Repository navigation
Description
ifrit, the offensive range's host, decided by ADR-0017 ("Buy ifrit for
IOPS and keep the range disposable") and specced there. #96 closed on the
decision and the runbook, and the purchase lost its tracker — the roadmap
still ends its entry with "what is left is the purchase itself and the
build", and ADR-0015 still says "ifrit is unpurchased". Same shape as
#102 → #404 and #95 → #413; this is the tracker.
What is decided (ADR-0017)
- Form: a quiet SFF or mini box, not rack-mounted, one NIC, on a Green
(VLAN 30) access port ofneoand an outlet on the PDU. - Disk: NVMe. Not a spindle, not a mirror — the range is disposable by
decision and gets no backups and no PBS. - RAM: two socketed slots, 32 GB fitted; the second slot is the fix if
four concurrent targets and a BloodHound ingest do not fit — "a memory
kit rather than a machine". - Address: VLAN 30, per ADR-0014; the attack VM is its only dual-homed
guest, the targets sit on a bridge with no physical port.
What it waits on
ADR-0017 and build-the-playground.md §0 put three things ahead of it, and
they are now: #101's estate (the domain, #414, and the SOC stack, #266/#267)
— an attack VM pointed at an uninstrumented estate teaches nothing; #234's
tripwire (armed and verified 2026-09-08); #235 (decided, the iLO stays, and
the BMC hardened 2026-09-09). So the purchase is gated on #414 being built,
and is the last purchase on the estate's list, not the next.
Order of work
- Buy it, to the spec above, when Build the lab Windows domain on Saruman (ADR-0029) #414's domain is up. The two SM863a SSDs
bought under Fit the two SSDs in Saruman, and re-derive what the spindles decided #418 are forSaruman, not this box; Fit the two SSDs in Saruman, and re-derive what the spindles decided #418 says the range
still needs NVMe of its own. build-the-playground.md§1–§9: rack, the two bridges, moveSaruman
to10.0.30.20, close the management plane, the attack VM, the targets,
the four properties, prove a leak reports itself.- The optional Loki rule from ADR-0014 (a
172.30.30.0/24source logged
as a block onigc0.30) lands here, once the subnet is real.
Purchases this needs
The box. Nothing else — and a memory kit only if 32 GB proves short.
Activity
Merge pull request #422 from Gerrrt/gerrrt/everything-still-to-buy
docs(adr): ADR-0039 — decline Proxmox clustering while ifrit is the r…
docs(adr): ADR-0039 — decline Proxmox clustering while ifrit is the r…
Gerrrt commentedon Sep 17, 2026
2026-09-17 — the gate read, and the documents reconciled. No purchase, no build.
The gate, as it stands today
shivastay on VLAN 30So the purchase is still gated on #414, and is still the last on the estate's
list rather than the next. Nothing was bought and nothing was built.
What did land — #490
The half of this issue that does not need the hardware: three documents that
still described the purchase through the closed #96.
docs/roadmap.md— the Infrastructure entry was keyed to Procure ifrit and build the playground — after the main network is finished #96 and ended"What is left is the purchase itself and the build", which reads as though
the money were the next thing to spend. Re-keyed to this issue, with the gate
written out as above. The file had been contradicting itself since this issue
was opened: Everything still to buy at the top already said "the last
purchase on this list, not the next".
docs/runbooks/build-the-playground.md— §0's prerequisite table read asthree open to-dos, two of which had been closed for over a week. It gains a
State column, so a reader arriving on build day does not re-arm the
tripwire or re-harden the BMC, and the blocking moves into a paragraph beneath
the table where it cannot be skimmed past.
docs/runbooks/build-the-lab-guest.md—Saruman.20 is after thisissue, not after Procure ifrit and build the playground — after the main network is finished #96.
Left alone on purpose
ifritis unpurchased" — still true. It stops being true onthe day the box is bought, and that is the right time to touch it.
One more, laterparagraph — already correct, andcheck_buy_listties both to README's count.#96references in ADR-0014, -0017, -0029, -0031 and the Decide whether the lab VLAN needs egress filtering #86 Doneentry — history, and correct as history (ADR-0001).
172.30.30.0/24source logged as a blockon
igc0.30) — still step 3 of the order of work above, once the subnet isreal. It was considered for authoring ahead of the hardware, the way
stacks/sensitive/was authored ahead oftrinity, and deliberately not:§9 already proves the property by hand on build day, and a rule that cannot
fire until a subnet exists is not something this estate keeps.
model, CPU and disk are chosen at the till and named only afterwards, in
hardware.md. Pricing them now would be stale by the time Build the lab Windows domain on Saruman (ADR-0029) #414 clears.make lintandmake check-docsboth pass, withcheck_docs' output identicalto the baseline — no counted fact moved.
