Skip to content

FedRAMP 20x Class C: every applicable rule Implemented or Not Applicable - #501

Closed
sam-aydlette wants to merge 123 commits into
mainfrom
feat/frr-records
Closed

sam-aydlette wants to merge 123 commits into
mainfrom
feat/frr-records

Conversation

@sam-aydlette

@sam-aydlette sam-aydlette commented Nov 5, 2026 •

Copy link
Copy Markdown
Owner

SCN-Type: adaptive
SCN-Rationale: carries the rule-completeness change to main: two IAM roles for unattended publishing, runtime-emitter grants, CloudFront access logging and a DAST scanner user; no new external service, no federal data, no class change.
SCN-Record: SCN-2026-007

This is the one pull request to main for the FedRAMP 20x Class C rule work. It carries the rule register and its deploy gate, plus everything merged into feat/frr-records:

PR Contents
#503 Certification package and quarterly cycle in FedRAMP's schemas (CPO, OCR, SDR, CCM)
#505 Vulnerability detection, evaluation and reporting
#508 Notifications: SCN, incidents and the simulated FedRAMP Security Inbox
#509 Trust center, scope and the simulated assessor
#510 Integration of all groups; every rule Implemented or Not Applicable
#511, #512 Fixes for the CodeQL alerts the combined change raised

Each of those PRs describes its own changes, verification and residuals.

Changed

  • Rule register data/frr-records.json: one record per applicable 20x Class C provider rule (154), derived from the pinned corpus by scripts/_rules.applicable_provider_rules(). scripts/_frr.py validates it and fails closed on:
    • a missing, extra or duplicate record;
    • "Implemented" without evidence, or evidence that does not resolve (a test that exists, a workflow step, a freshness-governed /.well-known/ artifact, or an in-force DEC-NNN);
    • Not Applicable outside a MAY permission not exercised or a subset FedRAMP scopes to a step not taken (Marketplace listing, applying);
    • a MAY rule without may_rationale;
    • record text carrying an AWS account ID or a personal email address.
  • Deploy gate: scripts/_frr.py --complete runs in the unit-test job, which the deploy job needs. It fails while any record is Partially or Not Implemented, so nothing deploys with a gap. make check-complete runs it locally.
  • simulated {counterparty, how}: a record that relies on a simulated outside party (assessor, FedRAMP, agency, incident recipient, mail, SAM.gov) says so. It is allowed only on Implemented records with deploy-time evidence, and the SDR publishes the count. Nothing is sent to FedRAMP or anyone.
  • Records describe the state after a successful deploy of their commit. Anything the deploy depends on fails the deploy before the SDR is built.

Effect

Rule records, from python3 scripts/_frr.py:

Status main this PR
Implemented no register 141
Partially Implemented 0
Not Implemented 0
Not Applicable 13
  • Not Applicable (13): 6 MAY permissions not exercised; MKT-CSO-MLR and MKT-CSO-PML (no Marketplace listing sought); FRC-APP-MLF, AFC, FIA, FCP, NTP and USA (not applying). Each states its reason.
  • Simulated counterparties: agency 4, assessor 7, FedRAMP 2, incident recipient 4, mail 7, SAM.gov 1.
  • After deploy, the site publishes the CPO, SDR, OCR, VDT, AVI and history in FedRAMP's 2026-06-24 schemas. The SDR lists all 154 rules, 46 KSIs (each with at least two automated methods, except KSI-CED-RAT, accepted in DEC-191) and the 323 Moderate controls. The SCN feed, incident reports, the inbox simulation and the trust center's access and availability views are published alongside.
  • Removed from published output, each with a gate: AWS account IDs, IAM user names, GitHub run IDs and the personal email address (the security contact is now the operator's LinkedIn profile).

