Poorup Redesign - Devlog #6

This phase was mainly about adding more depth to the player interactions, improving the bot system, and making the rankings feel more alive.

Deal Negotiation

I redesigned the deal flow so trades, loans, equity, and hybrid deals all work in the same way. When you receive a deal, a modal appears with accept and negotiate. If you close it, the deal does not get declined and stays inside the Finance rail. Both players can now see the same deal collapsed in Finance. The receiver can negotiate, while the sender can adjust or cancel the deal. You can open a full read-only view, change the parameters, and send the new version back. The server also checks old deal IDs so an outdated modal cannot change a newer deal.

Sponsored Purchases

I also added sponsored purchases. If a player cannot afford a property, they can ask the table for help. Other players can reserve contributions, and once the full amount is available, the buyer accepts and is forced to buy that exact property. The money is automatically returned if the request is canceled, the property changes owner, or a player disconnects. It is not a loan and does not create equity.

Bots

I continued improving the AI and no-AI bots. The bots now receive more context about the board, current obligations, loans, events, opponents, future rent risk, expected cash flow, and their own previous decisions.AI bots can now also counter trades, negotiate player contracts, choose debt rescue actions, and make auction decisions. If the AI provider runs out of credits or fails, the no api key AI takes over

Rankings

I added recent trend indicators to the ranking rows, so you can see if a player is moving up, down, or staying flat. I also added a seasonal ranking scope based on the current calendar season.

Rules Fixes

I also fixed the Double GO setting so landing exactly on GO now pays the boosted amount, while simply passing GO still pays the normal amount.TestingI expanded the tests and ran long bot campaigns with auctions, loans, casino, market, global events, sponsorships, and negotiations. The latest regression suite passes and the longer 2,500-game campaign had zero stalled games.

Next Steps

The next major step is making the bots simulate several complete future turns and improving long-term negotiation behavior. After that I want to add deeper browser testing and seasonal ranking rewards.
Also to lazy to do screenshots of new UI so you will get everything in the next devpost

Abi Valut - Last Devlog

This phase about cleaning UI elements changing the language of the whole demo website to english and fully preparing it for the offical stardance submission as the offical private one was alredy online for a good while and had near to 0 bugs.

Cards

This isnt a feature newly added but I just wanted to mention that our system in theory supports cards through gocardless but that isnt setup so its not usable at the moment though the backend exists and for a while even Ui existed but I have removed that.

Translation

This wasnt so hard just pretty time consuming going over everything and translating every single text.

Refelection

Pretty proud of myself as I managed to push out a full working apllication of course this isnt 100% selfcoded for example backend features but it is more then fine for what I was aiming to do and It works perfectly.

Trying out

here is a link to the demo for everybody wanting to try it out. https://abi-vault-demo.vercel.app/

Poorup Redesign - Devlog #5

This phase was mainly about fixing all bugs still exisitng, refactoring further with codescene and letting AI run over my code, also fixed some small missing features and simple bugs myself.

Choosing Icon

I noticed a bug in which because of logic I made before setting up custom icons the programm mapped custom icons people created onto persets and blocked them so for example your icon had red in it, the game decided to map it onto the red preset in the backend blocking it but showing yours in game, this came because I never rewrote this section after introducing custom icons.

Ownership icon

Before this change owenship on board was only shown with a simple like quadrat in the top right corner of the tile in the color you picked this is also something left over from the old preset logic. I decided to make it so It shows a smaller version of the face of the player which makes it look a lot cleaner.

Missing Buttons

This was mainly because I let AI write to much of my code, the finical trading tab which is centered around equity and everything missed a dropdown to choose who you send the trade to and even the button to send the trade which was a dumb oversight on my end. I also added a simple backend system for the dropdown inside that modal to only show the propertys a person owns.

Bots

I also continued working on the bot system though this was mainly watching tutorials and brainstorming further

Next Steps

Fix last bugs, fully add bot system and make sure UI looks good to release the first beta test hopefully the next post will be the last

Poorup Redesign - Devlog #4

This phase was mainly about setting up codescene doing a full restructuing with the help of AI and fixing bugs and adding missing features. I also added a simple, gambling and stock market system.

