Opponentura · a panel of critics
Everyone around your decision is paid to agree
We are paid to disagree. An independent AI panel attacks the assumptions, the numbers and the downside of a decision you are about to make — and tells you what to check before you commit. A chat assistant answers your question and stays on your side; Opponentura is your opponent.
The first verdict is free. It comes with the single most dangerous assumption behind your decision and the first check worth doing before the irreversible step. The full report — the reasoning, the action items, the sources — is 15 € / $15.
5–15 minutes to describe it · the verdict the same day · the material can be deleted an hour after delivery. A session is convened before an irreversible step — a deal, a signature, a dismissal, a launch.
Opponentura is a panel of independent AI critics. You have a decision — the panel assembles for your situation, attacks it from every angle, argues with itself, and returns a Decision Review Report with a verdict of APPROVE / REVISE / REJECT, action items and an integrity seal. You have none yet — the panel puts several options on the table, you pick one, and that one gets attacked just as hard; on the higher-tier plans it takes on all of them, so you can see which held up best.
Criticize, then propose — the old rule works both ways here: we propose, and we are the first to attack what we proposed. Not an assistant that agrees — an opponent that checks.
The problem
Everyone around you says yes
Your team
Agrees, because you sign their paychecks.
Consultants
Agree, because they want to renew the contract.
A chat assistant
One voice, and no one to argue with it. Give it two opposite decisions and it will work through both without calling either weak — it answers your question, not someone else’s.
No one has an incentive to push back. But the cost of the error you miss is yours: a blown deal, a lost partner, a lawsuit, money, years. A session with Opponentura costs less than an hour of a consultant — and far less than the mistake it catches.
How it works
How a review runs
You bring a decision you are about to make — or a situation in which one still has to be found. What follows is a procedure.
Intake
Five questions are always asked: what exactly you are deciding; what happens if you do nothing; what an error costs you; which alternatives you rejected and why; and by what date the decision is needed. Beyond those, no more than three questions about your case, and nothing you already said is asked twice. Before a personal or non-obvious question we say in one line why it is needed. Name the company or the market and the panel gathers the public part itself instead of asking you about what is already published; decline and we work anonymously — not a single query about you leaves for the web. And if it is all already written on your website, just say so and give the address: we read the home page and the main sections ourselves and put what we read into the case file, so you do not retype what is published.
Options
A stage for those who arrive with a situation rather than a chosen move. The panel puts two to four genuinely different options on the table and says plainly what is dangerous about each. You pick one — or word your own. The panel then attacks what you picked: the criticism lands on our draft, not on what you agonized over. If you already have a decision, this stage is skipped.
The case
Your question gets a number and a name — “Case 3 — the office purchase.” You can return to a case any time: the panel remembers past sessions.
Attack
The panel assembles for your situation: the other side of the deal, a domain expert, a veteran of a similar failure, a methodologist. Engines from two independent vendors sit on every panel: a model agreeing with itself is not evidence. We do not promise an even split — with three seats there is no such thing; but the second provider sits on every panel: one of the critics and the independent fact review before printing. From the Full panel upwards it also fields a decision-bias reviewer, who looks not at the decision but at how it was arrived at — anchoring, sunk cost and confirmation bias in the record, not personality traits; the Panel plan and the free first review do not include it. And the options stage too, when drafts were tabled.
Debate
The critics argue with each other — a real structural dispute, built-in defense against flattery. Dissenters file a separate opinion. Right after the transcript comes the map of disagreement: where the panel agrees, where it splits, which fact would settle the dispute, and who changed position under an argument. Without that summary a debate reads as decoration; with it, as evidence of work.
Verdict
A roll-call vote: FOR / CONDITIONAL / AGAINST — and the panel’s decision: APPROVE, REVISE or REJECT, with action items and dates.
The report
The report reaches you in the bot within 24 hours of your answering the intake questions — usually much sooner. Before the file itself a decision card arrives in the chat — the verdict, what the decision rests on, what to check first and what to do, on a single screen. On the first page of the document, right after the verdict, comes “The essentials”: the three most dangerous assumptions your decision rests on, the cheapest check to run before the irreversible step, and the condition under which the decision stops being right. Every review, every plan. Then a document for your partner, your board, your lawyer — with a seal and an integrity hash. The original is the PDF file — only it counts as the original. Any recipient can attach it on our verification page and see whether it is the same document: change a single character and the check fails. The document has no other form: a printout or copied text is a reading copy. If the review ran short of its promise — a critic dropped out, the fact review did not complete — we say so on the first page of the report and sort out your quota.
