v1.7.0 — What Is That Model, Actually?

This release is about a single question: what is that model, actually? Until now, the game knew and the player didn’t — a brain’s value was a black box that only opened when you sold it. Benchmark cracks that box open: you measure the model, you get a grade, and paper on that grade has a price at the vendor’s desk. And because there are two ways to measure, we also added a wager — a cheap in-house probe that lies downward, and a certification that tells the truth but ends the brain by writing it down. Plus a rebuilt half of the screens and a wall of achievements with their own icons.

A quick catch-up note before we start. July shipped v1.6.6 and v1.6.7 with two operational changes that didn’t get their own web devlog. The rewarded-ad boost slot was replaced with a Packet Filter minigame round — same slot, same cadence, no ads. And a beta German localisation joined the interface language list, alongside the existing English, Czech, Spanish (Latin America) and Portuguese (Brazil). Both are already live; this note is your web-side heads-up.

Now, on to 1.7.0.

🎯 Benchmark — finally, you know what you built

A new screen (opened from the Telemetry tile) puts a brain on the table and computes how well it turned out. Not “how you’re doing against other players” — how this specific brain turned out. Rolled traits count into the result; benchmark’s job is to reveal them, not normalise them away.

Everything the model does as a model goes into the score:

• The rolled weight quality of installed modules and upgrades.
• Domains you actually mastered — one point per domain that landed.
• The MP stats (Learning / Ru / Domain / Tolerance / Spec) — the biggest single line item, which is exactly why rarity shines through into the score even before you run the brain at all. An emergent is a better brain than a baseline before anyone does anything to it — that isn’t a hole in the design, that’s the whole difference between them.
• All of that is then multiplied by brain quality (untrained through maxed-out) and the software installed in the perk slots.
• And frustration and exhaustion drag the number down — a burnt-out brain measures worse. At high overload the benchmark refuses to measure at all; you don’t get a bad grade, you get “come back when the brain isn’t screaming.”

The result lands as a grade on a twelve-step scale:

D · D+ · C · C+ · B · B+ · A · A+ · S · S+ · S++ · S+++

🔬 Probe vs. certification — a wager, not a button

There are two ways to measure, and each one has a catch:

The in-house probe — available any time, repeatable, priced in Ru. It’s not a certified instrument and it reports like one: every line of its output sits below the truth, not just the final number. It never overshoots, so it’s a lower bound — enough to decide “is this worth certifying?”, useless for pretending you already know the answer. Repeated probing on the same brain won’t close the gap, either.

Probe cost climbs steadily across your career — a global counter that never resets, not even when you sell a brain and start a new one. There’s a hard ceiling on the price so it doesn’t become an arithmetic wall on long careers, but the ramp is real and it’s on purpose: probing has to feel like a decision, not a reflex.

Certification — free, but only at a vendor’s desk (the certificate is their requirement, not yours) and irreversible: a certified brain will never run again. It records the truth without distortion and, in doing so, closes the brain. So the dilemma is: keep training and risk overshooting, or write down what you have and go sell?

And here’s where the two paths meet — the probe reports low, which means if it shows the grade a vendor is asking for, certification will confirm it. That’s what the probe’s Ru is for.

📄 Papers — and the price of not having one

Sell a model without a certificate and the buyer shaves the price — a modest cut, visible on the contract preview rather than sprung at delivery. The fiction is straightforward: nobody has verified what this brain can actually do, so nobody pays a premium.

Any certificate cancels the shave — even a D. On a weak brain a certificate is still worth taking; it just doesn’t earn you a bonus on top.

Above the shave sits a grade premium, and it’s paid two different ways depending on who’s buying:

• Vendors who don’t read the paper pay a modest premium tied to the grade band. They never open the report, so their offer can’t be a strict function of the grade itself; it’s a glance and a haggle.
• Regulators who require the grade pay a substantially stronger, fixed premium. Their payout is a promise, and a promise can’t be a dice roll.

🏛️ Three ways vendors look at certification

We split the market deliberately narrow — better to start with too little than too much:

• Private buyers never ask. Bring the paper if you have one; they’ll pay for it, and if you don’t, they still buy.
• The state does not buy uncertified brains. Every government vendor requires a certificate — no paper, no offer. But they don’t read the grade.
• One corporate vendor per sector has the same requirement.
• And three regulators, across the whole game — Federal Market Authority, Strategic Response Authority, and Ministry of Continuity — name the grade: minimum B, and pay a substantially higher premium for meeting it. The bar is deliberately low because the probe reports low: if you’ve already measured a B on your own bench, you know it’ll pass.

🏆 Weekly leaderboard — probe voucher + GPU shake-up

• A new leaderboard reward: Probe Vouchers, scaled to bracket — 1 / 2 / 3 / 5 per tier. A voucher lets you skip a probe payment without denting your Ru.
• A rebalance pass on Cloud GPU — the leaderboard’s GPU tiers were flattening out, so each rung is now meaningfully different again.

🖼️ Achievements — a wall of icons

The row-per-achievement list was replaced with a three-across tile grid, and all 79 achievements got their own custom icon (the same ones you already see on Play Games).

• Locked tiles don’t show the art — the icon hides behind the NeuronIdle mark. Want to see it? Unlock it.
• Combos collapsed into a single ladder. Four steps of the same idea (10 / 25 / 50 / 100) used to draw four tiles with the same motif; now it’s one tile carrying the highest step you’ve reached. It shows how many you have unlocked, never how many exist — otherwise it’d give away how many secrets sit behind it.
• Hidden achievements no longer say ”???” when they unlock (which looked like a bug); they now say SECRET.
• The unlock popup shows the achievement’s own icon on a more readable background, instead of a generic plaque.
• Titan Brain has been renamed Colossus Brain — the name matches the model tier, and the Play Store achievement was renamed in step.

🧰 Rebuilt screens

Half the game got the same treatment: the header and footer stopped scrolling away, lists became grids, and details open inline where you tapped.

• Hardware: a three-screen chain collapsed into one screen. Values in Current Build are coloured by rarity. And the card no longer auto-opens on low durability — instead you get a dot from 75 % and a pulsing battery from 50 %, so it informs you without deciding for you.
• Market: five big cards became HUB-style chips arranged 2 + 3. Contracts is green against the Black Market’s red — the two you use most, in the colours you already read them by.
• HUB: the purple “brain slots” accordion is gone; slots are visible all the time.
• SW slots: the FILL SLOT screen is now a grid with a detail panel instead of a truncated list, with a Defense / Hack filter. Blocked papers are tappable and explain why (missing fragments are named, no more “insufficient”). The recipe reads as have / need.
• Telemetry: a status stripe and a tile grid using your custom icons. The Benchmark tile shows your career-best grade, not the current brain’s — because certification is terminal, the current brain’s grade would almost always read ”?”.
• Best-of results: benchmarks hold the top five per architecture, and certified results now sort in Best Brains by grade, not by raw score. A Micro with an A is a better piece of work than a Huge with a C, even if the raw number is an order of magnitude lower — and the board now reads that way.

---

Neuron Archiv records the peaks. Module Optimization records the habit. Benchmark, now, records what the model actually is — and paper on that record has a price. The long game just got a new kind of receipt.

— NeuroVault Research Systems