1. Brain
  2. Decision
  3. Arena
  4. Commit
  5. Prediction
  6. Outcome
  7. Calibration

Decision history

What you decided, and what reality said.

Record the real numbers when a deadline lands. Every outcome you enter sharpens the challenges in your next decision.

  • May 28, 2026 · committed

    Ship IndieTerminal as a public slash-command dashboard, or keep building a PTY terminal in private?

    choseShip /command publicly

    A URL a founder can type into is a better test than another month of emulator work nobody can run.

    your confidence 74 · arena 61

    • Technical Co-Founder · for

      A PTY you cannot run is not a product. /command is.

      The connected repo is a Next.js app with a command registry, not an emulator. Shipping the dashboard tests the actual codebase. Another quarter on a PTY tests a codebase that does not exist yet.

    • Product Co-Founder · conditional

      Ship /command, but stop calling it a terminal on the landing page.

      People who want Ghostty will bounce. People who lost a ChatGPT thread will stay. The name can wait. The object on the page cannot be a lie.

    • GTM Co-Founder · for

      A public URL is the first distribution event IndieTerminal has.

      Stealth on the personal site produced a status line. A domain produces a number. You cannot calibrate a product that only you have seen.

    • Financial Co-Founder · for

      The cheaper experiment is the dashboard. The emulator is a year.

      OpenAI drafts, Clerk, Supabase — that is this quarter's burn. A PTY is a different hire, a different stack, a different year of runway you have not named.

    • The Contrarian · against

      Shipping /command under the name terminal trains the wrong customer.

      You will spend the next year explaining you are not iTerm. That is not a positioning problem you can patch with a changelog. If you ship, ship under a name that matches the object.

    Review in the Arena

    Predictions

    • indieterminal.com gets 500 unique visitors in the first 14 days.

      expected500 visitorsactual186 visitorsmissed · 2.7× off

      growth · due Jun 11, 2026

    • /command ships in 21 days from the commit.

      expected21 daysactual27 daysclose · 0.8× off

      timeline · due Jun 18, 2026

    • Four slash pipelines live at launch: research, launch, post, operate.

      expected4 commandsactual4 commandshit

      technical · due Jun 18, 2026

  • Jul 9, 2026 · committed

    Let /post send to X and Slack as soon as the draft is ready, or keep the approval queue?

    choseKeep the queue

    If the product is a gate, removing the gate is a different product.

    your confidence 82 · arena 70

    • Technical Co-Founder · for

      The queue is already in the pipeline. Removing it is a new product.

      External actions already fork to the approval table. Auto-send means rewriting that fork and living with the first bad /post that hits X at 2am.

    • Product Co-Founder · for

      The gate is the reason to leave ChatGPT.

      ChatGPT will post. You will not, until you tap. That is the only sentence on the site a competitor cannot copy this week.

    • GTM Co-Founder · against

      A queue nobody uses is friction without a story.

      If /post is run twice a month, the queue is a screenshot. Speed-to-post is how indie hackers actually work. You may be protecting a ritual.

    • Financial Co-Founder · for

      One unapproved post is more expensive than a slow week.

      There is no paid plan yet. Reputation is the account. Auto-send turns a draft model into the public voice of a 14-year-old founder with no undo.

    • The Contrarian · against

      You said agents should act. Then you built a product where they wait.

      The LinkedIn line is 'lets agents act, not just chat'. The README line is 'nothing posts until you hit approve'. Those cannot both be the product. Pick which sentence you are willing to delete.

    Review in the Arena

    Predictions

    • Zero unapproved external posts in the first 30 days of the queue.

      expected0 postsactual0 postshit

      technical · due Aug 8, 2026

    • 12 approved /post sends in the first 30 days.

      expected12 postsactual3 postsmissed · 4.0× off

      distribution · due Aug 8, 2026

  • Aug 4, 2026 · committed

    Open-source kiaan-mittal/indieterminal, or keep the repo private while the site is public?

    choseKeep it private

    The site is the demo. The registry is still changing too fast to freeze in public.

    your confidence 68 · arena 44

    • Technical Co-Founder · for

      A private repo means this floor is arguing from a README.

      The Brain can name registry.ts. It cannot open it. Judges running WebMCP on Decision Arena can see 17 tools. They cannot see IndieTerminal's. That is a choice, not a law.

    • Product Co-Founder · against

      Opening the repo freezes a command surface that is still moving.

      Public GitHub is a promise about names. /research still changing its output shape every week is a bad first impression to pin.

    • GTM Co-Founder · for

      The site is already the leak. The private repo is theatre.

      Anyone can sign up. The code is the part you are hiding from agents, not from copycats. Copycats will use the product, not git clone.

    • Financial Co-Founder · against

      There is no revenue to protect. There is also no reason to donate the pipeline.

      OpenAI prompts inside /launch are the only scrap of secret. Public GitHub gives those away for a README badge.

    • The Contrarian · for

      You cannot claim agents act on a system they are not allowed to read.

      Decision Arena ungated itself so a judge could use it. IndieTerminal still asks for Clerk, then GitHub, then hope. The connected repo being private is the same move with a different password.

    Review in the Arena

    Predictions

    • The private repo still receives a push at least weekly.

      expected7 daysactual2 dayshit

      technical · due Aug 30, 2026

    • 40 people sign up after the private-repo decision.

      expected40 accountsactual11 accountsmissed · 3.6× off

      growth · due Aug 25, 2026

  • Aug 31, 2026 · open

    Should /research and /scan run without a Clerk session, so a browser agent can operate IndieTerminal?

    your confidence 57 · arena 39

    • Technical Co-Founder · for

      An agent cannot call /research through Clerk. Guest /research is the smallest honest WebMCP surface.

      The command is a GET-shaped read: a query in, citations out. Capping tokens per IP is an engineering problem. Requiring a Google account is a product problem you are exporting to every agent that lands on the page.

    • Product Co-Founder · conditional

      Guest /research is the demo. /post and /operate stay signed-in.

      Reads can be anonymous. Writes to someone else's X account cannot. The mistake would be treating every slash command as the same permission. Split the registry: public reads, authed writes.

    • GTM Co-Founder · for

      A command that works in the URL is the waitlist.

      indieterminal.com already exists. Asking people to sign up before they have seen /research is the stealth habit with a nicer domain. One ungated command is a number you can be wrong about.

    • Financial Co-Founder · against

      Unauthenticated OpenAI calls are an open tab on the only real cost.

      There is no price. The bill is tokens. A guest /research with no cap is a way to light money on fire for strangers scraping the endpoint. A signed-in user is at least a person you can turn off.

    • The Contrarian · against

      A guest session has no company. /research without a brain is a search box.

      The product is stored context. A stranger hitting /research has no org_brain_records. You would be demoing the one part of the stack that is not the product, then calling it agent-native.

    Open in the Arena

    Predictions

    • If guest /research ships, 40 unauthenticated runs in 14 days.

      expected40 runswaiting on reality

      growth · due Sep 15, 2026

Past decisions · Decision Arena