Verified

  • On this PR's head: every CI check passes, including the unit tests (the --complete gate among them), OPA compliance, Checkov, the security scan, CodeQL with no open alerts, and scn-tag.
  • Locally on feat/frr-records: make check passes (1847 tests, 1 skipped); make check-complete exits 0; Regal, check-policy-annotations, check-exceptions and check-wasm-safe pass.
  • Every gate invariant added across the stack has a unit test and a known-bad fixture.

Residual

  • The Sigstore bundles and provenance envelope carry the GitHub run URI in the signing certificate; that is part of keyless signing.
  • During each deploy, the staged inventory and runtime signal sit briefly under deploy-staging/<sha>/ in the website bucket and may be served by the CDN until deleted. They hold the same redacted content the deploy publishes.
  • Evidence resolution shows that evidence exists; FRC-CSX-VVR ties the cited tests to their results in the conformance run.

Needs human

  1. Complete the ordered setup checklist in Integration: all groups combined; every rule Implemented or N/A #510's description: bootstrap applies, repository variables, environments, the SILK_DAST_* secrets and the labels. The deploy's header, logging and availability checks fail until they are done.
  2. Review the AI-drafted text: the 154 records, the decision records listed in Integration: all groups combined; every rule Implemented or N/A #510, SCN-2026-007 and DEC-193.
  3. Merge, let CodeQL run on main, dispatch a ZAP run, wait for the first availability-probe run, then approve the deploy.

Couldn't verify

  • Anything that needs AWS or the deploy itself: the bootstrap and main-stack applies, the staged runtime revalidation, the simulations' deploy steps, and the published output. verify-published.sh and storycheck run against the live site after the deploy.

🤖 Generated with Claude Code

sam-aydlette and others added 3 commits November 5, 2026 13:32
…n is shown

data/frr-records.json holds a record for each of the 154 FedRAMP rules that
apply to a 20x Program Class C provider, derived from the pinned corpus's own
subset applicability metadata (scripts/_rules.applicable_provider_rules).
scripts/_frr.py fails closed on a missing or extra record, an Implemented
record without evidence, evidence that does not resolve (test, workflow
step, freshness-governed artifact, in-force decision), a non-MAY rule left
optional, or published text carrying an account ID or a foreign email.

Records are AI-drafted from the 2026-11-05 assessment, corrected for W0/W1
and #496, for the operator's review.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A MAY record now requires may_rationale and any other record refuses it;
the text is screened like the rest. The two quarterly-review MAY rules the
operator intends to exercise move from optional to W4.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A record may be Not Applicable, with its reason, only for a MAY permission
not exercised or a rule in a subset FedRAMP scopes to a step the provider
is not taking: Marketplace listing (MKT-CSO) and applying (FRC-APP). A
test checks that scoping phrase is FedRAMP's own subset description.
"optional" is no longer a closer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sam-aydlette and others added 13 commits November 5, 2026 14:32
…ence-only history, period activity

VER-RPT-VDT: carry the historical rating and rating history (providerExtensions;
the schema has no field), take the next reduction from a decision's new
planned_reduction before the remediation deadline, populate supplementary
information from the decision, explain overdue status rather than the
suppression, and date a record from a decision that names it (the manual ZAP
scans behind DEC-039/040).

VER-RPT-AVI: a .checkov.yaml suppression resolves to the resources its decision
names, so its IRV is decided from the inventory; suppressed checks are dated by
their decision.

VER-TFR-MRH: a record that leaves the VDR is Remediated only from evidence (a
closed decision, a false-positive suppression, or its scanner running again);
otherwise it is recorded as no longer reported with the reason and no final
disposition. Closed decisions are backfilled.

VER-RPT-PER: each VDT/AVI carries an activity summary for its period, including
scans with times and the runtime V&V run, linked.

