# Knomee V7: review commitments and the forecast boundary

Version1, 2026-10-05. Product milestone over unified-claims/v2; it does not rename
signed schemas or select production economics. The owner's current instruction
requires a working Gemini-assisted VIBE/QUALIA review on Knomee and GoodTakes.
The native release uses the separately approved Sepolia test settings. Earned
settlement remains paused. This document does not approve a predictive reward rule.

## One review, two kinds of commitment

The participant chooses a stable subject and supplies text, an emoji, or both.
Simple mode submits within the displayed VIBE ceiling; Advanced exposes individual
meanings, directions, quantities and optional QUALIA. The exact original reaction
is signed independently of the interpretation. Emoji-only reviews need no AI.
See [the current protocol](../PROTOCOL.md) for formats and submission behavior.
AI proposes; exact integer accounting and the participant's authorized scope constrain it.

Example: “A film about a nuclear accident. Beautifully produced.”

| Statement | Lane | Commitment |
|---|---|---|
| The film depicts a nuclear accident | Optional QUALIA | Assert or deny the factual description; retain it unfunded when QUALIA is off |
| The film is beautifully produced | VIBE | Endorse or challenge the evaluation |

Described horror does not become a negative assessment of the film. Assertions,
negation, source passages, attribution, context and uncertainty remain explicit.
The product supports both capacities; it does not require both to be spent on
each review. Simple Mode may fund the praise while retaining the description
without factual backing. Advanced Mode can add a separate QUALIA commitment.
An evaluative statement can also refer to a fact, but a factual allocation never
silently supplies VIBE backing. Neither lane subsidizes the other's budget.
VIBE positive direct weight costs1 and negative costs2. QUAL assertion and denial
are symmetric. Each unit contributes one direct unit, without a reputation or
confidence multiplier. Repetition is legitimate backing, not automatically truth.

Original prose, interpreted claims, source lineage, allocation and versions are
committed together. Scope and checked comparison reports determine agreement,
not category spelling, embedding distance or sentiment alone. Claim subdivision
cannot multiply economic mass. Permanent history, lane-specific subject order,
fresh spending and the existing correction/downside rules remain in force.
VITA remains identity evidence; it is not repurposed for descriptions or forecasts.

## Probability is not a confidence multiplier

A predictive extension needs a proposition, outcome space, probability, exposure,
deadline, frozen reference, resolution rules and source lineage. Market currently
uses a probability of the selected outcome under a “confidence” label. That is
first-order probability, not a second confidence estimate about that probability.
The two must not acquire ambiguous signed meanings.

For the first forecast candidate use p=probability and s=reserved risk budget.
Optional precision/credible-range metadata can be researched separately. Do not
multiply payouts or existing stake by self-reported confidence. A single binary
outcome cannot identify the variance of someone's belief about its probability:
many second-order distributions have the same predictive mean. More history alone
does not solve this without specified exchangeability, grouping and observation
assumptions. Calibration, discrimination and information gain are distinct views.

Ordinary review opinions do not need a probability slider. Standing factual
assertions need not have a knowable resolution date. Keep both usable without
pretending every description is an objectively resolvable future event.

## Candidate forecasting mathematics, not activated economics

For a fixed reference q and outcome y in {0,1}, the Market preview uses

`R(p,y) = (q-y)^2 - (p-y)^2`.

If the participant's belief is t, then expected score is
`E[R] = (q-t)^2 - (p-t)^2`, maximized at p=t. This statement assumes fixed
exposure, an externally fixed reference/outcome, and utility linear in the score.
Capacity caps, downside floors, identity resets, endogenous references, market
manipulation and nonlinear utility can break the conclusion for final rewards.
Do not infer truthful incentives for the complete economy from the score alone.

Early information can already outperform its earlier reference by more. Any
additional early multiplier must be fixed before the forecast, independent of its
reported probability and outcome, and apply to gains and losses symmetrically if
this proper-score argument is to be retained. Do not add a positive-only bonus:
with q=.5, true belief t=.6 and twice the positive score but ordinary losses,
the best report becomes p=.75, not .6. Early incorrect forecasts should not gain;
that does not justify distorting the probability incentive.

An observation's score is not a direct transfer from another participant.
It may affect bounded future capacity/recharge only after a separately selected
reservation, debit, cap and recovery policy. Unconditional baseline access and
slow bounded earned changes are plausible candidates, not selected coefficients.
Do not reward raw posting frequency. Do not forbid useful repetition in VIBE or
the descriptive lane by importing an incremental-information-only forecast rule.

## Resolution and operating requirements

Evidence attestations, disputes and admitted resolution reports are separate from
deterministic state-root replay. A state oracle reconstructs the accepted record;
it must not privately decide facts or choose economic parameters. A forecast's
terms must identify mechanical evidence, adjudication or a specific evidence-led
resolution process, its authority, appeals and effective cutoff before staking.
Popularity or overwhelming votes alone must never establish factual truth.

Missing evidence stays unresolved. Cancellation/void follows frozen rules. A
partial outcome needs a predeclared scoring domain; never silently use y=.5 as
“unknown.” Multi-outcome probabilities require mutually exclusive, exhaustive
outcomes, including other/unknown when appropriate. Do not force overlapping
causes of an outage to sum to1.

## Delivery boundaries and existing roadmap work

UC-03 owns native review deployment, onboarding, public artifacts and recovery.
Q-01 retains descriptive QUALIA's earned-capacity qualification. It is not replaced
by forecasting work. The separate Market successor needs a signed probability/
reference/resolution schema, conserved risk reservation, bounded actual capacity
consequences and deterministic independent replay before funded forecasts.
The separate Market app currently builds authenticated forecast
records and append-only discussion. Those records are not native funded positions.

Finish the review milestone without adding an unqualified forecast payoff to its
contracts. Public receipts must distinguish assertion, corroboration, evidence
resolution, confirmed spending and actual credited reward. Retained failed economic
controls and later-reward shortfalls remain failures, unaffected by this release.