Bots

I first started with setting up the initial idea of bots so to the UI and a bit of the backend logic for bots and how they will interact with the game iam nowhere finished and I havent tried it out yet.

Codescene

After checking out the github studentpack again after a while I saw codescene, which is a code reviewer similar to coderabbit. I struggled a bit with the Setup but after running an analysisi I noticed how abysmall my codebase really is and how its missing structure, archtiectuer and etc, for example sepearting certain getures from game logic.js and creating their own thread for it. I also setup a full pr pipieline which I can use with my AI agents which aim will be fixing my architecture, and finding bugs before i push them.

bugs

I ran Qwen 3.8 flash on my pr pipileine with codescene and let it work indepnedlty to full restructer my codebase so the results are better and I have better architectural design I also let it fix any bugs it found.

Bugs I found

While trying out the game i found mutipile BUGS
one of them affecting my secret minigame palor patrol. For some reason I was losing hearts for no appearent reason. The reason behind it was that even though UI was hidden the helicopters were still flying through. For some reason the heart bug still exists afterwards and even the AI isnt able to find out what it is so Iam completating making architecture changes to it.

#Gambling
I added a smile gambling system where you can bet money on a roulete on red, black, and green I didnt add any animation for it as I found it unecessary and I also made sure iam legally clear to make sth like this

Stock market

I created a simple stock market tab and backend which allows people to buy indexes of a property and made sure to connect it with global events so global events can affect the stock market but that is nowhere near usable.

Next Steps

Iam going to let my AI continue improving my codebase with codescene and continue the Stock market and afterwards add the bots.

Poorup Redesign – Devlog #3

This phase was about expanding the social side of Poorup, adding more depth to the game, and planning future systems that make every round feel more alive.

Social, Rankings, Achievements

I added a complete friend system. Players can search for unique usernames, send and accept friend requests, remove or block players, report users, invite friends to rooms, and view public profiles without leaving the lobby.

Profiles now include statistics such as wins, win rate, games played, average money, bankruptcies, achievements, and match history.

I also added global rankings for wins, win rate, games played, achievements, and other statistics. Rankings and Social are now independent tabs instead of modals, with wider layouts that use the available screen space more effectively.

Achievements now have custom pixel-art and SVG icons, rarity levels from Common to Mythical, filters, dates, categories, and clickable detail modals. Mythical achievements are extremely rare and can trigger a server-wide announcement for every player.

Easter Egg

I added the hidden Parlor Patrol minigame. Pressing CTRL + P on the home screen opens a secret mode where you shoot down police helicopters, planes, and drones. It includes custom explosions, animations, sounds, score tracking, and a secret achievement.

Global Events

Global Events are rare events that affect every player instead of only the person who landed on a tile. They can be triggered by round conditions or Surprise cards and can last several rounds.

Ideas include:

• Housing Bubble Pop
• Credit Freeze
• Inflation Spiral
• Bank Run
• Interest Rate Shock
• Airport Strike
• Energy Crisis
• Tax Scandal
• City Election
• Rent Control
• Tourism Boom
• Supply Chain Breakdown

Events can combine with each other, create unexpected situations, and unlock secret achievements. I also want a global-event setting that is simply on or off, while event strength and duration scale with the round length.

Rules Page

I added a complete Rules tab designed like a pixel-art game manual. It explains the board, movement, properties, loans, cards, achievements, global events, social features, and settings. It uses an internal book-style scroll so the main application does not become a long scrolling page.

Future Systems

I am also planning a casino system using virtual money, with red, black, and green bets where the odds are not equal.

Another planned feature is a fictional stock market connected to properties, countries, airports, utilities, and global events. Events could crash markets, increase rent values, affect airports, or create temporary investment opportunities.

I also want to add unethical and outrages bank loans with high interest, strict repayment rules, and serious consequences.

Music/sound icons

I also added simple music and sound icons for the buttons instead of having the music on/off text

Next Steps

The next phase will be making these systems work together while keeping the game balanced, and also adding AI bots which utilise AI like deepseek v4 and so on so they able to actually judge the situation and think of possible ways for them to win. The aim is to use hcai to power them. I will also add normal BOTS as a fallback if my usage runs out

Abi Vault - Devlog 10

This phase was just about creating a demo website for the stardance people. Creating One new tab and cleaning up the final UI.