Ratings, deadlines, overdue status and blocking are unchanged on the live inputs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…mulated counterparties flagged

scripts/_frr.py --complete fails the unit-test job (which the deploy job
needs) while any record is Partially or Not Implemented. A record relying on
a simulated outside party carries simulated {counterparty, how}, allowed only
on Implemented records with deploy-time evidence. Known-bad cases and tests
for each invariant.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ed rules

The vendored ksi-catalog.json predated FedRAMP's 2026-10-13 mapping revision:
40 of 46 KSIs mapped fewer (or different) controls than the pinned
Consolidated Rules. scripts/sync-ksi-catalog.py now derives the KSI block
from the pinned rules (Class C statements where a KSI varies by class,
titles from the NIST Rev5 catalog), and a test fails while they differ.

Within the Moderate baseline no implementation status changes; SC-45 and
SC-45.1 become shared. The projection reaches 59 controls beyond Moderate
(was 10). s3.rego drops si-12, which no KSI maps any more. The SSP parity
test runs on the catalog as it stood at the SSP's retirement.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every applicable 20x Class C rule, every KSI and every Rev5 Moderate
control, each no stronger than what backs it; semver versioning; consumers
moved to the schema shape.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…eploy

SDR-CSO-MTD Implemented; SDR-CSO-FRR, SDR-CSX-KSI, FRC-CSX-VVK and
FRC-CSX-VVR gaps restated for the post-deploy state.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…notifications and deadline checks

Every change now carries an explicit significant-change evaluation (SCN-Type,
SCN-Rationale, and SCN-Record for adaptive/transformative/emergency); a
missing tag fails instead of defaulting to routine-recurring, and class-change
blocks. The deploy appends every merged evaluation to a signed, append-only
log and publishes every register row in FedRAMP's SCN schema with a Markdown
view and an Atom feed, listed in the CPO. Business-day deadlines (US federal
holidays) are read from the vendored rules; a scheduled workflow fails on
anything overdue. Notification is publication only; nothing is sent.

SCN-Type: not-significant
SCN-Rationale: change-management process and compliance reporting only; no component, boundary, data flow or access change to the system.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ard)

SCN-Type: not-significant
SCN-Rationale: input validation in the new SCN log builder; no system component, boundary or access change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…torical notices

The record describes the system after a successful deploy of this commit: the
deploy builds and publishes every notice (any failure fails the deploy; the
round-trip check compares the served index and feed), so the mechanism is in
force. SCN-2026-001..005 were published after their deadlines and are named.

SCN-Type: not-significant
SCN-Rationale: rule-record text and a test only; no component, boundary, data flow or access change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…/NRI triggers

- data/incidents.json + schema + fail-closed loader (scripts/_incidents.py):
  every incident carries its FedRAMP reportability evaluation and reason;
  default N5 until an IEC-CSO-EFI estimate (checked against the rule's rubric).
- scripts/build-incident-report.py renders Initial/Ongoing/Final reports in
  FedRAMP's incident report schema (validated before write) and the incident
  log; Class C clocks and the rating scale are read from the pinned rules.
- policies/triage.rego incident_triggers: VER-TFR-IRI and VER-TFR-NRI, with
  thresholds parsed from the rule text (fail closed on a changed text).
- Reconciliation invariant (m): an open incident's findings are in the VDR.
- incident-reports.yml publishes unattended (publish-only; nothing is sent);
  bootstrap role scoped to .well-known/incident-*; the deploy also publishes;
  watchdog metric and alarm for overdue reports; freshness entry.
- Dated exercise (2026-11-05, exercise: true): IIR -> 2x OIR -> FIR, plus a
  not-reportable evaluation; run end to end in tests.
- FRR records: EFR, DPR, EFI, VER-TFR-IRI, VER-TFR-NRI Implemented; IIR, OIR,
  FIR, AIR Partially Implemented (publish-only; bootstrap apply pending).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s state the deployed system

The deploy fails before applying anything unless AWS_INCIDENT_REPORTS_ROLE_ARN
names the bootstrap incident-reports role and the incident-reports environment
deploys only from main (scripts/check-incident-publisher.py; known-bad fixture
with the variable missing). After the deploy job, the deploy calls
incident-reports.yml (now also workflow_call) on its own role, passing only the
CloudFront secret, so a green deploy proves the publisher publishes.

