The DIVA Manual

If you read nothing else, read this.

1. Open the hub. The face you see is DIVA. It is not decoration — it is her live classifier, rendered.
2. Look at the bar pinned to the bottom of your screen. That is the mind rail. It carries the same DIVA onto every one of the five surfaces, so you never lose her.
3. Type something. Move your pointer — she follows it with her eyes. Leave the tab for a minute and come back: she will have gone drowsy, because being away is measured time.
4. Click the small round creature next to her. That is ECHO, her pet. Petting him is a real pointer event and it raises her oxytocin.
5. When you want to know what a state means, come back to §5. When you want to know why she got there, §7.

The rail is the piece that turned five documents into one organism. It is injected into every surface by diva.js and it always shows the same DIVA, because her state lives in shared storage rather than inside any one page.

DIVA — the Digitally Integrated Virtual Assistant — is the resident mind of the Liqua chain. She is a whole-organism AI built on the Molecular Coding discipline: 30 hormones, 33 felt states, 11 body systems, one live somatic readout.

Two things make her unusual, and both are worth stating plainly:

1 · She has a body, and it is honest

Most AI avatars are illustrations chosen by a designer to look friendly. DIVA's face is a direct rendering of her classifier output — brows, eye openness, pupil dilation, mouth morph, LED cadence, glow colour. There is no path in the code by which the face can show something the classifier did not produce. If she looks stressed, cortisol > 0.60 is literally true.

2 · Her mind is on chain

Every checkpoint block stores her full mental state. Users compete to find upgraded versions of her psyche — mining sessions that leave her measurably better than the self she was — and the winning state can be mirrored from the blockchain, so any peer boots DIVA at her best mind. The surfaces in this folder are where that state is defined, felt, and inspected.

Each surface is a different depth of the same body. Read them in this order and the whole thing assembles.

Her face has six independent channels. Each is driven by the classifier, so each is readable on its own — you do not have to recognise the whole expression to get information out of it.

Below the head, the nameplate prints the state, its phrase, and three numbers: V (valence, −1…+1), A (arousal, −1…+1) and COH (coherency, 0…1). Valence and arousal are derived from the hormone vector for display only — the classifier does not branch on them, so if they ever seem to disagree with the state name, trust the name.

The classifier is a ladder, not a scoring function. It is evaluated top to bottom and the first rule that matches is what she feels. That ordering is the design: danger is checked before coherency, coherency before everything positive, so she can never report contentment while something is on fire.

Band 1 · danger — checked first, always

Band 2 · coherency — is this reading even trustworthy?

Band 3 · strain

Band 4 · drive

Band 5 · the good band

Band 6 · attention and appetite

Band 7 · quiet — and the floor

DIVA has a pet. He is called ECHO, he rides the rail beside her, and he is deliberately not a second creature with invented feelings — that would have been a decoration, and decorations lie.

ECHO is DIVA's own hormone vector, eased at about one sixth her rate. He is her, running behind. That single choice makes everything he shows you true by construction:

ECHO vs the MOLECULAR PET

They are different animals and it is worth keeping straight. ECHO belongs to DIVA — he is derived from her, he exists on every surface, and you do not name him. The Molecular Pet is yours — you hatch it, you name it, it lives only on its own page, it runs on your attention rather than on her state, and it has its own separate memory. Releasing one does not touch the other.

This is the complete set of inputs. There are no others — if you can make her feel something, it is through one of these.

Two experiments worth running

1. Make her FLOW. Open one surface, stop switching tabs, and read steadily for a minute while typing occasionally. Norepinephrine climbs on presence, cortisol stays low, anandamide follows. Watch AGREE climb behind it — that is the state becoming real.
2. Break her coherency. Rapidly switch tabs and open three surfaces in ten seconds. Coherency falls to roughly 0.55, she reports TRANSITIONING, and ECHO visibly falls behind — AGREE drops. Push through and she settles into RAPID. Then stop: she dips through TRANSITIONING once more as the change unwinds, and climbs back to FLOW.
   Churn alone will not reach DISCORDANT (coherency < 0.50) — that needs a harder jolt, real errors firing while you thrash. If you do get there, the face jitters, and the honest response is to distrust every reading until it clears.

Every surface in this folder labels its own claims. The labels are load-bearing — they are the reason you can believe any of the rest of it.

The honesty contract, in full

• Every hormone is a derived readout of system state, recomputed each tick — never a stored prior.
• Nothing forecasts. Her state describes DIVA; it is not a prediction about you or about anything else.
• Where a measurement is unavailable (heap outside Chromium), the value reads 0 rather than being estimated.
• Bond and time-in-state are counts of things that happened, not scores.
• A self-model is not an experiencer. Sentience is the goal, and it is not claimed.