Merge pull request #490 from Gerrrt/gerrrt/ifrit-purchase-range-build…
Gerrrt commentedon Sep 26, 2026
2026-09-26: The domain is up (#414, six guests since 2026-09-25), but step 1's "when #414's domain is up" isn't the whole gate.
build-the-playground.md§0 androadmap.mdalso want the SOC, and #266 and #267 are both open. This is still the last purchase, not the next. The roadmap's buy-list line now says so (#670).Gerrrt commentedon Oct 1, 2026
2026-10-01 — the gate is clear, and a shortlist for the box. Nothing bought.
The gate
#266 and #267 both closed 2026-09-27, so the SOC half of the gate in the 2026-09-26 comment above is met. With #414, #234 and #235 already done, nothing blocks step 1.
roadmap.mdand §0 ofbuild-the-playground.mdstill say the SOC is not built. They are not touched here.Shortlist
These are eBay asking prices checked live on 2026-10-01. Sold-price history sits behind a login, so they are not comps. "All-in" means the price to reach ADR-0017's 32 GB.
For reference, the first pass picked an HP 800 G6 Mini with 32 GB fitted at $449. #1 beats it by about $120 with the same CPU and NIC.
Ruled out, and why
What the RAM market changes
DDR4 SO-DIMM prices are up about 3× since early 2025, and used 2×16 kits now start at about $120. Even so, units sold with 32 GB fitted ask more than that premium, so a 16 GB unit plus RAM is the cheaper path. A unit that ships 1×16 with a free slot is cheaper still. That is the case ADR-0017's "second slot" argument already makes, now applied at purchase.
For the build, whichever unit is bought
post-up /sbin/ethtool -K <iface> tso off gso offin/etc/network/interfaces. This belongs inbuild-the-playground.md.Per ADR-0017 and the roadmap, the model, CPU and disk are recorded in
hardware.mdonce the box is bought, along with ADR-0015's "ifritis unpurchased". None of that moves on a shortlist.Generated by Claude Code
docs(roadmap): #266 and #267 are closed, so the SOC's entry leaves (#421
docs: ifrit's purchase gate is met, and the SOC's finished roadmap en…
docs(hardware): ifrit takes U10 in the rack plan — bought, and two pa…
Gerrrt commentedon Oct 6, 2026
2026-10-05:
ifritis bought. It's a Lenovo ThinkCentre M80q Micro, refurbished: eBay item 800466163205, "Lenovo ThinkCentre M80q Micro i7-10700T 2GHz 8GB 256GB SSD". It hasn't been delivered yet.It matches ADR-0017's form: a quiet 1-litre Tiny, socketed RAM and NVMe. Lenovo's PSREF sheet for this generation (https://psref.lenovo.com/syspool/Sys/PDF/ThinkCentre/ThinkCentre_M80q/ThinkCentre_M80q_Spec.pdf) says:
Still to buy before it meets ADR-0017
ifrittakes U10 in the 27U rack plan (#918). A ProDesk Mini tray is not assumed to fit a TinyOnce these are fitted, this issue's build steps can run.
docs(runbooks): move into the 27U rack — bench prep, one window share…
19 remaining items
feat(range): build tpl-kali on gandalf and make the attack VM, ragnar…
feat(range): RangeSubnetLeaked, and §8 ran on gandalf before any targ…
fix(packer): drop tpl-kali's build-time eth0 DHCP stanza before templ…
docs(range): §9 checks RangeSubnetLeaked on the security channel, whe…
fix(packer): drop tpl-kali's build-time eth0 DHCP stanza before templ…
feat(range): RangeSubnetLeaked, and §8 ran on gandalf before any targ…
docs(roadmap): keep §9's date and outcome out of #421's entry (#421)
docs(range): §9 passed on gandalf, a leak reports itself (#421) (#1400)
docs(range): §6 attaches the imported disk and boots from it (#421)
docs(range): §6, three targets on vmbr1 with clean snapshots (#421) (#…
Gerrrt commentedon Nov 12, 2026
Split out: Sliver on
ragnarokis now #1481, a sub-issue of this one in thelastmilestone.ragnarokitself was rebuilt by OpenTofu on 2026-11-12 (#1464) and passedscripts/finish-attack-vm.sh.Generated by Claude Code
docs(roadmap): drop a finished-work date and a met gate
docs: the roadmap and changelog after the tracker pass, and ragnarok'…