: make it user-friendly, "better calculate," add more ACG-fit questions,
and surface upsell gaps to both onboard a customer and find service gaps. The tool was a clean
single-assessor PHP+MySQL wizard (Syncro lookup → risk-ordered questionnaire → 365/Google consent →
flat export) with NO scoring and NO gap→service mapping. Confirmed two design choices with the user up
front: scoring model = 0–100 risk score per domain + overall A–F grade; output = two views (internal
upsell worklist + client-safe risk report).

Designed a data-driven scoring engine: scoring rules live in `questions.json` (per-answer `risk`
weights + a `finding` with why/fix + the ACG `service` that closes the gap), and the engine is mirrored
in JS (live wizard) and PHP (export), both reading the same rules. Rewrote `questions.json` to v2 —
added `risk`/`finding` metadata to 25 fields, a `scoring` block (A–F grade thresholds, severity→weight
critical15/high10/medium6/low3, gapThreshold, `requiredControls`), and 6 new ACG-relevant questions
(3rd-party 365/Google backup, dark-web monitoring, DNS filtering, MDR/SOC, IT documentation, vCIO
cadence). Validated the engine in node against weak (0/F, 25 findings) and strong (100/A, 0) sample
clients.

Implemented the engine + UI in `index.php`: live posture chip (overall grade) in the topbar, per-domain
grade badges in the rail, and a rich "Posture & Findings" results view (overall A–F card, per-domain
score bars, prioritized findings each mapped to the ACG service, and an internal↔client-safe toggle,
with a collapsible full intake). Mirrored the engine in `api.php` (`score_assessment`) and rewrote the
export to render posture + findings, honoring `?view=internal` (default; shows ACG service per gap +
raw intake) vs `?view=client` (client-safe risk/recommendation report, no upsell language).

Then drove a 6-item todo list to completion of the code-doable work: responsive/mobile layout
(breakpoints; rail → horizontal stepper); replaced all `alert()` dialogs with an in-page saved-list
modal + toasts + light input validation; compliance-aware framing (when the client reports compliance
drivers or sensitive data, findings in `requiredControls` get a REQUIRED badge + a context banner).
Committed the build on `feature/posture-scoring-and-findings`, pushed, then (per Howard) merged
fast-forward to `security-assessment` `main` (`c82a3c9`) and deleted the branch. Two items remain:
GuruRMM endpoint prefill (deferred — infra) and the live IX deploy (gated on go + access).

## Key Decisions
- **Scoring rules as DATA in questions.json, engine duplicated in JS + PHP.** Keeps the wizard's live
  score and the export identical without a shared runtime; the only duplication is ~30 lines of engine
  in each language.
- **Risk score + A–F grade** (not CIS/NIST maturity or pure insurance checklist) — user choice;
  simplest + most sales-legible.
- **Two export views** (`?view=internal` default vs `?view=client`) — internal carries the ACG service
  per gap; client view strips upsell language and frames gaps as risks/recommendations.
- **Domain score only counts ANSWERED scorable fields** (unanswered = not assessed, not penalized) so a
  mid-consult partial intake still shows a meaningful posture; "Unsure" answers map to a mid risk frac.
- **Compliance framing via a `requiredControls` field-id list** (not per-framework rules) — when the
  client has compliance drivers/sensitive data, those baseline-control gaps get a REQUIRED badge. Robust
  + low-maintenance vs brittle per-framework mapping.
- **GuruRMM prefill deferred, not faked** — no Syncro→RMM client mapping exists in the RMM schema and no
  reachable read API from the public IX host; a fuzzy name-match could prefill the wrong client's counts
  into a client-facing report. Count fields stay manually editable, so nothing is broken.
- **Merged to submodule main, no auto-deploy** — DEPLOY.md is a manual cPanel upload, so merging to main
  does not change the live site; the live deploy stays a gated, confirmed step.

## Problems Encountered
- **Export Edit failed to match** — the live rows used unquoted `class=r`/`class=k`; re-read the exact
  block and matched it.
- **node `require('./questions.json')` failed** from a /tmp test script (require is script-relative, not
  cwd) — used an absolute path.
- **No PHP locally** — validated JSON + the wizard JS in node (engine is identical logic) and the export
  PHP via careful review + a bracket-balance check; final PHP gate is the live host.

