energy: the operating point, the residual, and alignment with network-layer identities #86

Merged

dcharlot merged 1 commit from wai/energy-operating-point into main 2026-09-10 12:22:43 -04:00

Conversation 0 Commits 1 Files changed 2 +106

2026-09-10

Transaction Science

energy: the operating point, the residual, and alignment with network-layer identities ...

The energy field is contested, not empty. The prescriptive half is a decade old
and ratified — ISO/IEC 23001-11 carries decoding-operation-reduction metadata
whose purpose is to make a decoder do less work, and a companion carriage part
now moves it. WAI's half is the measured one. Neither half currently scores
against the other, and that gap is where a joule figure stops meaning anything.

DEGRADATION-BLINDNESS (§7.1). A receipt that records joules spent, and nothing
about how much work the decode was asked to do, rewards the wrong behaviour: a
sink can always spend fewer joules by decoding worse. energy-measurement.md §5's
recomputable-work-unit rule blocks the substitution at the byte-exact tier, but
not at `entropy-consistency`, and not for the conventional codecs WAI already
carries. So the receipt has to carry the operating point, not only the cost.

REQUEST / COMMITMENT / DELIVERY (§7.2). Three distinct fields, because the
interesting cases are exactly the ones where they differ. A receipt MUST NOT
carry `op_delivered` alone — that is the degradation-blind case. `op_committed`
MUST be what the receiver stated even where it differs from the request, since a
commitment that silently becomes the request is not a commitment. And a sink that
honoured a reduction request MUST NOT claim `decode-equivalence`, per the tier
rule.

This converts a recorded number into a number measured against a promise. The
prescriptive standards say what should be spent; the measurement standards say
what was; nobody scores the second against the first.

THE RESIDUAL (§7.3). Where a complexity descriptor and a measured binding cover
the same reconstruction, the receipt MUST carry the signed difference between
predicted and measured energy — over the same work unit, never without the
acquisition class of the measured term, and never suppressed for being large.
The large ones are the informative ones, and suppressing them turns a falsifiable
claim into an unfalsifiable one.

A complexity descriptor is a prediction about energy, and until something records
what actually happened against it, the prediction can never be wrong. §7.3 closes
that loop inside the artifact rather than in a report.

COMPANION CAPABILITY (§7.4). Green-metadata semantics register the way
wai.meta.hdr10plus and wai.meta.dovi do — name the descriptor class, sink supplies
the decoder, no carriage adopted. ISOBMFF/DASH carriage is somebody else's layer.

NETWORK-LAYER ALIGNMENT (energy-measurement §2.1). WAI's HwShunt/ModelBased/
Estimator is a private vocabulary, and coarser than the graded accuracy hierarchy
an adopted IETF GREEN draft defines for device energy. An implementation
reporting under both SHOULD derive rather than assign independently, MUST say
which identity it derived from, MUST NOT let a derivation improve the class, and
MUST declare the tolerance the class implies — a class without a tolerance is a
label. WAI does not adopt that model; the scopes differ (device vs per-object)
and conflating them would breach §4. What is required is that two vocabularies
for one property do not contradict each other.

Ingest queue item 3 and item 8 from the landscape survey (#80 §5.3 a–e).

Co-Authored-By: Claude Opus 5 <[email protected]>

54db9c6f4d

• View at this point in history

2026-09-10 12:01:53 -04:00