Skip to content
C.W.K.
Stream
Lesson 04 of 05 · published

Deferred, and Said So on the Surface

~12 min · deferral, substitutes, honesty, scope

Level 0Raw Ore
0 XP0/36 lessons0/12 achievements
0/100 XP to next level100 XP to go0% complete
"Deferred, stated on the dashboard as deferred. No synthetic CAPE." — the gauge spec, stating an intention this lesson will end up checking

When the data genuinely is not there

The previous lessons were about approximations that are honest and available. This one is about the case where no honest approximation exists at all.

The long-window valuation gauge needs ten years of real earnings at index level. For the US, an academic assembled that series. For the Korean and Japanese markets, no equivalent public series exists. Not hard to obtain — absent.

The options are the familiar three, and the middle one is the trap.

Ship it with worse data. Assemble something from partial inputs and call it by the same name. Now the dashboard has a gauge that is not comparable to its US sibling, sitting right next to it, under an identical label. This is worse than not shipping and worse than saying nothing.

Ship nothing and say nothing. The card is absent. A reader notices the asymmetry and cannot tell whether it is a bug, a loading state, a permissions issue, or a deliberate choice.

Ship the absence. A surface that says the gauge is deferred, why, and what to look at instead.

Which of those this product actually does depends on which deferral you look at, and the split is worth more than the tidy version would be. The sector deferral is genuinely shipped: the dashboard carries it in prose, names all three probed paths, and dates the decision. The long-window gauge deferral is not — it is stated in the gauge spec, in exactly the words "deferred, stated on the dashboard as deferred," and the dashboard says nothing of the kind. There is no card, no reason, no named substitute in the interface at all.

So the model case here is the sector one, and the other is this lesson's own moral pointed back at the product: a deferral that lives only in a design document is the gap, not the design. Two deferrals, one team, one week apart, and only one of them made it to the reader — which is roughly the hit rate you should expect from a rule that lives in prose rather than in a check.

The substitute matters as much as the deferral

The third option is only strong because of what comes with it. The deferral names a substitute: the index price-to-earnings ratio, in percentile against its own accumulated history.

Notice what that substitution preserves and what it gives up. It gives up the smoothing — the long-window metric exists precisely to average out the earnings cycle, and the substitute does not do that. It preserves the question: expensive relative to its own history? That is what a reader actually wants to know, and it can be answered with data that exists.

So the deferral is not a hole. It is a downgrade, named, with the reader pointed at the best available answer to their underlying question rather than to the unavailable answer to their literal one.

A deferral with a named substitute is a design; a deferral without one is a gap. When you cannot build what was asked for, do not merely report the impossibility. Work out what the request was actually for, and deliver the best available answer to that — clearly labelled as the substitute it is. The difference for the reader is enormous, and the difference in effort is one paragraph.

Deferrals that were measured, not assumed

The same discipline appears in the sector-aggregate deferral for one market, and there the record is even more useful: three specific paths were probed and each was closed for a stated reason — one source required a session, one sat behind a bot wall, one carried no usable fundamentals on the available library.

That is a deferral you can act on. A future reader knows exactly which three doors were tried, which means they neither repeat the work nor assume it was never done. And the reopening condition is named: revisit if that source publishes a real interface.

An unnamed deferral is a silent drop. "We will do that later" with no reason, no substitute and no reopening condition is indistinguishable from forgetting — and a few months later, nobody can tell which it was. The named version costs three sentences and stays actionable indefinitely.

Code

Two deferrals, both stated so a future reader can act·text
DEFERRED: long-window valuation gauge, two markets
  reason:      no public 10-year real-earnings series exists at
               index level for those markets
  refused:     a synthetic version under the same name
  substitute:  index P/E percentile vs its OWN history --
               answers the same underlying question ("expensive
               relative to itself?") with data that exists
  surfaced:    the SPEC says "stated as deferred on the
               dashboard". The dashboard does not say it.
               Written down, not shipped.

DEFERRED: sector aggregates, one market
  probed and closed, three paths:
    1. official exchange endpoint -- session-gated
    2. archive route             -- bot wall
    3. local sector ETFs         -- no fundamentals available
  rule invoked: no scraping past a gate (the spec's own rule)
  reopen if:   that source publishes a real interface

# One of these two reached a reader. Both are actionable years
# later from the documents -- but only the second is actionable
# from the PRODUCT, which is where a deferral has to land to be
# a design rather than a note.

External links

Exercise

Find three things your team has deferred and check each for the four parts: the reason, what was refused, the substitute, and the reopening condition. Most deferrals have one of the four. Fill in the other three for one of them, and notice how the conversation about it changes once the substitute is named — usually the deferral stops being a source of low-grade guilt and becomes a decision.
Hint
The substitute is the part almost everyone skips, and it is the part the user actually experiences. 'We do not support X' and 'We do not support X; here is what answers the same need' are the same engineering state and completely different products.

Progress

Progress is local-only — sign in to sync across devices.
Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)
Sign inPlease sign in to comment.

No comments yet — be the first.