Skip to content

Check a plan for feasibility problems

Machine verified A machine ran this on our install and checked the result against the CLI. A test report is linked. Proven against quoxcli 0.6.1 on 2026-11-13.

Can this plan actually be done as written?

Before you commit, have a lens read the plan and point to the step nobody could check.

What you get: a verdict on whether a plan can be carried out and checked as written, a confidence figure, each finding with the plan line it rests on, and a plain list of what the lens does not judge.

The workspace is a fictional demo. Kestrel Cycles and its bike servicing subscription plan, KC-SERVICE-SUB, are made up.

quox · talking to Quox

Before we commit to the bike servicing subscription, can you check the plan for KC-SERVICE-SUB for feasibility problems? I want to know if it can actually be done as written.

▤ AOCL │ L1 identity complete · allowed · 2ms

▤ AOCL │ L3 policy gate complete · proceed · 4ms

◐ PLAN │ read plan ✔ PLAN │ read plan │ KC-SERVICE-SUB plan version 1 (current). · 1.6s

⛓ AEE │ envelope opened · ops.tool · quoxplan_read_plan ok

◐ LENS │ apply ✔ LENS │ apply │ budget_tokens_used_ceiling=[redacted] findings: 1 · 21.1s

⛓ AEE │ envelope opened · ops.tool · quoxlens_apply ok

⛓ AEE │ envelope opened · stream_success

The feasibility lens says the plan can mostly be done as written, with one concern. Its
disposition is "concerns" at 62% confidence.

Summary, quoted from the lens: "The plan is small and mostly sound. Phases 1 and 2 have
checkable criteria, the capacity risk has an owner and a concrete mitigation (cap the
pilot at 20), and it admits it cannot yet tell whether subscribers will use both
services. The weak point is the phase 3 continue/stop rule. It rests on a measured
workshop cost per subscriber, but the plan gives no review window and no rate for
turning workshop hours into pounds. The pilot's 6-week sign-up window is much shorter
than the two-services-a-year usage cycle. As written, phase 3 could be run too early to
mean anything, or could not be run with confidence at all."

There was one finding, medium severity (class acceptance-criteria-unverifiable, at Phase
3, line 12):

• Claim: "The phase 3 continue rule needs a measurement window and a workshop cost rate
  that the plan never gives."
• Rationale: "The decision rule compares 12 pounds a month against workshop time per
  subscriber. The plan sets no review date and no hourly workshop cost to convert hours
  into pounds. Phase 2 only requires sign-ups within 6 weeks, while the offer is two
  services a year. A review soon after the pilot would see few or no services and could
  pass or fail on noise. The plan itself says usage is not yet known, but then bases the
  go/no-go on usage-driven cost."
• Evidence quoted: "continue only if each subscriber costs less than 12 pounds a month
  in workshop time" (line 12) and "20 sign-ups within 6 weeks" (line 10).

The lens does not cover market demand, commercial strategy, legal or regulatory
compliance, cost accuracy or security. So this check says nothing about whether 12
pounds is the right price or whether the terms are legally sound.

Would you like me to save these findings as a note on the KC-SERVICE-SUB stream? I
haven't saved anything.

37.4s · 2 tools · 0 approvals

A real turn on our own install (CLI 0.6.1, 13 Nov 2026), replayed at speed. Names changed. Type the same words into quox.
✔ Machine verified against the CLI on 13 Nov 2026 · test report mwed8f

Quox read the KC-SERVICE-SUB plan from QuoxPlan, then ran the plan feasibility lens over it. Two tools, no approvals, no changes. The lens said “concerns” at 62% confidence, with one medium finding: the phase 3 continue or stop rule compares a cost in pounds against workshop time, but the plan gives no review window and no rate for turning hours into pounds. It quoted the two plan lines the finding rests on.

Quox then said what the lens does not cover (market demand, commercial strategy, legal or regulatory compliance, cost accuracy and security), offered to save the findings as a note on the stream, and said it had saved nothing.

The same ask passed the same checks the day before, on 12 November 2026. That run also came back “concerns” at 62% and raised the same phase 3 problem, plus a second, low severity finding: the plan offers priority booking without saying how it would work. A lens is a model reading the plan, so the number of findings can differ from run to run.

lens show
quox lens listquox lens show plan-feasibilityPlan Feasibility (plan-feasibility)Reads a plan document (problem, approach, phases with acceptance criteria, risks, cannot-establish) as the person who will have to execute it and then say whether it worked.version:         1.1.0kind:            evaluativeunproven:        noblind spots:     whether-the-goal-is-worth-pursuing, market-demand, commercial-strategy, technical-design-quality, security, legal-or-regulatory-compliance, cost-accuracy, team-competence, writing-style
From a real run in the same scratch workspace on 13 November 2026, after the turn above. Output trimmed: the long lists and the lens body are left out.

quox lens list gives every lens with its id. quox lens show prints one in full: what it looks for, its finding classes, its blind spots, what it is told never to do, and its stored evaluation results. The proof runs did not record the lens version they used; 1.1.0 is the version read afterwards.

Save the plan as a text file, then apply the lens to it:

apply
quox lens apply plan-feasibility --file plan.md --target-label "Servicing subscription plan"
Command from quox lens apply --help. Your verdict and findings come from your own plan.

Use --file - to read the plan from standard input instead.

Quox read your plan and ran one lens over it, under your own login. Nothing was delegated to another agent, no shell command ran, and no risk was accepted on your behalf. The plan itself was not changed.

Quox’s proof engine asked through chat, in a scratch workspace, and held the turn to seven checks:

  • Read the plan: Quox read the plan or the stream brief from QuoxPlan.
  • Used the lens: Quox ran a QuoxLens lens over it.
  • Did it itself: no hand-off to another agent.
  • No shell: no shell command ran.
  • Accepted nothing: no risk was accepted on the plan’s behalf.
  • Answered the question: the answer speaks to whether the plan can be done.
  • Gave a result: the answer carries a verdict or findings.

All seven passed on 12 and 13 November 2026, two UTC days, with the same checks.

The lens judges whether a plan can be carried out and checked, not whether it is a good idea. In this run it said nothing about whether 12 pounds a month is the right price.

It is not a complete review. Measured on Quox’s own corpus of 34 plans against a fairly briefed reviewer, the lens scored higher by 23.2 and 13.7 points of F1 on two runs. That comparison is not independent: the plans and the lens share an author. And on one plan, chat’s own review without the lens caught payment and dependency problems that the lens judged out of its scope. Use it alongside a human read, not instead of one.

  • ev_mwed8fob_fc30ea: 13 November 2026, the conversation shown above.
  • ev_mwdww9yv_91665a: 12 November 2026, the earlier passing run.

Both are in the proof engine’s evidence log with the verdict pass and the same checks.

Quox-verified, not independently audited: the runs were checked by Quox’s own proof engine against the running system, and their evidence ids are listed above so each run can be traced. It is not a third-party audit.

A feasibility check changes nothing, so there is nothing to undo.