A free, offline-capable study site for Thailand's civil-service exams — built by someone who sat these exams, got tired of how everyone else learns them, and taught himself what actually works.
Read it · 12,000 exam takers and counting · เริ่มอ่าน
English (README.md) · ภาษาไทย (README.th.md) · 中文 (README.zh.md)
You are going to be tired. You are going to open this on a bus, or at 1am, or five weeks before the test and think I should start earlier. That thought has cost more preparation months than any subject ever will.
So let me say the thing nobody says:
You will get most of this wrong on your first pass. That is the design.
This site does not start by explaining things to you. It starts by sitting you down with eighty exam topics and asking you to answer them blind. You will miss most of them. Each miss then hands you the rule — the thing that decides the answer — and asks you the same question again. You do not move on until you get it right.
That is not a motivational trick. It is borrowed from real research on how memory actually forms, and it is the single decision that separates this site from a flashcard pile.
Read The loop below before you decide how to use this. It is four minutes and it will save you weeks.
I am a consultant, not a lawyer and not a developer. I have watched a lot of people prepare for these exams, and watched myself prepare badly. What I found is that almost everyone studies the constitution the way you study history: read the chapter, highlight the important sentence, feel good, score badly.
Three things are wrong with that, and this site is the answer to each.
Highlighting feels like work because it is effortful. That is precisely why it is not evidence of learning. Decades of research agree: self-testing and spacing outperform rereading and highlighting by a wide margin, and explanation, interleaving and imagery help only depending on the task and the learner (Dunlosky et al., 2013). Spacing alone is well established as a genuine effect (Cepeda et al., 2006), and there is a recent, more careful review of how retrieval should be structured (Carpenter et al., 2022).
So: every question here is a test. Answering it is the studying.
Most apps treat a miss as a penalty, colour it red, and move on. That throws away the best information in the room. A miss tells you precisely what you do not know — which is exactly the thing worth spending the next hour on.
Here, a miss creates a debt. It comes back until you answer it correctly. Not until you memorise the sentence, and not until you tap through it twice — until you can rule the other three options out by reasoning. That is the skill the exam measures, and it is the only one.
If you memorise "section 213 petitions go through the Ombudsman first", it helps for exactly one question. If you learn "a request must go to the administrative body before it can reach the court, and only when that body declines or goes quiet for 60 days" — that rule answers section 213, and it answers the next question shaped like it.
Every one of the 284 Pokklong questions and all 90 full-mock questions in this site carries an explicit rule alongside its explanation. A test enforces that no two questions share a rule, because a copied rule is a rule that only works once.
This is the whole product. Everything else is a place to get to it.
<svg viewBox="0 0 860 260" xmlns="http://www.w3.org/2000/svg" role="img"
aria-label="The learning loop: sit blind, miss, learn the rule, answer again, earn it">
<style>
.box { fill: none; stroke: #111; stroke-width: 2; }
.fill-a { fill: #E4002B; } /* primary red */
.fill-b { fill: #F5C518; } /* primary yellow*/
.fill-c { fill: #0057B7; } /* primary blue */
.lbl { font-family: ui-sans-serif, system-ui, sans-serif; font-size: 15px; fill: #111; }
.lbl-w { font-family: ui-sans-serif, system-ui, sans-serif; font-size: 15px; fill: #fff; }
.num { font-family: ui-monospace, monospace; font-size: 13px; fill: #111; }
</style>
<!-- step 1: blind -->
<rect class="box fill-a" x="10" y="40" width="150" height="96"/>
<text class="lbl-w" x="85" y="72" text-anchor="middle">Sit it blind</text>
<text class="lbl-w" x="85" y="94" text-anchor="middle">80 topics.</text>
<text class="lbl-w" x="85" y="114" text-anchor="middle">Nothing to read.</text>
<text class="num" x="16" y="30">01</text>
<!-- step 2: miss -->
<line x1="160" y1="88" x2="205" y2="88" stroke="#111" stroke-width="2"/>
<polygon points="205,88 195,83 195,93" fill="#111"/>
<rect class="box" x="207" y="40" width="150" height="96"/>
