Skip to content
PHRONEN
Create account
Menu

Product

ProductPricingCompanyEnterpriseDownload

Proof

EvidenceSecurity

Account

Sign inCreate accountSalesContact

Evidence

Done means checked

Real runs from the published evidence bundle — predicates, a second review, a failure that was not rounded up, and helpers that ran in parallel. These are published runs, not invented demos.

  • VerifiedAll checks passed on a real run.
  • Not verifiedOne failure is enough — shown, not rounded up.
  • PreviewSecond-agent review is labelled when partial.

Action Audit in Carry Continuum is the same honesty: apply stays reviewable. Continuum is Partial where labelled.

  1. AdviseFAR shows the plan. Nothing runs until you choose.
  2. AskConfirm gates before act. Irreversible clicks still wait.
  3. ActAgent carries the work. FAR stays a separate brain.

Checks

Every result comes with its checks

Before PHRONEN says a task is done, it checks the result itself — the files are there, the page answers, the command succeeds, the numbers match — and shows you every check.

Verified · 4 of 4 checks passed

  • ✓File existspackage.json exists (531 bytes)
  • ✓File containsApp.tsx contains 'Lisbon'
  • ✓Command succeedsexit 0 · built in 62 ms
  • ✓Site answersHTTP 200 in 1.7 ms, contains 'Lisbon'
From a real run: A comparison site for a two-week remote-work stay in Lisbon, Mexico City or Taipei, for someone with an Indian passport: costs, internet, visa rules, neighbourhoods and the hours that overlap with India.

Review

A second agent looks before you do Preview

When a task produced something, a second agent that did none of the work reviews it against your request and sends back what's wrong, twice if needed. It can read pages but can't click through them yet, so flows are checked by running tests.

  1. First review3 things to fix

    • The cost comparison and recommendation are not valid because the displayed benchmarks use different currencies and inconsistent definitions.
    • The visa content does not adequately answer the Indian-passport-specific requirement for Taipei or Mexico City.
    • The claimed browser verification could not be reproduced in this read-only review.

  2. Second review3 things to fix

    • The Lisbon India-overlap figure is incorrect … Lisbon 09:00–17:00 overlaps India 13:30–17:00 for 3.5 hours, not 4.5 hours.
    • The recommendation that Taipei has the lowest monthly burn is not supported by comparable figures.
    • The visa sections do not actually state the applicable entry rule for an Indian passport holder in two important cases.

  3. What PHRONEN fixed

    • Every city's cost converted to one currency at a stated rate (₹84 to the US dollar, on 24 September 2026) and labelled as a planning estimate
    • The entry rule for an Indian passport in each country, with links to the official government pages
    • Every page opened on a desktop screen and at phone width, with no console errors

  4. What it said was still true

    The cost figures are now normalized and explicitly caveated, but they are still planning estimates from different city-cost datasets … the recommendation should be treated as directional rather than a guaranteed rental quote.

Not verified · 1 of 10 checks failed

  • ✓File existsindex.html exists (2695 bytes)
  • ✓File containsindex.html contains 'Recommendation'
  • ✗Command succeedsexit 1: kill: (10241) - No such process
  • and 7 more that passed
A later run of the same request, on its last step.

When a check fails

It says so, instead of saying done

One failed check is enough for a result to be marked not verified. The failure is shown with what was observed, next to everything that passed — never rounded up.

Only an independent check can say it's done — never the part that did the work, never a model's word.

Helpers

Big jobs are split into parts that run at once

Splits big jobs across helpers that work at the same time.

  • Ran python3 -m pytest: 142 tests passed in 0.33s
  • Ran CLI manually: 5 km mi -> 3.10685596119; 2 lb kg -> 0.90718474; 100 C F -> 212
  • Verified CLI error handling: non-numeric input, unknown units, and missing arguments all return exit status 2 with clear stderr messages
  • Ran ruff check src tests: All checks passed!
  • Built wheel with python3 -m build: wheel built successfully

Worked in parallel5 parts

  1. Create package scaffolding and conventionsDone

    1m 03s

  2. Implement length conversion moduleDone

    after foundation

    1m 51s

  3. Implement mass and temperature conversionsDone

    after foundation

    1m 12s

  4. Build command-line conversion toolDone

    after foundation, length, mass-temperature

    1m 17s

  5. Write README and run full verificationDone

    after length, mass-temperature, cli

    1m 41s
A real run: A Python package that converts length, mass and temperature, with a command line, tests and README examples. About 10 minutes; all five parts done on the first attempt.

Give it something, and read the checks.

Start on FreeHow you stay in controlDocs