Community access / Message evidence
Duplicate events and original instruction identifiers
Count repeated delivery separately from original calls so noisy rooms do not inflate coverage or sample size.
Define original identity first
The original population needs a declared identity rule, not a count of every visible notification. A repeated delivery can concern the same instruction; an amendment can change that instruction without being a new original. A source that genuinely issues a new instruction needs a record supporting that distinction. Save object relationships before aggregating counts. The rule should be stable across favourable and unfavourable items. A screenshot collage or message-volume total cannot independently establish how many unique authored calls existed under the observation window.
An event log can be larger than its population
Keep delivery attempts, retries and repeated messages as events linked to the original. The event count can legitimately exceed the number of originals, but it should not inflate the received numerator or published denominator. A duplicate is not necessarily suspicious; it may be an operational repetition. The consumer finding concerns whether the record reconciles, not an invented intent judgment. This publication's synthetic packet has eight originals, six received originals and two duplicate events. Those counts describe the example only and supply no measured vendor activity.
Do not infer identity from text alone
Two messages can look alike while referring to different source objects, and one object can change wording through revisions. Preserve the actual supported relationship where available. If identity cannot be resolved, keep that uncertainty instead of deduplicating in whichever direction makes a metric look better. A reader-observed excerpt and a relayed original may need separate provenance even when their content overlaps. The site does not scrape room histories or attempt to reconstruct private IDs from anonymous text. Its invented O-labels are teaching labels, not real Discord identifiers.
Coverage is still a limited result
After reconciliation, the count can support a delivery finding under the stated observer and period. It does not establish a win rate, member-wide notifications or an executed account result. A large duplicate log might increase attention cost without changing informational coverage. If the task requires linked context, apply the separate missing-context rule within the timely uncancelled subset. Keep original, event, version and context definitions in the export so the reader can reproduce the arithmetic without mistaking an operational count for financial validation.
Completed fictional member-view case
Fictional case: O1 is received once and then repeated twice. The packet has one received O1 and two duplicate events, not three independent calls. O7/O8 remain unreceived, so the repeated O1 does not repair their absence. No real server or message IDs are used. The completed reconciliation preserves every event while keeping eight originals and six unique received. Its conclusion concerns identity and denominator integrity, not a trading edge, exact server-wide reliability or a claim that retries are inherently deceptive.
| Object | Record | Bounded finding |
|---|---|---|
| O1 first receipt | One unique original | Included once |
| O1 repeats | Two duplicate events | Separate log |
| O7 / O8 | No receipt in case | Not repaired by repeats |

Primary references and scope
Public documentation read 13 October 2026. It supports the stated mechanism, not a bought room, tested member device or the invented case.