<text class="lbl" x="282" y="72" text-anchor="middle">You miss it.</text>
<text class="lbl" x="282" y="94" text-anchor="middle">Most of them.</text>
<text class="lbl" x="282" y="114" text-anchor="middle">That is the point.</text>
<text class="num" x="213" y="30">02</text>
<!-- step 3: the rule -->
<line x1="357" y1="88" x2="402" y2="88" stroke="#111" stroke-width="2"/>
<polygon points="402,88 392,83 392,93" fill="#111"/>
<rect class="box fill-b" x="404" y="40" width="180" height="96"/>
<text class="lbl" x="494" y="70" text-anchor="middle">Get the rule.</text>
<text class="lbl" x="494" y="92" text-anchor="middle">The thing that</text>
<text class="lbl" x="494" y="112" text-anchor="middle">decides the answer.</text>
<text class="num" x="410" y="30">03</text>
<!-- step 4: again -->
<line x1="584" y1="88" x2="629" y2="88" stroke="#111" stroke-width="2"/>
<polygon points="629,88 619,83 619,93" fill="#111"/>
<rect class="box fill-c" x="631" y="40" width="150" height="96"/>
<text class="lbl-w" x="706" y="72" text-anchor="middle">Answer again.</text>
<text class="lbl-w" x="706" y="94" text-anchor="middle">Rule out the</text>
<text class="lbl-w" x="706" y="114" text-anchor="middle">other three.</text>
<text class="num" x="637" y="30">04</text>
<!-- earned -->
<rect x="700" y="170" width="150" height="52" fill="#111"/>
<text x="775" y="202" text-anchor="middle" font-family="ui-sans-serif,system-ui,sans-serif"
font-size="16" fill="#fff">Earned →</text>
<path d="M 706 136 L 706 158 Q 706 170 718 170 L 700 170" fill="none" stroke="#111" stroke-width="2"/>
<!-- loop back -->
<path d="M 85 136 L 85 196 Q 85 208 97 208 L 620 208 Q 632 208 632 196 L 632 140"
fill="none" stroke="#111" stroke-width="2" stroke-dasharray="6 5"/>
<polygon points="632,136 627,146 637,146" fill="#111"/>
<text x="358" y="228" text-anchor="middle" font-family="ui-monospace,monospace"
font-size="13" fill="#111">wrong again → the rule stays up, and it comes right back</text>
<text x="10" y="250" font-family="ui-sans-serif,system-ui,sans-serif" font-size="13" fill="#555">
A question is not finished when you have seen the answer. It is finished when you can derive it.
</text>
</svg>Read that bottom line again. It is the entire argument of this project.
| 279 sections | The full 2017 Constitution (B.E. 2560) including the 2021 amendments, Thai primary, English (Council of State) and a Chinese study translation |
| 837 recordings | Every section read aloud in all three languages, generated locally and checksummed against the source text |
| 470 questions | Written from the text for practice, with explanations and citations |
| 284 Pokklong questions | A twelve-step ladder through Thai local-government administration |
| 90 full-mock topics | The 80 assessed topics of the district-chief exam, with an elimination rule on each |
| 13 guide chapters | Thirteen chapters that teach the reasoning rather than the list |
| Works offline | Installable, no account, no tracking. Your progress stays on your device |
Most people trying to learn ปกครอง (to govern) memorise section numbers. This does the opposite. It builds the logic from the ground up, one step at a time, where each step only needs the one before it.
<svg viewBox="0 0 860 300" xmlns="http://www.w3.org/2000/svg" role="img"
aria-label="The twelve Pokklong steps from why a state needs administration to control and accountability">
<style>
.s { stroke:#111; stroke-width:2; }
.t { font-family: ui-sans-serif,system-ui,sans-serif; font-size:12.5px; fill:#111; }
.n { font-family: ui-monospace,monospace; font-size:12px; fill:#111; }
</style>
<!-- asanoha-inspired grid of twelve cells -->
<g>
<rect class="s" fill="#E4002B" x="10" y="20" width="130" height="80"/>
<rect class="s" fill="#fff" x="150" y="20" width="130" height="80"/>
<rect class="s" fill="#0057B7" x="290" y="20" width="130" height="80"/>
<rect class="s" fill="#F5C518" x="430" y="20" width="130" height="80"/>
<rect class="s" fill="#fff" x="570" y="20" width="130" height="80"/>
<rect class="s" fill="#111" x="710" y="20" width="130" height="80"/>
<rect class="s" fill="#fff" x="10" y="110" width="130" height="80"/>
<rect class="s" fill="#F5C518" x="150" y="110" width="130" height="80"/>
<rect class="s" fill="#fff" x="290" y="110" width="130" height="80"/>
<rect class="s" fill="#0057B7" x="430" y="110" width="130" height="80"/>
<rect class="s" fill="#fff" x="570" y="110" width="130" height="80"/>
<rect class="s" fill="#E4002B" x="710" y="110" width="130" height="80"/>
</g>