## Configuration Changes
In the `security-assessment` submodule (committed `c82a3c9`, now on `main`, pushed):
- `app/questions.json` — v2: scoring rules + `requiredControls` + 6 new questions.
- `app/index.php` — scoring engine (JS), posture chip + rail badges, Posture & Findings view, responsive
  CSS, toast/modal, validation, view toggle.
- `app/api.php` — PHP scoring engine (`score_assessment`/`grade_for`) + rewritten export (posture +
  findings + `?view` internal/client + compliance banner).
- `README.md`, `DEPLOY.md`, `app/config.sample.php` (the last two carried pre-existing Jun 19 changes:
  howard@ added to the allow-list; DEPLOY tweaks).
No code committed to the ClaudeTools repo from this work (only this session log + the submodule pointer,
which `/sync` already tracks at `c82a3c9`).

## Credentials & Secrets
None created or discovered. The tool's config uses existing vault entries (referenced in
`config.sample.php`): DB `msp-tools/security-assessment-db`, Syncro `msp-tools/syncro-mike`; M365
Security Investigator app `bfbc12a4-f0dd-4e12-b06d-997e7271e10c` (read-only, multi-tenant). `config.php`
is gitignored / lives on the server.

## Infrastructure & Servers
- security.azcomputerguru.com — PHP + MySQL on the IX cPanel host (172.16.3.10), behind Cloudflare
  Access (Zero Trust app `8ce5f31c-4f4e-4883-bae1-f7606e5b06c0`; allow mike@ + howard@). DB `acgsec_assess` / user `acgsec_app`.
- Consent redirect: `https://security.azcomputerguru.com/consent-callback.php`.
- security-assessment repo: Gitea `azcomputerguru/security-assessment`; `main` now `c82a3c9`.

## Commands & Outputs
```
# engine validation (node)
node engine-test.js   # WEAK -> 0/100 F, 25 findings ; STRONG -> 100/100 A, 0 findings
# parse/lint checks
node -e "require('.../questions.json')"            # JSON OK (25 scorable, requiredControls 13)
new Function(<index.php <script>>)                 # wizard JS parses OK (200 lines)
# land
git checkout -b feature/posture-scoring-and-findings; git commit; git push -u origin <branch>
git checkout main; git merge --ff-only <branch>; git push origin main   # c3ca261..c82a3c9
git branch -d <branch>; git push origin --delete <branch>
```

## Pending / Incomplete Tasks
- **#1 GuruRMM endpoint prefill** — DEFERRED (infra): no Syncro→RMM mapping + no reachable RMM API from
  IX. Unblock: add a Syncro id to GuruRMM clients (or name-match + confirm) and expose a read-only RMM
  API reachable from IX with a key, then a config-gated lookup with graceful fallback.
- **#6 Deploy to IX + live smoke test** — GATED on go + access: manual upload of `app/index.php`,
  `app/api.php`, `app/questions.json` to the cPanel docroot (config.php already on server); then smoke
  test lookup, live posture/findings, and both export views behind Cloudflare Access. HOWARD-HOME may
  lack SFTP/SSH to the IX docroot (IX SSH key is from GURU-5070).
- Optional: number-field scoring (global_admins), client-report exec-summary/branding polish.

## Reference Information
- Submodule commit: `c82a3c9` on `security-assessment` main (claudetools pins it as of sync `68a05d3`).
- Export views: `api.php?action=export&id=<id>&view=internal|client`.
- Scoring: domain = 100·(1 − Σ(weight·riskFrac)/Σweight) over answered scorable fields; overall = domains
  weighted by points; grades A≥90 B≥80 C≥70 D≥60 else F; finding when riskFrac ≥ gapThreshold (0.5).
- Companion logs this session: `2026-06-21-howard-unifi-pfsense-control-verbs.md`,
  `2026-06-21-howard-gururmm-bug-018-019.md`.