Finish

I finished with fully creating the full application privatly and fixing all bugs Ui and backend and also created one new tab which was interseting.

Stardance Demo

I made a second copy of my repo removed all clerk dependecies translated all the text from german into english and setup a second Supabase and vercel project this will be used without loging which allows for everybody to use and test it.

Next Steps

creating the readme, cleaning up a bit and its ready for ship

Poorup Redesign – Devlog #2

This phase was about continuing the complete redesign of the website, deciding on a UI theme, and implementing a new feature that I had wanted from the start but was not advanced enough to create before.

New UI

I am very proud to announce a complete redesign of the application. I decided to fully embrace a pixelated theme, and I personally think it looks great. It is genuinely one of the best UI systems I have ever designed.

New Feature

I decided to expand the trading system. You can now give out loans in exchange for premiums, equity, money, and more. This allows you to create fully customizable loans, such as:

I give you $200 for 30% permanent equity in this spot.

This means that you have to pay 30% less rent when you land on it and gain 30% of all rent paid on that plot. This is a system I have never seen in a Monopoly game, and I wanted to implement it from the start when I first had the idea for Poorup.

New Feature #2

I have also decided to take the advice I was given and implement a room system. This allows players to join and create rooms, as well as make them public or private. I also added a quick-table feature that matches you with other people who have pressed the button, allowing you to play a quick round with random players.

New Feature #3

I have created a complete profile system. You can now create a profile, view your statistics, win percentage, average money, and more. You are also able to create custom 32×32 icons to use while playing the game.

New Feature #4

I also decided to add sounds and an open-source soundtrack I found called Pondering the Cosmos. which i think fits my new theme perfectly but maybe iam going to change it

Next Step AI Bots

The next Step will be adding bots but normal bots are boring they are just glofired search engines why i want to be a bit creative.

I will use hackclub ai provider and give the bot a brain by combinging them with an AI like deepseek v4 flash this means the bot will be able to check how is he performing what are the most logical steps sell buy properties and trade and if wanted even chat with the user to make a deal would anybody like that?

Poorup Redesign

So after my last ship i read through all the reviews and have decided to continue this project even though i really didnt plan for that. So there were 3 addition people wanted. A new UI, being able to join random people, better Animations and a Dev mode to test in solo player possibly with mods.

What did i do?

So i first started with thinking of a new design route so i asked chatgpt for some references of a good design. I decided on making the Whole Ui pixellated similar to the Arcade style and fully redesingen the SVG witth and now iam going to go over it and adjust it to fit ours. Of course this took like around 6 hours but I use figma on desktop and the browser tracking of hackatime doesnt seem to work how do you like the new svg?

Next Steps

Next Steps i redesignen the Full Ui to fit this pexelated style and making sure it looks good

Abi Vault - Devlog 9

This phase was mostly about moving everything from development to production, using the actual domain, finalising the system, rewriting a bit of the structure and fixing bugs.

UI

So I finalised the iPad UI and added multiple simple spinner animations and slide-in animations. I also worked a bit more on the blank states, what the placeholder text should say and how everything should look.

Right now I am still in the process of finalising the phone UI, as the blank states fucked it up a bit.

Logic

I rewrote 2 large sections.

As I realised in the last devlog that I couldn’t use a bank API, I needed to rewrite the logic of how the platform worked.

I decided to keep the cards and make them a visual representation of the physical cash.

I also decided to fully transition the system to being ledger-based, which was pretty obvious since we now don’t have an API.

For anybody asking, ledger-based basically just means that the system calculates the balances and tracks how everything is performing based on the values that users manually enter.

i also decided to create an invite system link so people cant just signup if they somehow find the software and it only works by getting a link from the Admin. For the Supervisor role its a 30 day link for up to 30 people and for admin its a onetime link for up to 7 days and 1 user.

Domain

I also decided to buy a domain, which I originally didn’t want to do as I thought Vercel was all I needed.

