Skip to content

docs(roadmap): #294's entry names the proof visit it is actually gated on - #609

Merged
Gerrrt merged 1 commit into
mainfrom
gerrrt/second-age-recipient-sops-7a4ac7
Sep 22, 2026
Merged

Gerrrt merged 1 commit into
mainfrom
gerrrt/second-age-recipient-sops-7a4ac7

Conversation

@Gerrrt

@Gerrrt Gerrrt commented Sep 22, 2026

Copy link
Copy Markdown
Owner

What changed

docs/roadmap.md's entry for #294
now states the gate the issue is actually waiting on — the first proof of the
second recipient's copy — instead of describing the lab as it was before
2026-09-09. Docs only; one bullet.

Why

The entry landed a few hours ago in 3b691fe, written from the issue's
original body rather than its state, so three of its claims were false when
they were written:

The entry said Actually
"what it needs is somewhere to put the second key" Decided 2026-09-08 (#397) — the technical second named on the break-glass card
"Decryption depends on one key held by one person until it lands" #412 re-keyed on 2026-09-09; either recipient opens the secrets
"ADR-0048 already assumes that medium exists" ADR-0048 reports it: "the medium exists, is off-estate by construction, and has not yet made its first visit"

It also never named the gate, which is the one thing roadmap.md says an
entry is for: "A link, the gate, its place in the order, and why. Nothing
else."
The gate is the proof visit — verify-key-backup.sh refuses the live
key by device and inode, so no timer can clear it and the medium has to be
carried here. That is the same visit #573 rides on, and the entry now carries
a → [runbook] pointer as #531, #573 and #574 already do.

#294 is not closed by this. Its remaining box is the visit, so the trailer
is Refs rather than a close keyword.

Blast radius

  • No change to network segmentation or firewall rules
  • No new port published to a VLAN that could not already reach the service
  • No credential added outside secrets/*.sops.yaml

Nothing outside docs/roadmap.md is touched. .sops.yaml, secrets/,
back-up-the-age-key.md, successor-handover.md and secrets/README.md were
all read and are correct as they stand.

Verification

Measured on the host before writing, so the entry's claims are read off the
lab rather than off the issue:

$ ./scripts/key-recipients.sh --list
age1cutv5qvkwxxy9n5clznnvm2hgseeuw9ecswgnspkmxs6d69hrpqqx2wxp4
age1yrdu996u5mhdh0qf93l7s8zz8stneqnqxpncrcarrmgxvsy264rqmkcs6x

Two recipients, matching .sops.yaml's catch-all rule — no drift, which is
what makes "carry two, and either opens the secrets" true. Also read:
key-recipients.prom has age1cutv5… at 0 (never proved),
SecretsKeyBackupUnproven is firing and naming it, and
homelab-recipient-state.timer is active and last ran 09:20 today — so the
gate really is the visit and nothing has drifted since the 2026-09-20 reading
on the issue.

  • make validate passes — all checks passed (3 skipped — NOT run, not verified); the skipped three need a docker daemon and CI runs the full
    set. check_docs.py treats roadmap.md as a record rather than prose,
    so only its buy-table count is checked, and this change does not touch
    that table.
  • Deployed to the lab and confirmed working — n/a, documentation only
  • Docs updated (docs/roadmap.md)

No changelog entry: roadmap.md sends history to changelog.md on the day an
issue is done, and this one is not.

Refs #294, #573.

🤖 Generated with Claude Code

…d on

The entry landed hours ago in 3b691fe, written from the issue's original body
rather than its state, and it described the lab as it was before 2026-09-09.
Three of its claims were false when it was written:

- "what it needs is somewhere to put the second key" — the home was decided
  2026-09-08 (#397): the technical second named on the break-glass card.
- "Decryption depends on one key held by one person until it lands" — it does
  not. #412 re-keyed on 2026-09-09 and either recipient opens the secrets.
- "ADR-0048 already assumes that medium exists" — ADR-0048 does not assume it,
  it reports it: "the medium exists, is off-estate by construction, and has
  not yet made its first visit."

It also never said what gates the issue, which is the one thing roadmap.md
says an entry is for — "a link, the gate, its place in the order, and why".
The gate is the first proof of the new copy: verify-key-backup.sh refuses the
live key by device and inode, so no timer can clear it and the medium has to
be carried here. That is the same visit #573 rides on, which is the ordering
fact this section exists to carry, and the entry now points at the runbook for
it as #531, #573 and #574 already do.

Measured before writing, rather than read off the issue:

  ./scripts/key-recipients.sh --list   two recipients, matching .sops.yaml's
                                       catch-all rule — no drift
  key-recipients.prom                  age1cutv5... = 0, never proved
  SecretsKeyBackupUnproven             firing, naming that recipient
  homelab-recipient-state.timer        active, last run 09:20 today

"Honest reading" is attributed to back-up-the-age-key.md and not to ADR-0024
on purpose: the ADR's "honest state" is about the period before a second key
existed, which ended on 2026-09-09, and reusing it here would have been a
second false claim in a commit fixing three.

No changelog entry. roadmap.md sends history to changelog.md on the day an
issue is done, and #294 is not: the remaining box is the visit, not this
merge. The trailer below is deliberately Refs and not a close keyword.

Refs #294, #573.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@Gerrrt
Gerrrt merged commit 27f56e7 into main Sep 22, 2026
3 checks passed
@Gerrrt
Gerrrt deleted the gerrrt/second-age-recipient-sops-7a4ac7 branch September 22, 2026 16:47
Gerrrt added a commit that referenced this pull request Sep 22, 2026
… its text

Both branches corrected the same false clause — "decryption depends on one
key held by one person until it lands" — after the re-key had landed. #609
came from the session that owns #294 and says more: ADR-0024, the holder,
the device-and-inode refusal that means no timer can clear the proof, and
the runbook pointer. It already carries this branch's correction, so the
resolution is its text unchanged and this branch stops touching roadmap.md.

The other three corrections here are untouched by the merge.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
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.

1 participant