IEC-CSO-IIR/OIR/FIR/AIR gaps now state only what holds after a successful
deploy: published, not delivered to FedRAMP or agencies, by the operator's
decision.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The inbox is simulated for this proof of concept by the operator's decision:
nothing is received or sent. scripts/_fsi.py is the inbox logic
(classification, designators, Class C deadline, register, header-safe
acknowledgement sent once, paging). Every deploy runs it over committed
synthetic messages before the SDR build, fails on any unexpected outcome, and
publishes the signed, dated result as fedramp-inbox-simulation.json, labelled
as a simulation. The published contact says so and points to LinkedIn; the
AFC-CSO-NOC check applies to that contact. AFC records are Implemented with
simulated {counterparty: mail}.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e SDR

Every deploy and every unattended publish now "delivers" each incident report
the log lists to a simulated FedRAMP and agency recipient
(scripts/deliver-incident-reports.py): read back, checked against the
published sha256 and FedRAMP's schema, and recorded in the signed delivery log
/.well-known/incident-deliveries.json. Nothing is sent. A failed delivery
stops publication.

The deploy job assumes the incident publisher's role before it builds the SDR
(scripts/check-incident-role.py), so a deploy whose publisher role does not
work stops before publishing records that claim it. The role's trust gains the
reviewer-gated deploy environment (bootstrap apply).

IEC-CSO-IIR/OIR/FIR/AIR are Implemented with simulated {incident-recipient}.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sam-aydlette and others added 11 commits November 5, 2026 16:48
…y; SDR carries it; cited tests must pass

Totally Real Assessment Services, LLC (simulated) runs as a reusable workflow
before the deploy job, works from the generated assessor guide, re-runs every
cited check, and signs one set under its own workflow identity, which
infrastructure/keys/assessor-identity.json pins. The SDR build requires and
verifies the set. The unit-test job fails when a test cited as rule evidence
did not pass (FRC-CSX-VVR).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ds in allocated blocks

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ds in allocated blocks

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ence-only history, period activity

VER-RPT-VDT: carry the historical rating and rating history (providerExtensions;
the schema has no field), take the next reduction from a decision's new
planned_reduction before the remediation deadline, populate supplementary
information from the decision, explain overdue status rather than the
suppression, and date a record from a decision that names it (the manual ZAP
scans behind DEC-039/040).

VER-RPT-AVI: a .checkov.yaml suppression resolves to the resources its decision
names, so its IRV is decided from the inventory; suppressed checks are dated by
their decision.

VER-TFR-MRH: a record that leaves the VDR is Remediated only from evidence (a
closed decision, a false-positive suppression, or its scanner running again);
otherwise it is recorded as no longer reported with the reason and no final
disposition. Closed decisions are backfilled.

VER-RPT-PER: each VDT/AVI carries an activity summary for its period, including
scans with times and the runtime V&V run, linked.

Ratings, deadlines, overdue status and blocking are unchanged on the live inputs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… dates for DEC-007/009/010 from git history

The CKV_AWS_117 skip is global, so DEC-010 covers every Lambda function; it named
a resource no inventory component carries. The inline skips for CKV_AWS_68,
CKV_AWS_86/310 and CKV_AWS_117 were added on 2025-06-14 (commit f453e9d), so each
was detected by then. AI-drafted for the operator to confirm.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…VDT (VDR-CSO-AKE, VDR-TFR-KEV)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d disclosure per finding

policies/triage.rego:
- N0 for a fully mitigated finding, read from VER-EVA-EPA's scale (VER-EVA-EPA;
  REQ-PAI-1). N0 has no timeframe.
- Groups (VER-EVA-GRV): a finding's group (CVE, else decision, else check) is
  evaluated at its most exposed open member, on the clock of its earliest
  member; the group's deadline binds every open member (known-bad fixture: a
  new instance of an old group is past the group's deadline and blocks).
- VER-TFR-MAV "will not be": a decision's planned remediation past day 192
  accepts the finding now (accepted_basis).
- VER-TFR-EVU: evaluation_late when first evaluated more than the rules'
  window after detection, or still on a default after it.

scripts/_evaluation.py: the evaluation record per finding (criticality,
reachability, exploitability, detectability, prevalence, privilege, proximate
findings, known threats, automatable, exploitation); the disclosure
determination and the redaction it drives (VER-RPT-NID/RPD). A decision record
may carry automatable {value, evidence}; 'no' without evidence is refused.