However, Clerk required a custom domain so I could move everything from development to production.

Also, for anybody asking, I will not reveal the name of the domain as this is a closed platform. That’s also why I made it as badly SEO-optimised as possible in the metadata.

And for anybody wondering, this domain cost me A WHOLE $0.07 a month.

Quite expensive if you ask me.

Clerk

Clerk was quite the hassle because, as mentioned before, my plan was to just use Vercel, but Clerk essentially forced a custom domain onto me.

Another thing I didn’t know was that under my student plan I was only allowed 2 roles and wasn’t able to add another custom one.

This became a bottleneck as I originally had 3 roles in mind:

• Student
• Admin
• Supervisor

In the end I decided to scrap Admin and make it only Student and Supervisor.

Supabase

This part was quite simple as most of it was already finished.

I mainly needed to work on the role permissions and also add a caching system because I noticed that some pages would load for 1–2 seconds every time.

That’s just bad UX, especially when it happens even during a simple tab switch.

The caching system now keeps already-loaded data available where possible instead of unnecessarily fetching everything again.

Next Steps

The next steps is now creating a demo version and a better opensoure version so people can use my software without depending on my vercel, supabase and etc. Or maybe I will host another workspace of this exact software for hackclub so people can try it what would be your advise for that?

MIRA - Revision 2

This phase was just about bug fixin, way to many bugs i dont know where they even came from.

Pipeline

I first started with the research pipeline for some reason it wasnt really activated like the CLI wasnt running because I initialied the pipeline in the wrong order so I fixed that.

Camera Engine

After the reviewer came back to me and told me that my camera engine for the dashboard wasnt working I say down and did a whole redesign to make sure it works and looks good. The main bug was just as before that I didnt cover edge cases and initailised the camera engine in the wrong order.

#Website
I did another redesing of the website though not much changes i just rewrote logic and cleaned up the docs page again.

Next Steps

So first of all for some reason this devlog is 10 hours idk how its this much prob a kind of bug which I need to analyse. I will now continue cleaning up the entire codebase for bugs and hopefully push it one last time and get accepted

Abi Vault - Devlog 8

This phase was mainly about the backend which i have to confess i had to use AI for as it was complicated with supabase clerk and just rotuing everything together.

Failed bank API

My plan was to allow the use of real bank cards and getting all their information through a bank API, but i missed sth important either you handle all bank apis from all providers yourself or you get a service to do it. The problem is there is no free Opensource service which allows for this and to keep it a buck iam broke asf. So i scraped the Idea and made it only ledger based which means it gets calucalted automatically but you have to write down all transactions and etc.

SupaBase

This was quite hard as Supabase is hella confusing prob one of the most confusing UI i have ever seen
but after some back and forth with the Ai i was able to set it up though i still need to learn a bit as I really didnt undestand what i did there.

Clerk

this was quite easy clerk really allows for easy synchronisation and it was quite easy to set up the 3 classes admin, supervisor, and student I also made it so the software is request only for the moment so i can send a request with a certain authorisatzion and the User creates their account. But will probally change that soon

Next Steps

I will probally either get a domain grant or just buy it from my own money and prepare to launch the website over nest. Its nearly finished and the onlz fixes are small UI mismatches, better animation adding things like skelleton loaders and etc.

NIMBL - Devlog #7

for some reason i got way to many hours on this project and this is prob because my hackatime tracks my benchmarks so Iam gonna request that they like cut it in half idk, dont wanna be fraud reviewed. This phase was about resetting and reconfiguring

Benchmarks

The benchmarks were horrendus and I realized that something is going insanly wrong so i decided to do a full reset. Scrap my architecture and go back to a opencode baseline architecture and just remove and add diferent parts. this took only 6 hours but as said my hackatime is bugging prob hoing to try removing 30hours because i just noticed it doesnt make sense.

Next Steps

Optimizing the archtitecutere and not running anymore benchmarks till Iam sure of this architecture also contactiing stardance cause aint now way iam getting fraued reviewd.

Abi Vault - Devlog 7

This section was mainly about adding all other pages creating a fully nice looking functionial phone UI and just cleaning up the UI in general.

