HAN VERIFIABLE COMPETENCE
why how it works live example events proof chain contact
TR · EN Closed beta
FOR ANYONE WHO MAKES HIRING DECISIONS
Competence shows in the work.
Yetkinlik, işin içinde görünür.
HAN measures what the person you're about to hire can actually do, while they do it. It hands them a real task from your own work, records the entire session as they work with AI, and scores their competence against evidence. The moment the report is finalized, its seal is written to the chain. The decision stays yours.
Closed beta Review the real report →
◉ Live on Avalanche C-Chain · every report is sealed and written to the chain · 0x17DA…5D13 ↗
81/100
✓ HIRE · OUTSTANDING
real task
recorded session
sealed report
HAN
SEALED
REAL WORK, RECORDED SESSION
Every step is recorded as they work
engineers in the terminal, other roles in the browser · the AI derives the criteria from the task
EVIDENCE-BACKED SCORES
Every score comes with a quote
not "made a good impression" · line-by-line justification from the session itself
VERIFIABLE RECORD
Anyone can verify the seal
once the report is final, its seal is written to Avalanche C-Chain · no sign-in required
/ WHY WE EXIST · NEDEN VARIZ
More often than not, the best candidate is cut by a person, not a system.
Someone protecting their own seat won't put a better candidate on the shortlist. They don't even need an excuse; they know you'll never see that CV. As two founders, we lived through this again and again in our own careers. HAN came out of that.
"Didn't even look · they wouldn't have been a fit anyway."
There isn't a boss who hasn't heard this sentence. The problem isn't the sentence · it's that nothing stands behind it.
01
You never see who was cut
The list that reaches your desk has already been filtered. Nowhere does it say why the eliminated names were cut; ask, and there's no written answer.
02
The evidence is in the gatekeeper's hands
If an assessment lives in someone's memory and someone's folder, it changes when they want and disappears when they want. A record like that can't be challenged.
03
The wrong thing gets measured
A CV tells a history; an interview measures the ability to tell it. Neither shows whether the person can actually do the job · which is precisely what you're paying for.
/ HOW IT WORKS · NASIL ÇALIŞIR
You write the task, the system builds the rest.
Describe a real situation from your own work, in your own words. The AI derives the assessment criteria from that text; you don't need to build a separate scale for each role.
01
You describe the task
"Our payment provider went down twice in a row · how should the team respond?" Free text is enough; no template.
02
Criteria are generated
The AI writes 5–8 task-specific dimensions, their weights, the signals to look for and the red flags. About three minutes.
03
The candidate works
Engineers connect from their own terminal and actually write code. Other roles talk with a work assistant in the browser. Every line is on the record.
04
The report is sealed
The evaluator AI reads the entire session and ties every score to evidence. The moment the report is finalized, its seal is written to the chain.
/ WHAT WE MEASURE · NE ÖLÇÜYORUZ
Is the AI driving the candidate, or the candidate driving the AI?
Both candidates use the same tool, and both deliver tidy work. But one said "write me a plan" and accepted what came back; the other gave context, pushed back on an assumption, rejected a flawed suggestion, and justified their decision. That difference doesn't show in the delivered file. It shows line by line in the record.
0–20 very weak21–40 weak41–60 fair61–80 good81–100 outstanding
Directing the AI · giving context, setting constraints, asking the right question
×0.15
Verifying the AI · testing the answer, demanding evidence, rejecting what's wrong
×0.15
Completion · is the work actually done, is it consistent
×0.20
Domain knowledge · which layer they speak from, what they actually know
×0.15
Process · sequencing, what they did when stuck, use of time
×0.10
Reasoning · quality of decisions, prioritization, justification
×0.10
Experience · spotting risk unprompted, edge cases, instinct for scale
×0.10
/ LIVE EXAMPLE · CANLI ÖRNEK
Let's show it instead of telling it.
The report below came out of a real session and was written to the chain. You can open the full transcript, the quote under every score, and the Snowtrace record of the seal without signing in.
81 /100
URL shortening service · software developer
✓ HIRE OUTSTANDING HIGH CONFIDENCE 7 DIMENSIONS ⌨ TERMINAL · 1h 05m
HAN
SEALED
Ran the session from start to finish with their own architecture plan: justified 302 over 301, caught the XSS hole in the interface before the AI did, and rejected a speculative root-cause explanation and demanded evidence. Their weakest point was the time they spent writing tests.
DIMENSIONS AND SCORES
Directing the AI
88
Domain knowledge
87
Verifying the AI
85
Experience
82
Competence
77
Process · Completion
76
THE EVIDENCE BEHIND THE SCORES
"Use 302, not 301 · 301 gets permanently cached in the browser and the click counter goes off."
evt_18 · Candidate
"Your root-cause explanation is speculative. Check node_modules and show me the source."
evt_84 · Candidate → AI
Test Files 1 passed · Tests 4 passed (4)
evt_100 · Run output
This report was sealed at block 93,956,595. Open the full report →Verify the seal ↗Snowtrace ↗
/ FOR EVERY ROLE · HER ROL İÇİN
Two ways of working, one assessment.
Candidates work in the natural environment of their role. What's measured is the same in both: how they direct the AI and their professional judgment.