The golden triage report is regenerated: the only changed values are the three
fully mitigated findings (N1 -> N0); every other change is an added field.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…oy (MAS-CSO-IIR/FLO/MDI/TPR, FRC-CSX-MAS)

- Inventory: the app's browser origins (jsDelivr, Google Cloud Storage) and the
  VDR's remaining feeds (FIRST EPSS, Grype database) become external services;
  the CloudTrail trail joins the logging KSIs' evidence domain.
- scripts/build-mas.py publishes /.well-known/minimum-assessment-scope.json
  (+ .md, signed, freshness-governed): information resources, every CSP
  off-origin host placed or out of scope with a reason, flows with a
  high-water-mark security category, a metadata inventory, the third-party
  records with FedRAMP's four fields, and the KSI-to-scope application. It
  fails the deploy, before the SDR, on any gap.
- Classification: 17 new flows covering every external service and data store;
  an app-delivery group (and its service in the service model).
- Third-party records: the four MAS-CSO-TPR fields on all 11; decision fields
  AI-drafted for the operator to confirm; a decided record without them fails.
- scripts/_tpr.py: the CPO's thirdPartyInformationResources, for the CPO hook.
- Five rule records flip to Implemented with resolving evidence.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
providerExtensions.fedRampId, read from data/cds/offering.json fedramp_id (the
field _offering.fedramp_id() reads on feat/sdr-schema). A missing ID stops the
build. Adding these files to check-fedramp-id.py's list is left for integration.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ng standard; DEC-165..167

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sam-aydlette and others added 25 commits November 6, 2026 12:39
- Assessor binding (IVV-CSO-FIA, IVV-CSX-AIA, IVV-CSO-ICP, SDR-CSO-FRR):
  each simulated assessment set names the commit and KSI signal it covered;
  _assessor.load_assessment refuses any other. The deploy reads the live
  signal itself before the SDR is built, and the SDR states what the
  pre-deploy assessment covered. A post-deploy assessment of the
  just-published state runs after publish and fails the run on any failure.
- Append-only histories: the access summary, certification report index and
  submission log are fetched fail-closed (only a 404 starts a history).
- CDS-CSO-FRC: the simulated FedRAMP issues a stand-in report only in
  response to a submission (first, after an adaptive or transformative SCN,
  annually), logged with its received date; publication past 2 weeks fails
  the deploy. Reports say NOT ISSUED BY FEDRAMP and not certified; the CPO
  states certification.status "not certified (proof of concept)".
- AFC-CSO-NOC: the deploy files inbox-change notices with the simulated
  FedRAMP (log entries only) and fails unless each is within one business day.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ource

The SDR's providerExtensions.certification becomes {program, status}, read
with the CPO's from scripts/_offering.py; the trust center JSON carries the
same and its page reads it. submit-package.py refuses an SDR or CPO without
the status.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Bind assessments to what they covered; enforce append-only histories, CDS-CSO-FRC and AFC-CSO-NOC
Removes the operator log review from DEC-181 and the VDR-CSO-RES record
(operator's direction, 2026-11-06: no recurring manual work). The before
text is kept under source.reworded. New screen: no in-force decision or
rule record commits the operator to a manual review, with a known-bad
fixture.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n the guide