Pages

I created the rest of the pages and created place holders for them and now a whole system is in place the only thing left is writing real logic and conecting it to a server to get user auth.

Phone UI

This was the biggest phase of this Project i first started with scraping our already existing phone UI and started gathering references I decided to use klarna as they got quite the clean UI. I first started with adding the sidebar to the bottom to make it easier accsessible. I also needed quite A long time to make the UI look clean with adding animation for the modals and making them appear from underneath and not spawn in the middle of the screen

Next Steps

Now that the UI is kinda finished and school starts in 1 week the next step is creating a fully functional backend syncing the application with superbase and other services so I have a first working demo which I can continue working on during the school year.

NIMBL - Devlog #6

I noticed that the last few devlogs were quite lazy and to be frank trash. So this section is just getting you upto date. Explaining the problem, goal, benchmarks and architecture and just giving you a short summary.

What is NIMBL?

NIMBL which stands for nimble which describes a person as being fast, agile and efficent.

Nimbl’s goal is it to allow users to get a coding harness which is half the cost and dobule as fast and allows for them to keep around 80% of the intelligence.

It also includes 4 modes learn,explain,plan and build. Plan and build are just the typical modes you see in all harness like opencode and explain is a copy of plan but adjusted, so it shows examples and is able to retrieve the codebase ligthning fast and offer detailed explanation. Learn is the same as build but with measurements in place so you cant just prompt it rather it give you example code or code with simple mistakes and you have to fix them yourself.

What have I done?

So since fully porting the Opencode tui I have decided to work on nimbles effectivnes I first starded with indepth analysis of other harness and how they work their benefits and their drawbacks. So for example how Pi is ligthweight or how Aiders gathers context. Afterwards I summarized everythign i learned and implemented it, and then came the benchmarks i first started with a local 75 test benchmark across all kinds of test from lh, mf to retrival tasks on which nimbl performed great with being 3x cheaper and achieving a 98% solve rate.

Benchmarks

The next step was running NIMBL on offical benchmarks for that I choose SWE Lite as it one of the more reputable ones and smaller ones. OF course I wouldnt be able to run even that size as it around 120GB and I dont have that much money for benchmarks. Just so you know evey benchmark I have run used openrouter credits and Iam the person paying for them. So i sliced it, Slicing means instead of running all you take out a certain part and download only that and run only that.
On that slice NIMBL was miserable solving 1 out of 15 task. After further optimizing the logic behind it NIMBL is now at a point where it reaches around 80% accuracy and is a third of the price.

How does NIMBL work?

NIMBL's main architecture is built around a layered context and execution pipeline.

The project is indexed using lexical search, structural parsing, and dependency-graph relationships.

When a task is received, NIMBL classifies it and retrieves only the most relevant files, symbols, and related dependencies.

The retrieved context is compressed and placed into a stable prompt structure, allowing provider-side caching to reduce cost and latency.

The agent then works through the task using tools, tests, diagnostics, and bounded execution loops.

The result is tracked through token usage, cache hits, cost, latency, and success metrics.

This allows NIMBL to stay lightweight while still having access to the important parts of a codebase. The goal is not to send less context blindly, but to send better context.

Combined with the different learning modes, this gives NIMBL the foundation to become a faster, cheaper, and more educational coding harness.

in the screenshot you can see how much credits i got left i had 10 at the start of the development. Hopefully this made NIMBL easier to understand and gave a short overview

MIRA - Revision

This devlog is about fixing bugs and redoing parts of my website. It seems like my project won’t be approved because of the website and the amount of AI used for it, which I fully understand.

There is just one problem: the website was never the focus of MIRA. It is purely aesthetic and was made so people can get a quicker overview of the project.

Still, I complied and rewrote around 1000 lines of code, changing spacing, logic and other parts of the website. But it is still not being approved, so the next step is creating a separate, simpler website just for Stardance so the project can finally be approved.

