Final Video Runbook¶
Use the production custom domain only after /api/status and the Mission package both report the
reviewed live snapshot. The recording must be shorter than three minutes and may use voice-over
only; TJ does not need to appear on camera.
Pre-recording gate¶
Do not use a snapshot ID, collection timestamp, count, or finding copied from this document. Read the current values from the reviewed production package immediately before recording. If evidence is stale or the displayed fleet aggregate cannot be explained, follow the protected refresh runbook before recording.
- Open
https://provifact.tmcoconsulting.com/api/statusin a clean browser profile. Confirmdata_modeisLIVE SANITIZED TENANT DATA,narrative_modeisopenai,modelisgpt-5.6-terra, andmodel_call_availableistrue. - Open
/assets/data/mission-control.jsonand compare itssnapshot_idwith/api/statusand the reviewed audit artifact. Do not show the raw JSON in the final video. - Confirm the package is current, the public scanner passed, and the browser console has no error.
- Close notifications, password managers, account menus, terminals, cloud consoles, and unrelated tabs. Use only public-safe pages.
- Ask one short Assistant question before recording to confirm the bounded live path. Avoid repeated chargeable calls.
Narrated script — target 2:50¶
0:00–0:20 — problem and provenance. Open Mission Control at the top. “Regulated endpoint teams should not reconstruct months of configuration history before every audit. Provifact joins approved intent in Git to read-only Intune observation, deterministic drift, sanitized publication, and evidence-grounded explanation.” Point to the live-data badge, collection time, snapshot, and fixed-model availability.
0:20–0:55 — action-first findings. Point to Approved Level 1 rules, Exact Intune joins, Deterministic drift, Implementation backlog, and Collection gaps. Read the current live values exactly as displayed; never use fixture counts in narration. Open the highest-priority genuine finding. Show observed value, approved target, assignment state, exact reviewed provider definition, evidence IDs, fingerprint, and read-only operator guidance.
0:55–1:20 — baseline and honest limits. Open Baseline Plan. “The approved demo baseline contains 98 pinned macOS Level 1 rules. Four Microsoft Intune provider mappings are currently reviewed; the other 94 are visible work to plan, not hidden and not mislabeled as failures.” Filter one baseline section, then show an evaluated row and a planning row. Mention that framework identifiers are cross-references, not a score or assessment verdict.
1:20–1:45 — change history. Return to Changes. If the current live package has a prior snapshot, show its exact new/resolved/unchanged result. If no manual Intune change was completed, say: “The workflow is ready for a later human change and second read-only collection; Provifact will not fabricate a resolved finding.” Explain that browser refresh checks only a newer published snapshot and never calls Microsoft Graph.
1:45–2:20 — actual GPT-5.6 answer. Open Provifact Assistant and ask What requires my attention?
Show the direct answer and click one evidence reference. “The Worker—not the browser—selects bounded
sanitized facts for fixed gpt-5.6-terra. Typed claims and evidence references are checked against
the deterministic package. Free prose remains generated analysis subject to human review.”
2:20–2:42 — security boundary. “Microsoft Graph is GET-only through GitHub OIDC. Raw tenant
responses and tokens are ephemeral. Production cannot fall back to synthetic evidence. The OpenAI
key is server-side, requests use store: false and no tools, and Provifact has no apply or Intune
write command.”
2:42–2:50 — close. “Provifact does not decide organizational compliance. It makes approved intent, observed state, drift, limitations, and evidence references reviewable before the audit.”
Exact click sequence¶
/evidence-dashboard/— point to the global provenance strip.- Select the first item under What requires attention now.
- In the dialog, point to observed → target, assignment, provider ID, evidence, fingerprint, and guidance; choose Close.
- Select Baseline matrix in product navigation.
- Choose Review details on a reviewed provider mapping; close the dialog.
- Return to Mission Control and select Changes.
- Choose Check for newer published snapshot and state that it does not collect from Intune.
- Open the lower-right Provifact Assistant launcher.
- Choose What requires my attention?, submit once, then open one returned evidence link.
- Close with the visible human-review/no-write methodology boundary.
Drift-to-resolution option¶
Use this only when TJ completed the separately approved low-risk Intune change and a second protected
audit used snapshot A as prior_sanitized_audit_run_id. Show snapshot B’s Resolved item, then ask
Assistant Which finding was resolved? The answer must cite current deterministic evidence. Never
describe an Intune change as resolved based only on the human action or Git history.
Privacy review checklist¶
- [ ] No tenant ID, client/object ID, policy/group/device name, serial, user, email, private domain, notification, account menu, browser history, API key, token, or cloud-console content is visible.
- [ ] The public snapshot ID and evidence pseudonyms are shown only from the scanned public site.
- [ ] The recording says TMCO Consulting; legal/approval copy uses TMCO Consulting, LLC.
- [ ] Live and fixture modes are never confused.
- [ ] No framework cross-reference is described as a pass, certification, or completed assessment.
- [ ] The Assistant answer shows human-review language and links back to deterministic evidence.
- [ ] The final video is public, under three minutes, and uses only authorized media.
Backup procedure¶
If the custom domain is temporarily unavailable, record the already reviewed production capture if
one exists. Otherwise demonstrate the credential-free local build below and label every frame
SYNTHETIC DEMO DATA; do not imply that the backup proves live Intune or OpenAI operation.
python -m evidenceops run-mission-demo --output-dir build/mission-demo
python -m evidenceops rebuild-static-demo
mkdocs build --strict
npm run dev
After recording, run /feedback in the primary Codex task and preserve its identifier privately.
Never commit the Session ID.