- simulated-assessor.yml: the assess job verifies every bundle with cosign
  (assess.py cosign_verify) but no longer installed it after the signing
  split; install it (verify only, no id-token). New test: every job that runs
  cosign installs it first, with a known-bad workflow fixture.
- secure-configuration.json 2.3.0: the incident reports role's recommended
  trust adds the prod environment the bootstrap already grants for the
  deploy's role check; the exact-set comparison would have failed the deploy.
  New test: the guide's trust for the maintenance and incident roles equals
  the bootstrap Terraform's, with a known-bad case.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…le text

- deploy-with-opa.yml: the OCR, SDR and CPO previous copies (version, KSI
  metric and training history) are fetched with fetch-published-history.py:
  a 404 starts the history, any other failure stops the deploy. Wiring test
  extended.
- notify also watches the unit tests (which hold the rule-completeness gate)
  and the security scan on main, where their failure skips the deploy rather
  than failing it; on a pull request their failure is the PR's own check.
- DEC-025: the hardware-key purchase plan is withdrawn (DEC-150 accepts
  virtual TOTP), kept under source.plan_withdrawn, so the OCR stops
  publishing it as a planned change. Test added.
- Text that contradicted the records: continuous-monitoring-plan (the
  Quarterly Review and feedback rules run with simulated parties, not N/A),
  rules-of-behavior (changes reach main through pull requests),
  VDR-CSO-DET / VDR-TFR-PCD (non-machine resources verified every deploy
  and night), CLAUDE.md invariant (h), DEC-026's POAM-025 citation, the
  incident-response AWS credential step (CI holds no keys), README script
  count and the exact cosign identity, a workflow comment citing DEC-191.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The trust manifest and the CDS-TRC-HMR / CDS-CSO-CBF records said every
published JSON has a .view.md. Per-incident reports, certification reports
and the inventory schema (deploy), the frozen issue, snapshot and Quarterly
Review files (package maintainer) and per-incident reports (incident
publisher) had none, and the manifest listed views only from the deploy's
own folder. Each publisher now builds, gates and puts the view of every JSON
or CSV it puts; the manifest lists views from the publisher table in
scripts/_signers.py and fails if a JSON is built without its view. The two
maintainer-only indexes take the nightly (scheduled unattended) cadence.
A test runs each workflow's view, gate and publish steps with stubbed AWS
and fails on any JSON put without its view; known-bad fixture included.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The notice deadline ran from the history entry's `since`, and a failed
deploy publishes nothing, so each later deploy filed later and the
append-only history could not be corrected: a permanent block. A change now
takes effect when a deploy first publishes it. The submission log records
that deploy (inboxContact: date and commit), each notice carries it
(publishedOn, publishedIn), and the deploy fails only when a notice was
filed more than one business day after the publishing deploy. Regression
test (a change first deployed ten business days after its date passes) and
a known-bad notice filed two business days after its publishing deploy.
Record and runbook text updated (AI-drafted).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scripts/check-published-redaction.py now fails any published JSON, CSV or
Markdown that carries an email address not at samaydlette.com (or a
subdomain), counting without printing the address. Domain labels must start
with a letter, so package versions and workflow refs are not read as
addresses. Known-bad fixture with an example.com address.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The package submission log, the certification report index, the inbox
simulation, the incident delivery log, the incident log, every incident
report (providerExtensions.engagement, allowed by FedRAMP's schema) and the
SCN index now carry the no-engagement statement verbatim. The simulated
certification reports themselves are left as issued: each is published
byte-identical to the logged copy, so the statement goes on the index and
the log. A test builds each output and fails if one lacks it; known-bad
fixture included. The delivery known-bad fixture's report hash follows the
changed report bytes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tools/boundary-map/corroborate_checks.py expected the deploy identity for
every deploy-cadence file and the evidence nightly's for every nightly one,
so files the package maintainer, the incident publisher or the simulated
assessor sign were reported as signed by an unexpected identity. Each
file's allowed identities now come from scripts/_signers.py (held to the
publish steps by tests/test_publishers.py); the TAP-side verifier tries
each allowed identity. Known-bad fixture: two files signed by workflows
that never write them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The register row and the impact analysis list the further engine changes
(DEC-181 automated sources and the manual-work screen, fail-closed package
history, failure alerts on tests and security scan, complete views from
every publisher, the email gate, the DEC-194 statement on simulated
outputs, writer-based corroboration) and the AFC-CSO-NOC clock now running
from the publishing deploy. AI-drafted.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DEC-181 automated sources, first-deploy blockers, and final-review fixes (views, NOC clock, email gate, DEC-194 outputs)
Every nightly reads CloudTrail write events and GitHub changes to main,
workflows, rulesets and environments, evaluates each against what the
pipeline is expected to do, and records unexpected events as process
findings. A planted, plainly simulated intruder event must be caught on
every run. The nightly role gains a read-only LookupEvents grant
(bootstrap apply), compared by the secure-configuration check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Annual Security Review (with the rules-of-behavior acknowledgement)
and the monthly recovery exercise (annual tabletop in November) are built
by scripts with a simulated cast; an overdue check fails the run from the
day after a due date with no review published. Recurring costs are
AI-drafted estimates for the operator to correct.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Wires the audit-log review, recovery exercises and security reviews into
the nightly and maintainer workflows: signer identities, freshness,
FedRAMP ID and redaction gates, views, bootstrap object keys, published
verification, and VDR monthly activities generated from run records.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
build-sdr.py replaces the KSI-CED-RAT special case with a KSI-to-method
table: the training checks, plus the audit-log review (MLA-RVL, CMT-RVP),
the Annual Security Review (PIY-RIS, PIY-RES, SVC-EIS, INR-RPI) and the
recovery exercise (RPL-TRC), each counted only when its published record's
signature verifies against its publisher and the record is current and
passing. The deploy and the maintainer rebuild pass the records in.