Bugs

While working on the website, I wanted to test the dashboard and noticed that it didn’t seem to work anymore. I analyzed what was wrong but couldn’t really understand why.
The dashboard was my first experience using FastAPI, but from the logic I understood, it should have worked. It just didn’t. So instead of trying to patch it, I decided to redo the whole backend logic of the dashboard and adjust it to work more like MIRA Live.
This fixed the bug, but even afterwards I still couldn’t fully figure out what exactly caused the old version to stop working, even after asking AI and going through the code multiple times. As it had been working before

Website Redesign

Before deciding to create a completely separate website as mentioned above, I tried redesigning and cleaning up the current website as much as possible.

I didn’t touch the color palette because I chose it myself for a specific reason, and I honestly can’t identify MIRA with another color scheme anymore.
Instead, I removed things I wouldn’t have been able to code myself, like certain slide-in animations and more complicated spacing. I also rewrote parts of the logic that I didn’t properly understand before and replaced them with simpler solutions that I actually understand.

Next steps

Hopefully i can get approved soon and will be able to finnaly work on the next project. Iam probally goint to try one more time to redesign the website and if that doesnt work I will create a new one which i just think is unecssary, because as I said MIRA is a computer vision model and research pipeline coded mostly by hand and the website was just generated quickly for asthetic reasons and truth be told isnt necessary at all

NIMBL - Devlog #6

This phase was still about benchmarking optimizing reading papers to see if i can implement some new logic and benchmarking again. I burnder through atelast 400mil tokens which were used in the benchmarks whihc like 10$ openrouter credits and a bit of opencode go

Benchmark

The results were overhemlingy postive with nimbl achieving a 300/300 fix while being 2.2x cheaper accros all task combined this came mainly from retrival which is understandable as nimbls architecture is build on the idea of rag and we further refined it but even on lh(long horizon) and mf(multi files) it was great

#Next steps
Damn i take everything i just said back iam runnign offical swe lite tests now and nimbl is performing like shit because of our inbuild steps limit which i forgot its unable to solve the tasks and is failing, so its back to bug fixing see you in a few days.

NIMBL — Devlog #5:

This Devlog ran another benchmark: 25 tasks × 3 samples, hidden golden tests only, and the same model on both harnesses (deepseek-v4-flash). I also used these results to fix the mistakes I found and run the benchmark again, which is running rn.
NImbl is ca 40% more token efficent half the price and caches better.

The Results

NIMBL: solved 68/75 tasks, with total tokens of 2,386,450, wall time of 37.7 min (2,264 s) and throughput 1,054 tk/s.

opencode: solved 75/75 tasks, with total tokens of 2,436,642, wall time of 71.2 min (4,270 s), throughput 571 tk/s.

NIMBL finished 1.9× faster and actually used fewer total tokens than opencode. The tradeoff is that it didn’t finish 7 tasks, and those failed runs still spent real tokens, so this wasn’t a cheap excuse to give up early.

On the 4-mode run NIMBL also used around 40% fewer tokens per solved task (26.2–28.0k vs opencode’s 46.8k, −40% to −44% in every mode). That shows the savings were never only coming from retrieval tasks.

By Category results

bug-fix tasks: −7.2%
retrieval tasks: −51.1%
multi-file tasks: −16.3%
running agent delegation: +130.2%
long-hunt: −35.3%
shell-loop: +6.3%
test-write: +116.9%

What It Says

The results are pretty decent, and they are getting better. Retrieval now uses half the tokens at a third of the latency: 7 seconds compared to opencode’s 19 seconds.

NIMBL is faster in every category, even delegation, where opencode takes 116 seconds per task. The architecture is doing great, and reference cost lands at about half of opencode’s (~$0.26–0.28 per solved task vs ~$0.51).

The losses are concentrated in two specific places. Delegation costs 130% more tokens — 373,493 compared to 162,256 — and test-write costs 117% more and it did all this and still failed so it practically wasted tokens

The reason for that is simple: we had a hardcoded limit of 8 tool calls, which wasn’t enough for these long tasks, opencode ran ~100. So NIMBL kept reading files and crashed when trying to run an edit, because it ran out of tool calls. The fix was basic: instead of hardcoding every tool call limit universally, adjust them according to the task.