After the report
The work does not end when the document arrives. The action items from the “Resolved” section become an action plan in the bot: /orders lists them, /done_N marks one finished, /failed_N marks one that did not work out. Dates by which the decision can still be changed — signing, a board meeting, the end of a pilot — you add yourself with /calendar, and we warn you in advance. And we ask how it ended: three short questions — what happened to the decision, which objection was new to you, what the panel missed. Those answers are what the method is corrected by; we count a review as successful not by the number of findings but by what happened to the decision. The action plan and the outcome question are free on every plan. Reminders that arrive on their own — at the checkpoints and on order deadlines — are decision supervision, included from the Full panel upwards.
What sits behind the words “debate” and “verdict”
The hearing runs in a courtroom genre, but the genre is the shell. Inside every report there are five ordinary things, and they matter more than the words:
The panel attacks the decision, not the person. What gets taken apart is the move, the assumptions, the numbers and the deadlines — not your competence; there are no judgments of character, no condescension and no reproach for what is already done. There is no defendant at this hearing. This is not politeness for its own sake: a person who has been stung stops reading on page three, and the most expensive finding goes unread.
- The risk is named — not “complications are possible”, but what breaks and under which condition.
- What backs it — a source, or a plain “not confirmed”: before printing, the report is checked by a model from the other vendor.
- What happens if the risk lands — in money, time and relationships.
- What to check first — instructions with dates, cheapest test first.
- The condition under which the decision holds — and when it is time to change it.
Who and what for
When it is worth convening a panel
Not for everyday calls. For the step that is hard to undo — a signature, a wire, a dismissal, a launch — and usually on the evening when the decision is all but made and still keeps you awake. Below are three of the situations people bring most often.
“I am buying out my co-founder”
The other side of the table shows where you will be held up. A separate question: what actually transfers with the shares — the equity, the client relationships, or somebody else’s liabilities.
“I have to let go of the executive our clients follow”
The HR lawyer sets out the sequence and the notice arithmetic. The domain expert estimates which accounts move and when. Your executive explains what he does next.
“The bank will lend to the company — against a personal guarantee”
The CFO works out at what fall in revenue the first breach arrives. Counsel reads the enforcement terms rather than the credit deck. A veteran of insolvency describes the order in which things are taken.
Six more situations — deals and money, people, growth and commitments — on the Cases page. Personal decisions (a guarantee for a relative, a divorce settlement, an inheritance, moving the family) are handled separately: personal decisions.
Comparison
How Opponentura differs from ChatGPT
| A chatbot’s answer | An Opponentura report |
|---|---|
| No dissent | Dissenting opinions as separate documents |
| No sources | External facts are checked with live search by a model from the other provider; what could not be confirmed is marked “not confirmed” rather than passed off as fact |
| No measure of confidence | Probabilities and votes, name by name |
| Nothing to verify | The original is a PDF: any recipient can check the file against our register |
| Agrees with your framing | Intake plus an attack on the frame itself |
| Answers straight away | Critics work apart and never see each other’s conclusions — the argument starts afterwards, in the debate |
| The conversation ends and that is that | The orders stay on as a plan of action, and we ask how it went. On plans with supervision the reminders come on their own — for up to 90, 180 or 365 days |
| Gives you one option and defends it | Tables several options and hits each of them itself |
| Forgets the context | A ledger of cases: the panel remembers your sessions and deadlines |
| Unknown which model is inside | Current models from Anthropic and OpenAI, line-ups checked daily |
The thinking is done by the current flagship line-ups of Anthropic and OpenAI. They are also the most expensive ones: a review costs what their work costs. The top models of both vendors are included on the “Premium” and “Special matter” plans; the other plans run the working models of the same line-ups. Two independent engines argue with each other: agreement between models of different schools carries more weight than one model’s opinion. And the more cases you bring, the sharper the panel becomes: it remembers past sessions, your deadlines and outcomes. We watch for new releases daily, but an engine is switched only after a quality check and only by a human decision: newer is not always smarter for a given role.
The method
What the “O-18” method is
One critic, however good, looks one way. We broke the examination of a decision into 18 procedures — “lenses” — and handed them out across the panel so that no one duplicates their neighbour. Four lenses are mandatory in every case; the rest are assigned to fit your situation. A lens counts as applied only if it produced a result or a reasoned refusal: “looked and found nothing” is a report too.
Four lenses are mandatory every time: the strongest version of your own decision, a pre-mortem, an audit of key assumptions and a verification chain. The rest are dealt out to fit the case. The method is not invented — Gary Klein’s pre-mortem, Richards Heuer’s competing hypotheses, the ICD 203 probability language, Toulmin’s argument model, Popper’s falsifiability, the ISO/IEC 31010 risk map. Every source with an author, a year and a link, the lenses we disclose, the decision-bias reviewer’s role (from the Full panel upwards) and the three checks before printing — on the “O-18” method page.
Samples
Sample reports will be published at launch
They will be neither a demo nor a staged mock-up: real decisions from the founder’s own practice, run through the panel. Names and figures will be redacted, everything else will stand as it was. For now there is nothing to show — and we would rather say so plainly than fill the space with placeholder cards.