Tests with committed known-bad fixtures cover the planted intruder, an
unknown principal, a direct push, a restore mismatch, a configuration gap,
a tabletop gap, an unnamed new component type, invalid cost lines, the
overdue and acknowledgement checks, the nightly read grant, and parity of
the monthly VDR report with its previous output.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DEC-195 names the only recurring acts left to the operator (the monthly
boundary corroboration, decision review_by lapses, deploy approvals and
the 30-day deploy freshness windows) and the mechanisms that replace every
other recurring review. DEC-019 (formerly POAM-019) is restated as
rotate-on-compromise; DEC-032, DEC-166 and DEC-180 name their mechanisms.
Each rewording keeps its before-text under source.reworded.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Policies, plans, procedures, the control statements, the VDR suppression
rationales and the data records now name the script or workflow that
performs each review instead of committing a person to it. Hand-written
entries are kept under sections headed as history. AI-drafted.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scripts/_nonmachine.manual_work screens every sentence for a person, a
review-like act and a recurring cadence (or a review-cadence line); the
non-machine verification runs it on every listed document, and
tests/test_no_manual_work.py runs it over docs/, the README, in-force
decisions, rule records, control statements, VDR rationales and data
records. Only sentences citing DEC-195 and sections headed as history are
allowed. Known-bad fixtures cover the active and passive forms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds the nightly audit-log review, the monthly recovery exercise, the
Annual Security Review, the overdue check, the SDR method counting and
the manual-work gate to the register row and the security impact
analysis, with the nightly role's read-only LookupEvents grant.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Automate the recurring reviews (DEC-195)
The operator's rule for record statuses: Implemented, Simulated, Not
Demonstrated, Not Applicable; stand-ins only when a bad input changes the
result and a real input could replace it; stand-ins and scope cuts are
proposed, never self-approved; never choose a status because a gate
requires it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CLAUDE.md: proof-of-concept latitude
@sam-aydlette

Copy link
Copy Markdown
Owner Author

Closed without merging. The work is kept locally and will land on main in smaller slices, each deployed and verified.

@sam-aydlette
sam-aydlette deleted the feat/frr-records branch November 7, 2026 21:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants