Confidence and Observability ── How Far to Trust a Reading

Even when the L number is out, the job is not done. Two L3 readings can differ wildly: one backed by five concrete past events told without contradiction, another resting on a single event and only the person's own word. Without a number for that gap, a thin reading and a thick one get combined at equal weight. It is like a health check: you would not treat a blood pressure measured many times the same as one taken once. Episode 8 sets confidence C, which says how sure a reading is, and observability o, which says whether the rater could see that item at all, then passes their product as the weight into the tally.

Readings you may finalize, and readings to hold

In Episodes 6 and 7, each rater read the highest level a person actually showed (we call this level L). But readings differ in thickness. Some are backed by many concrete examples; others are shaky. The source (Part 3.2) names this certainty confidence C. When C is low, the L number is not finalized even if it has been computed. It is held, and you either add evidence or have someone else look.

An analogy: judging a cooking contest by tasting one dish and declaring someone a master is risky. Only after several dishes, all consistently good, can you say it with confidence. Why separate final from held? Because mixing a thinly supported verdict into the final result drags the whole thing toward it. Three conditions let you finalize: enough evidence, a story that holds together, and whether the person could see it at all. Miss any one and C falls, leaving the reading provisional. Let us take them in turn.

The three questions behind confidence C

A formal-looking formula appears, but it is three easy questions. C scores three things — (1) is there enough evidence, (2) is the account free of contradictions, (3) was the rater placed to see it — each from 0 to 1, then adds them up. Skip the formula; remember the three questions and you have it.

First, is there enough evidence? Call this saturation. Count the concrete events (supported pieces of evidence); three earns full marks, one earns a third. Why three? A single event might be luck. Only something repeated counts as real ability. Second, is the account free of contradictions? Call this coherence. Do the same person's experience, reasoning, and values fit together? If the "why I judged so" from a hard case clashes with another story, the score drops. Third, was the rater placed to see it? Call this observability, covered next.
(As a formula: C = weight x saturation + weight x coherence + weight x observability, where saturation is "count divided by 3, capped at 1.")

Observability o — no weight without sight

Observability o measures whether the rater could actually see that item. The source (Part 4.1) sets it as a product of two things: being well placed to see (positional fitness), and actually producing evidence. The product matters. Even a well-placed rater who cannot produce a single concrete piece of evidence drives the second number near zero, so the product o falls too.

Why a product? To refuse weight based on a title alone. In soccer, a referee in a good spot still cannot blow the whistle if they did not actually see the foul. Same here. A manager, for instance, is well placed to see how much trust a report carries (trust density). But if they cannot recount a single concrete scene where the person's judgment was pushed back or upheld, their o on that item is low. Conversely, a field owner who received a flag holds living evidence — how they revised their own first draft — on the narrow items of communication and getting people to act on their own. So o is high there.

Who sees what, by standpoint

Who can see which item is set by who was actually present, not by title. The source's staffing table gives a starting point for visibility. As with tasting food, whoever was in the kitchen sees the cook's skill best.

These figures are only a start. Real o appears only after multiplying by "did they produce evidence." Even in a high row, real o falls below the tabled number when no evidence comes out. Sitting in a good seat does not prove you watched.

Weight w = visibility x certainty — how strong a voice is

Finally, multiply observability o (could they see it) by confidence C (is it certain). This is the weight w, how strong a voice is when opinions are combined. The more a rater could see (high o) and the thicker and more consistent their evidence (high C), the more strongly they bear on that item's verdict. Thin evidence, or an unsighted standpoint, is made light automatically. Why a product? Because if either side is weak, we want the whole to be weak: seeing without evidence means nothing, and evidence without sight cannot be trusted. Next episode's corroboration sets the outside view (the other-level) from the level a majority of this weight supports.
(Formula: w = o x C.)

How to staff the raters

Once o and w are numbers, you can judge a panel before measuring. The source's rule is simple: each item must be covered by two or more people with visibility o of 0.7 or higher. Why two? If only one rates high and no second pair of eyes backs it up, the item was never really measured. Only with several judges can one person's oversight or mistake be caught. The table distinguishes a sound panel from a broken one.

What to do when C is low

Do not force a low-C reading to final. The source is plain: too few events, a contradictory account, or no standpoint to see it — any of these lowers C and leaves the reading provisional. The fix splits by symptom. Too little evidence (low saturation): draw a few more events from the same person. A contradictory account (low coherence): check the clash with the person and use concrete facts to settle which was the real behavior. No one who can see it (low observation): add someone who can see that item to the panel. The shared point across all three: each thickens the evidence rather than raising the score. Evidence sets the score. Raising it without evidence is like rewriting a health-check number instead of measuring again — not allowed.

Confidence C and observability o do not touch the L number. L stays the highest level the evidence actually shows; C and o are a separate measure deciding how much weight a reading may carry. Their product, w = visibility x certainty, becomes how strong a voice is in next episode's tally (third-party integration).

If, when staffing, each item is covered by two or more people at visibility 0.7 or higher, the formula makes thin evidence and unsighted readings light on its own. Hold certainty as a number. That is what stops one high rating from moving the whole.