What you can already verify: every report is sealed with a SHA-256 hash, and whoever receives the document checks the file against our register — it cannot be altered unnoticed.
The founder
Why I built Opponentura
My name is Vadim Kamashev. I am originally from Russia and I live in Spain. I own and run a business, and I make the kind of decisions I pay for myself — with my own money, not an investor’s. I built Opponentura for myself first and opened it to others only afterwards.
The reason is simple: everyone around a decision-maker has an interest in agreeing. Your employee — because you are their boss. Your consultant — because they want the next contract. A chat assistant — because it is one voice answering your question: give it two opposite decisions and it will back both. Nobody gains by objecting to you, and the cost of the flaw nobody named is entirely yours.
The first thing the panel took apart was my own launch plan for this very service. The verdict: revise — unanimously, not a single “approve”, and a list of things to fix before going live. I am still working through some of them. I did not enjoy it — which is exactly why I trust the thing.
Why I got into this, what happens to your data and what the Service does not do on any plan — on the about page.
Vadim Kamashev, founder of Opponentura
Privacy
What happens to your data
- No model is trained on your materials. Neither ours nor the vendors’ — that is a contractual term of the Anthropic and OpenAI APIs, not a promise on our word.
- You choose the retention period. In the bot,
/privacy: the standard term, “within an hour” after the report is delivered, or no-trace mode — every review leaves nothing behind. It switches at any time and takes effect at once. What always remains is the report number, the verdict and the file hash: the authenticity check of your own copy rests on them. - Telegram, on purpose. The review runs in Telegram, and that is a choice rather than an unfinished feature. Given what people bring us — a buyout, a dismissal, a dispute — we want as few surfaces as possible where the material could sit: no web account between you and us, no forms, no passwords, no sessions, no cookies, no third-party widgets, and every one of those is a place data escapes from. Telegram has a long record on security, it is run independently of any single state, and it is the one channel most of our customers already have. We ask for no phone number, no surname and no email beyond the one a card payment needs for the invoice: what we hold is your Telegram id and the name you introduced yourself by. Browser access and card payment without Telegram are in preparation — for those who cannot use it or would rather not. We will not name a date until it works; when it arrives the chain will have one more link in it, and we will say so plainly rather than quietly. What we do not claim: a conversation with a bot is an ordinary Telegram chat, not a “secret” one, and the model providers read the substance of the case. That is true on any channel; the only difference is how many links our own side adds.
- Not a single tracker. No Google Analytics, no Yandex Metrica, no social pixels — and no third-party scripts at all: your visit does not go to anyone else. It is enforced technically — the pages forbid the browser to load anything from other domains — so a counter would not work here even if one ended up in the markup by mistake. We host the fonts ourselves too.
What we honestly do NOT promise: “your data never leaves our server”. Requests are processed on the model providers’ servers under their policies, as with any service built on frontier models. Redaction before sending, how long attachments and report live, and deletion on request — in the privacy policy.
Scope
Which topics we take, and which we do not
Hard negotiations, dismissals, lawsuits, partner conflicts, bankruptcies — yes. Without moralizing: the panel critiques your decision, not you.
A short banned list is public: facilitating violence, human trafficking, terrorism, weapons, drugs, harm to children, stalking of private individuals. Everything else that is lawful — we take on.
Early access
Open the bot and start. The first verdict is free
Access opens immediately: write to the bot and the review starts from your own account of the matter. The service is young and we collect feedback on every case. In early access the number of free reviews per day is limited: if today’s are taken, your case joins the queue and starts by itself in the morning — you do not need to ask again.
We do not ask who you are. The Service does not need your real name, your company, your job title, an ID document, an address or any business details — call yourself whatever suits you and skip the company altogether. We explicitly ask you not to send such things: the less you have told us about yourself, the less there is to lose. The one thing we genuinely need is your Telegram contact: that is where the review runs and where the report arrives. That is personal data and we say so plainly — how we handle it.
How it looks
- You open the bot — access is granted at once, with no approval to wait for.
- You describe the decision in your own words. We meet you and ask what is missing.
- The first review is on us: 3 critics and the full procedure — the decision-bias reviewer starts with the Full panel plan — and you get the verdict APPROVE / REVISE / REJECT, the most dangerous assumption behind your decision and the first check to run before the irreversible step. The written report is already prepared and sealed — unlocking it costs 15 €, and you pay only if the verdict struck you as worth the reasoning.
Open the bot and simply write.
Questions
Frequently asked
Do you only criticize, or do you propose anything?
Both. If you have a decision, the panel attacks it. If you have none, the panel puts two to four genuinely different options on the table, you pick one, and that one gets the same treatment just as hard. The difference from a consultant is that a consultant gives you an option and defends it, while we give you an option and hit it ourselves. Every set of report contains a RESOLVED section — action items with dates, not a list of complaints.