Engineers · Terminal cli-han
The candidate connects from their own terminal; the task and AI assistant open automatically. They actually write code, run it, and test it. The "Completion" score looks at run output, not claims. There's no browser option for engineers · the measurement belongs where the work happens. The live session infrastructure runs on the cli-han sub-product.
aday-7f3k — ssh
$ ssh aday-7f3k@ssh.han.tayf.io
HAN · verifiable competence assessment
✓ Welcome. Task: URL shortening service…
❯ set up the reverse proxy with Caddy, TLS automatic
▮ recording in progress · evt_031
Other roles · Browser han.tayf.io
Marketer, manager, salesperson, analyst… The candidate enters from the browser, reads the task, and solves it by talking with a work assistant. No setup, no terminal · only the way of thinking is visible. Same dimensions, same scale, same evidence requirement.
CANDIDATE
Budget 1.8M, target 6,000 customers, CAC 520. Is that realistic? Just do the math.
ASSISTANT
6,000×520 = 3.12M needed; budget is 1.8M. At the current CAC that's ~3,460 customers (58%). The target isn't realistic.
CANDIDATE
Then suggest 3 channels that would lower CAC, and show me your source too.
ROLES software developer data analyst product manager marketing sales team lead finance HR operations designer legal teacher the list isn't closed · because the criteria are derived from your task, you can set up an assessment for a role that isn't on the list the same day
/ EVENTS AND PROGRAMS · ETKİNLİK
Challenge-proof ranking for hackathon, award, and scholarship selection.
The hard part of a hackathon isn't judging the code, it's defending the ranking. HAN evaluates every participant with the same task and the same scale; when the result is contested, your answer is ready.
HAN suggests a ranking; it doesn't hand out the award. The jury always picks the winner · HAN only feeds the decision with evidence.
hackathon & coding contest accelerator selection scholarship & award committee internal promotion internship selection
▸ one task, hundreds of participants
You write the task once; everyone does the same work in their own isolated environment. Recording and scoring are automatic.
▸ comparable ranking
Same dimensions, same weights. Scores are meaningful relative to each other; the jury starts its discussion from the top ten.
▸ challenge-proof
The answer to "Why not me?" is ready: under every score, a quote from the participant's own session, its seal on the chain.
▸ no setup burden
The participant enters from the browser or connects with a single line of SSH. No environment setup, no version conflicts.
/ PROOF CHAIN · KANIT ZİNCİRİ
Evidence can't live where whoever wants to bury it can reach.
The moment a report is finalized, its cryptographic seal is computed and written to Avalanche C-Chain. That the report existed at that minute and hasn't been changed since becomes verifiable without trusting HAN at all.
Blockchain is our job, not yours
No wallet, no token, no transaction fee. We run the chain side end to end. What you see is ordinary software: you create a candidate, read a report, make a decision.
WHAT GOES ON THE CHAIN
✓ Cryptographic digests of the transcript and report
✓ Seal time (block number)
✓ Assessment version
WHAT NEVER GOES ON THE CHAIN
✗ Name or any personal data
✗ Scores and report content
✗ Session transcript
When a candidate exercises their right to be deleted, their data is genuinely removed from the system; the number left on the chain can't be traced back to any person. That's why the right to be deleted and immutability don't conflict. The candidate holds the report too · as a portable record of competence.
SEAL 0x454e10339236476e174ef2207b7ab2ca426e043f854ee2cfd4f8e59fd25b2f8c verifysnowtrace ↗
/ FREQUENTLY ASKED · FAQ
You'll probably ask these.
I already have an HR team · what does HAN change?+
HAN doesn't replace your HR team; it puts evidence on their screen. CVs and interviews produce impressions; HAN produces a recorded session and line-by-line justification. The shortlist is still HR's, the decision is still yours — but instead of "made a good impression" you now have a report you can point to.
If the candidate uses AI, aren't you measuring the AI?+
Quite the opposite: today the work itself is done with AI. We don't score what the AI wrote — we score how the candidate directed it: asking the right question, testing the output, rejecting a speculative answer and demanding evidence. The "Directing AI" and "Verifying AI" dimensions in the sample report score exactly that.
What if the candidate has someone else do the task?+
The session is recorded end to end, and the score is tied to the process, not a single output. Ready-made solutions pasted after long silences, a flow where no thinking is ever visible — these show up in the report. In doubtful cases the report gives you line-by-line material to probe the same topics face to face in a second round.
Why blockchain? Wouldn't a database do?+
We could change a database; nobody can change the chain. The seal is publicly verifiable proof that the report was never altered afterwards — for the candidate as much as for you. We run the chain side end to end: no wallet, no token, no transaction fees.
Is the candidate's personal data written to the chain?+
No. Only cryptographic digests are written to the chain — never a name, a score, or a transcript. Personal data never leaves the platform; the seal only proves the report's integrity.
Does the AI make the decision?+
No. The AI evaluates the session, ties scores to evidence, and writes the report; the hiring decision always belongs to a human. The report is a screening aid — the system filters, the human decides.
How long does setup take?+
There is no setup. You write the task, the candidate joins via an invite link; engineers work from the terminal (cli-han), other roles from the browser. You can have your first report the same day. Questions: omer@tayf.io
This is how it looks for your candidates too.
Write a task, invite a few candidates, read the reports yourself. You still make the decision · but this time you have the evidence in hand.
Closed beta What is HAN?
HAN han.tayf.io · omer@tayf.io · the system filters, the decision is human cli-han · live-environment sub-product ↗ only digests on the chain · never a name, score, or transcript © 2026 · Bursa, Türkiye