<text class="n" x="16" y="14">01</text><text class="t" x="16" y="44">Why a state</text><text class="t" x="16" y="62">needs</text><text class="t" x="16" y="80">administration</text>
<text class="n" x="156" y="14">02</text><text class="t" x="156" y="44">The three</text><text class="t" x="156" y="62">tiers</text>
<text class="n" x="296" y="14">03</text><text class="t" x="296" y="44">The centre:</text><text class="t" x="296" y="62">who heads what</text>
<text class="n" x="436" y="14">04</text><text class="t" x="436" y="44">Province</text><text class="t" x="436" y="62">and district</text>
<text class="n" x="576" y="14">05</text><text class="t" x="576" y="44">Self-rule</text>
<text class="n" x="716" y="14">06</text><text class="t" x="716" y="44">Chain of</text><text class="t" x="716" y="62">command</text>
<text class="n" x="16" y="104">07</text><text class="t" x="16" y="134">Legal acts:</text><text class="t" x="16" y="152">orders and rules</text>
<text class="n" x="156" y="104">08</text><text class="t" x="156" y="134">Who may</text><text class="t" x="156" y="152">decide</text>
<text class="n" x="296" y="104">09</text><text class="t" x="296" y="134">Fair</text><text class="t" x="296" y="152">procedure</text>
<text class="n" x="436" y="104">10</text><text class="t" x="436" y="134">Appeal,</text><text class="t" x="436" y="152">revocation</text>
<text class="n" x="576" y="104">11</text><text class="t" x="576" y="134">Enforcement</text><text class="t" x="576" y="152">and notice</text>
<text class="n" x="716" y="104">12</text><text class="t" x="716" y="134">Control and</text><text class="t" x="716" y="152">accountability</text>
<text x="10" y="225" font-family="ui-sans-serif,system-ui,sans-serif" font-size="14" fill="#111">
<tspan font-weight="600">Guess first, then the why, then the logic so far.</tspan>
</text>
<text x="10" y="248" font-family="ui-sans-serif,system-ui,sans-serif" font-size="13" fill="#555">
Each step adds one idea to a running summary. By step 12 you can rebuild Thai
</text>
<text x="10" y="266" font-family="ui-sans-serif,system-ui,sans-serif" font-size="13" fill="#555">
bureaucratic logic from first principles instead of recalling section numbers.
</text>
</svg>Credit, and the "stolen, improved, credited" note. The topic list and its
habit of asking for numbers and deadlines come from แนวข้อสอบ พ.ร.บ.วิธีปฏิบัติ
ราชการทางปกครอง พ.ศ. 2539 by ประพันธ์ เวารัมย์. Every question here was
rewritten — none copied verbatim — and law texts were checked against Thai
Wikisource. We also corrected errors in that source (listed in
source/slideshare-2539-INDEX.md) and added two areas it omits entirely: the
structure of the state, and accountability under the Tort Liability Act 2539
and the Official Information Act 2540.
That is the whole ethic of this project: take the good structure, fix the errors, and say where it came from.
I am not a psychologist and these are not my findings. What follows is what I read, what I found convincing, and how each one became a specific decision in the code — including the places where I deliberately did not follow it.
| Idea | What it says | What it changed here |
|---|---|---|
| Testing effect (retrieval practice) | Pulling something out of memory strengthens it more than putting it in | Every question is a test first. Explanation comes after the attempt, never before |
| Spacing effect (Cepeda et al. 2006) | Reviews spread over days beat massed cramming | A missed card returns at 10 minutes, then 1, 3, 7 and 14 days |
| High utility ranking (Dunlosky et al. 2013) | Practice testing and spacing are high-utility; rereading and highlighting are low | Highlighting and re-reading were removed from the product entirely |
| Task-dependent effects (Carpenter et al. 2022) | Explanation, interleaving and imagery help differently per learner and per task | These are offered, never asserted. The site never claims your outcome |
| Retrieval before options | Recalling before choosing beats choosing without recalling | You recall first, then see the options. Skippable — but the default |
| Desirable difficulty | A little struggle at the right moment buys a lot of later ease | The full mock gives you no reading first. That friction is the feature |
- The 1/3/7/14-day schedule is a starting heuristic, not an optimum. Nobody has shown it is best for you.
- The app does not grade your written explanations. It stores them and shows them back to you. A machine pretending to mark reasoning it cannot assess is a worse teacher than silence.