Running Now

A new benchmark is running rn trying to use these fixes to finally close that gap of tool calls and get fully functional harness which is cheaper and better after local benchmarks work I will run it in a real benchmark like swe lite or sth

NIMBL - Devlog #4:

This devlog is about the one thing NIMBL was built for: proving it saves tokens without making answers worse. I built a real benchmark, ran it against opencode, and got cooked.

The Tier B benchmark

I generated a ca 150-file TypeScript “storefront” app with planted bugs, decoy files, and hidden golden test suites. 25 tasks spread across retrieval, bug-fixing, multi-file work, test-writing, shell loops, delegation, and long-horizon fixes.

The key part: every bug is only caught by hidden tests, so an agent can’t cheat by reading the grader. Ran it live on the same model for both harnesses: deepseek-v4-flash, 3 samples per task, 75 runs per side.

And critically — every single result is saved raw, not summarized. Full event streams, tool calls, token splits, everything, so nothing can be hidden.

The results

solved mean tokens NIMBL (hybrid) 68/75 28,004 opencode 75/75 23,352

opencode solved more and used fewer tokens. That stings, but it’s the truth, and the raw data is committed.

Where NIMBL genuinely won: retrieval tasks (+63% fewer tokens) and latency (26s vs 49s per task). Pure lookup-and-answer is where the context index shines.

What went wrong

• Shell mismatch (the big one). NIMBL runs commands through PowerShell but never told the model that. So the model wrote POSIX (find | sort, 2>&1, head) and 60 of 113 shell calls failed. opencode tells the model it’s PowerShell, and the same model wrote working commands. The fix is a one-line hint in the tool description.

• Analysis paralysis. One long-horizon task spent 78 tool steps reading files, never ran a single test, never edited anything. All 3 samples failed while opencode fixed it every time.

• Delegation overhead. Subagent “research” cost 7k-27k tokens per task with no payoff. opencode solved the same tasks inline for a fraction of that.

• Hallucinated answers. One retrieval task answered “42” with zero tool calls, never reading the file that had the answer.

Where it stands

NIMBL has permissions, sessions, providers, budgets, context selection, structural extraction, tokenizers, and a real benchmark harness that saves raw output for every run. It’s genuinely usable for simple coding.

But the benchmark says what the benchmark says: the token win is retrieval-only for now. The next devlog should be about the shell fix, the delegate budget, and seeing if the numbers move.

Abi Vault - Devlog 6

This section was mainly about cleaning up the UI, making the different pages more synchron and creating a more structured UI system for the whole website.

UI System

I worked on making the design more consistent across the dashboard, transactions and receipts page. Before that many aspects were styled slightly different like the spacing, typography, buttons, dropdowns and card sizes. I started creating shared rules and components so the pages follow the same visual identity and feel more connected.

Transactions and Receipts

I created and implemented the full transactions page with filters, a filter modal, pagination, placeholder transactions and a structured transaction table. I also created the receipts page with placeholder receipts, status filters, pagination and a modal for adding new receipts. Both pages use the existing Abi Vault sidebar, topbar, colors and general design language.

Dashboard Cleanup

I added more placeholder transactions to the dashboard so the transaction history feels more complete and does not leave as much empty space. I also added another spending category and adjusted the review section so the right side feels more balanced with the transaction history on the left.

1920px and 2560px

A big part of this session was cleaning up the layout for both 1920px and 2560px screens. The layouts looked very different depending on the screen size because of adaptive heights and spacing. I adjusted the cards, rows, table containers and page spacing so the website stays inside the viewport and does not become scrollable on the supported resolutions.

The 2560px version was especially difficult because there was a lot of empty space, but simply scaling everything up did not look good. Instead I had to create separate wide-screen adjustments, add more rows and balance the different cards while keeping the 1920px layout unchanged.

Results

This was a productive session but also quite difficult because many small CSS changes affected multiple pages. The UI now feels much more consistent and the Transactions and Receipts pages are much closer to the final design. I still need to refine some spacing differences between the pages and test everything more carefully, but the main structure is now in place and the beta version is getting closer. For anyone wondering about the phone UI I havent touched it in a while and decided to just create it at the end because its such a hedache