Is this legal advice?
No. Opponentura is decision-support software — not legal, financial or professional advice. For high-stakes matters, take the report to a qualified specialist; the document is built for exactly that.
Who makes the decision?
You do. The panel attacks and advises — the final call always belongs to the owner of the question.
What if the panel is wrong?
We keep a ledger of verdicts and openly reconcile them against real outcomes. A verdict is not a prophecy — it is a map of risks with check-by dates.
Isn't it in your interest to find objections so the guarantee never pays?
It is. We say so plainly rather than hide it. The guarantee is not a lottery about getting your money back — it is our commitment not to hand you an empty document: if the panel found no substantive objection, you do not pay for a page that says everything is fine. To keep severity from being inflated, findings pass two independent checks before printing — a fact review by another vendor's engine and an editorial pass — and the number of substantive objections is printed inside the report and stored in our ledger: one number in the document and in the books, with nowhere to diverge.
Is my decision even worth this?
It may not be — and we will say so first. The bot has a /worth command: five questions about the cost of being wrong, reversibility, timing, how many parties are involved and how much is unknown. If the decision can be undone and little is at stake, the answer is blunt: do not convene the panel — write down the three assumptions the decision rests on and find a way to test the most expensive one in a day. Your answers are scored by code, not by a model: a model asked whether our own service is needed will always find that it is.
What should I bring?
Nothing is required. There is a prompt — /checklist — listing what is worth attaching for your scenario: buying a stake, replacing an executive, a launch, a loan, property, a key hire, a dispute, a family decision. Documents make the analysis concrete: a critic argues with clause 7.2 rather than with your summary of it. But if you do not have them, the review still goes ahead — the panel works from your account of it and states plainly in the report what was taken on trust. And if something essential is missing, we say so when the case is confirmed — before the review starts, not after.
Why not just use free ChatGPT?
It is one model and there is no one to contradict it. Here — a panel, a dispute between critics, and a sealed document.
What does it cost?
The first verdict is free: the panel names its decision — APPROVE, REVISE or REJECT. The written report that explains it unlocks for 15 € and wait 7 days for your decision; after that, plans from 30 € per review — less than one bad decision.
And if the analysis turns out empty?
Then you do not pay for it. The rule is counted by software, not by eye: a material objection is a finding marked “critical” or “high”, or a directive from the “Resolved” section, and their number is printed inside the report next to the seal. See a zero — write to the bot. The guarantee covers your first payment (the first review carries no charge anyway): the claim is accepted within 7 calendar days of issue and the money is returned within 10 business days. Details — section 9 of the agreement.
My question is highly sensitive.
Redaction with labels before sending, plus a trace-free mode. Caveat: it is not available on the free first review — those reports are held with us until unlocked; if you need it from the very first case, start on a paid plan. Frame the case without names — the quality of the attack does not suffer.
When do you need a human expert rather than us?
When the decision turns on a fact only an insider knows: how a particular regulator actually behaves, the state of a site, whether your contractor can really do the work. And always when the law requires a signature — an opinion from a lawyer, a doctor, a valuer, an investment adviser. The record does not replace them and does not pretend to. It shows what you failed to ask and what is worth checking before you pay an expert by the hour.
Can everything be deleted afterwards?
Yes. In the bot, /privacy sets how long we keep things: the ordinary term, “an hour after delivery”, or no trace at all — every review erased at once. A change applies backwards: switch to the short term and the older material goes too. What stays under any mode: the record number, the verdict and the hashes of the files delivered. Without them the authenticity check stops working for the copy already in your counterparty’s hands. Support correspondence is separate — it is a conversation about money, and its retention period is named in the privacy policy.
Can I do this without Telegram?
Not yet — and that is a deliberate choice rather than an unfinished feature. The review runs in the bot because it puts fewer links between you and us: no web account, therefore no forms, no passwords, no sessions, no cookies, no third-party widgets, and no separate database of your credentials. We ask for no phone number, no surname and no email beyond the one the payment platform needs for an invoice. Payment and reviews through the website are in preparation: Telegram suits neither everyone nor everywhere. When they arrive the chain gets longer — and we will say so plainly rather than quietly. What we do not claim in the meantime: a conversation with a bot is an ordinary Telegram chat, not a “secret” one, and the model engines read the substance of the case. That holds on any channel.
I disagree with the verdict. What now?
You may do nothing: the decision is yours and the record does not bind you — that is stated in the agreement and on every page. If you want to argue on the merits, start with the dissenting opinions: they show which members of the panel agreed with you and on what grounds. You can also convene a second review with the same panel at a discount and bring what was missing the first time. The refund guarantee is a different matter: it answers emptiness, not disagreement — it applies when the record contained no substantive objections at all.