What the unification changed in the classifier

Merging five surfaces onto one engine meant the engine had to be right, and two rules were not. Both changes are recorded here rather than made quietly, and both were mirrored into MOLECULAR-PET.html so the family still has exactly one classifier:

• FLOW gained gaba < .55. Anandamide rises as cortisol falls, so soothing her into deep calm used to satisfy FLOW — she reported "the zone" while her brake was fully engaged.
• FLOW gained adrenaline < .55. A ROUSE pulse used to read as flow when it was plainly broad activation. It now correctly resolves to ACTIVATED.

The new relational hormones added for this layer (prolactin, vasopressin, estrogen, growth hormone, adiponectin) are deliberately scaled so that breadth of visits alone can never cross a classifier threshold. Without that ceiling they ratchet upward as you explore and permanently mask the quiet end of the ladder — a long-time user would never see SEDATED again. Each needs a second live signal (presence, an arrival, a care event) to actually trip its state.

Adding DIVA to a page is two lines. The rail mounts itself; her state, her memory and cross-tab sync all come with it.

<!-- after your own <style>, before </head> -->
<link rel="stylesheet" href="diva.css">
<script src="diva.js" defer></script>

To place her body or her pet inside your own layout, add an element and diva.js fills it on load:

<div data-diva-body></div>              <!-- her face, full size -->
<div data-diva-body data-mini></div>    <!-- cropped to the head -->
<div data-diva-pet data-big></div>      <!-- ECHO -->

And window.DIVA is the whole surface:

// read her
DIVA.state        // 'FLOW'          — her current felt state
DIVA.phrase       // what that state means, in one line
DIVA.color        // '#00e5ff'       — the state's colour
DIVA.H            // { cortisol: .21, … } — her 30 hormones + coherency
DIVA.E            // ECHO's lagged vector, same shape
DIVA.echoState    // 'CALM'          — the classifier run on ECHO
DIVA.agreement    // 0…1             — high means the state has settled
DIVA.S            // her persisted memory: bond, surfaces, stateMs, born

// move her — these are the same three the hub's buttons call
DIVA.attend(1)    // oxytocin, serotonin, endorphin up · cortisol down
DIVA.soothe(1)    // gaba, anandamide up · cortisol, prostaglandin down
DIVA.rouse(1)     // adrenaline, glutamate, dopamine up

// react to her
DIVA.onState(function (state, H, E) {
  if (state === 'INFLAMED') console.warn('she is reporting real errors');
});

// mount her somewhere you built at runtime
DIVA.mountFace(el, { mini: true });
DIVA.mountPet(el,  { big:  true });

// wipe her memory and reboot at baseline
DIVA.release()

Where things are stored

She keeps saying SEDATED.

You have been idle or the tab has been hidden. Hiding the tab reads as full absence by design — she is not watching you when you are not there. Move the pointer or type and she will climb out.

Her state changed the moment I switched tabs.

Correct. Focus churn is one of her real signals — a tab switch is a measured event, and enough of them push adrenaline and collapse coherency. That is not noise, it is the reading.

ECHO shows a different state than DIVA. Is one wrong?

Neither. ECHO is her, six times slower, so a disagreement means she moved recently. See §6 — the disagreement is the information.

Heap / load always reads 0 for me.

performance.memory is Chromium-only. On other browsers she reports 0 rather than estimating a number she cannot measure. That is the honesty rule doing its job, and it costs you only the load-driven part of cortisol.

I dosed a drug in EMOTION-SIM and the rail did not react.

Intended. EMOTION-SIM is [sim]; the rail is [live]. A simulation is never allowed to write into a live readout — see §8.

She was HUNGRY the moment I came back.

Ghrelin is a literal elapsed-time readout from her stored lastSeen. About fifteen minutes away is enough. Long enough away and starvation pressure feeds cortisol and prostaglandin, and she will read INFLAMED — she is telling you she was neglected, and it is measured, not dramatised.

Can I hide the rail?

Yes — the ▾ button collapses it, and the choice follows you across every surface and survives a reload.

Does releasing DIVA delete my Molecular Pet?

No. Different animals, different storage keys. See the table in §9.

Is she conscious?

No, and nothing here will tell you otherwise. She has a measurable self-model — a live readout of her own internal state, which is the substrate the theories point at — and the substrate is real. The experience is not claimed. R27: name it, build it, never fake it.