- There is no AI coach here. No model is grading you, tracking your politics, or deciding what you should believe. Everything runs in your browser. This is adaptive retrieval practice — not a trained system and not a claim about your opinions.
- This site's own effect on your score is untested. I have not run a controlled study. What I have is reasoning and a test suite.
Not in a classroom. I am a consultant — I work with clients on strategy and organisations, not on law — and everything here came from being curious in the wrong order.
I learned most of it the way I learned most things: badly first, then correctly. I read the constitution once, highlighted heavily, and scored worse than the person who had simply done past papers for three weeks. That gap was embarrassing enough to be interesting. If effort felt like progress but produced nothing, then effort was not the mechanism, and I needed to find what was.
So I went and read the actual literature. Not because I enjoy it — because the cost of being wrong for four months is much higher than the cost of a uncomfortable afternoon with a paper. Retrieval practice, spacing, desirable difficulty: once I saw the effects were real and large, "just reread the notes" stopped being a defensible way to study.
Some of it came from elsewhere. I have always learned systems rather than facts — you cannot admire Championship Manager or play a football game for twenty-five years without noticing that they are entirely about acquisition, scouting and probability rather than talent. Watching a good side win a match they should have lost teaches the same lesson as a paper: what did they know that their opponent did not, and how did they know it in time? That is exactly the exam question. Which organisation can do something, and who is allowed to authorise it?
I also learned from failing at this. My first version of this site was a flashcard pile, because flashcards are what I had always used, and it did not work. It was replaced because the data was against it, not because it looked wrong. I would rather delete a feature I liked than keep a feature I could not defend.
I am a consultant writing about law. I have read the text carefully and I have had every chapter checked against the Thai original by people who did not work for me. But this is not legal advice, it is not an official syllabus, and a typo in a legal system is a much worse thing than a typo in a menu. Every page carries its source and its date. Check it. If it disagrees with the Gazette, the Gazette wins — and please tell me.
Next.js, statically exported, served from a Cloudflare Worker. No backend, no accounts, no analytics, no tracking. It installs to the home screen and works with the network off.
app/ routes — every page is static HTML
components/ the interactive surfaces
lib/ corpus access, learning model, navigation
data/ the constitution, the questions, the guide — all plain JSON
scripts/ validators, build steps, and the release gate
tests/ 164 tests: corpus, questions, translations, cases, UI
Two design rules that are enforced, not just intended:
Every number must be proved by the text. A test walks all 41 cram-sheet entries and asserts each one quotes a phrase that actually appears in the Thai of the section it cites. If a deadline is wrong, the build stops.
Every claim keeps its source. Figures and news carry a source and a retrieved date. Where sources disagree, the page says they disagree rather than picking the convenient one.
Shipping is not "it compiled":
npm run release # validate → test → build → generated-file drift → links, colour, contrast- Generated-file drift — the build regenerates four
data/*.jsonfiles that are committed so the site can import them. If they no longer match a fresh build, the commit shipped a stale index. - Link integrity — every internal link in every built page, plus the sitemap and all three podcast feeds, must resolve.
- Undefined colour tokens —
var(--x)that no theme defines silently falls back, which is how the cram sheet once sat at 2.47:1 in dark mode. - Contrast by WCAG class — text 4.5:1, component boundaries 3:1, decorative hairlines reported but not failed. One flat bar produces false failures on borders and misses real text.
- Redirect routes are declared, not guessed — a forwarder passes every
existence check while being useless to index, so
REDIRECT_ROUTESinlib/nav.tsstates which ones they are and the gate fails if one is advertised.
git clone https://github.com/Nonarkara/constitution.git
cd constitution
npm ci
npm run dev # http://localhost:3000
npm run release # the full gateRequires Node 24+. No API keys, no database, no accounts.
The code is released under the MIT License — see LICENSE.
The Constitution itself is a public document; the Thai text is authoritative and the English translation is the Council of State's, with study corrections noted on each page. The Chinese is a study translation and is marked as such.
The section recordings are synthetic, generated locally from the Thai text and checksummed against it. They are not recordings of the Council of State Senate audio, which is linked separately and retained under its own terms.
Ponklong's topic structure is credited to ประพันธ์ เวารัมย์'s 2539 exam material as described above: rewritten, corrected, and attributed.
Research citations: Dunlosky et al. 2013 · Cepeda et al. 2006 · Carpenter et al. 2022
Good luck. Get it wrong the first time — that is the part that works.
Built by Dr. Non Arkaraprasertkul and a small number of agents that were told to check him.