One of the hardest risks to control in a 360 reporting system is a conclusion built from real evidence.

Every score and comment may be genuine. Somewhere between the evidence and the finished prose, however, one observation becomes a behavioural pattern, qualifying evidence disappears, or uncertainty becomes confidence.

This is broader than literal hallucination, which invents information that is not present in the source. A report can instead omit qualifying evidence, flatten disagreement or turn one observed event into a stable behavioural claim.

When that happens, diagnosing the problem means working backwards through the scores, comments and reporting rules to reconstruct why the conclusion was written.

The cost reaches beyond review time. A material conclusion can change how a participant understands a strength or development need, what a coach prioritises, or how an organisation interprets the result.

Consider a representative hypothetical report. It is not a description of a client report.

The participant avoids difficult conversations and may need to become more direct when performance falls below expectations.

The sentence is plausible. A manager may have rated the participant lower on constructive challenge, and one comment may describe a delayed conversation with an underperforming team member. Read on its own, the conclusion appears restrained and useful.

Now add the evidence that the sentence leaves behind. Peers rated the participant highly on candour. Direct reports described clear conversations when expectations were missed. The manager's comment described one delayed conversation. On its own, it supports only a situational observation.

One event has become a behavioural tendency. One respondent perspective has become the dominant account. Conflicting evidence has disappeared as the prose became smoother.

I use the term Evidence Chain for a reporting structure that keeps this movement visible. Each material claim remains connected to the approved evidence offered in support, the interpretation rule applied, any unresolved conflict and the decision state that allowed the claim into the report.

The chain makes the reasoning available for inspection before polished language hides the joins. It cannot certify that a claim is true.

Conceptual Evidence Chain: each material conclusion retains its sources, interpretation rule, disagreement and decision state as new context changes the reasoning.

A source is not yet a reason

A source link tells a reviewer where information was found. How that information was used requires a further record.

The difficult-conversations paragraph contains at least three possible claims: a behavioural pattern, an explanation tied to performance conversations and a development implication. A link to the survey file leaves unclear which score or comment supports each part, or which rule permits the move from a single event to a recurring behaviour.

The gap between the source and the claim creates a design requirement: establish the evidence, interpretation boundary and review state of a material claim before turning it into polished prose.

For the hypothetical claim, the working record might look like this:

Claim field	Working record
Proposed claim	The participant avoids difficult conversations.
Supporting evidence	One lower manager rating on constructive challenge and one manager comment about a delayed conversation.
Conflicting evidence	Higher peer ratings on candour and direct-report comments describing clear performance conversations.
Interpretation question	Does the approved method permit one situation to support language about a stable behavioural tendency?
Current state	Unresolved until the conflict is addressed; otherwise narrow, qualify or exclude the claim.

This is the difference between a reference list and a chain. The chain records both the source and the reasoning required to use it.

The distinction matters when several nearby signals appear to agree. A low item score, a lower group average on a related construct and a critical comment may all concern directness. They are still different forms of evidence. Their combination supports a broader theme only when the reporting method explains how they may be combined and weighted.

Without that rule, a model can blend three real inputs into one unsupported conclusion. More citations would make the paragraph look well sourced without exposing the leap.

The difficult evidence is the evidence that disagrees

I have seen polished 360 narratives flatten meaningful differences between manager, peer, self and direct-report perspectives. Fluency can make the loss harder to notice because the final paragraph reads as though the evidence always pointed in one direction.

Disagreement is normal in 360-degree feedback. Respondent groups observe different settings and relationships. A gap between them may reflect context, noise, role expectations or a meaningful difference in behaviour. The consultancy's method determines when that gap should enter the narrative and how strongly it can be interpreted.

An Evidence Chain gives the disagreement somewhere to remain visible. In the hypothetical example, the peer and direct-report records stay connected to the proposed claim as conflicting evidence. The applicable method rule shows whether the difference may be interpreted, must be qualified or remains unresolved.

The resulting language might become:

One manager response suggests that directness may become harder in some performance conversations. This pattern is not reflected consistently across other respondent groups.

That is one possible resolution. Other readings may justify a request for context, a different explanation or no narrative claim at all. What matters is that the disagreement remains part of the drafting process.

Professional judgement also becomes easier to locate. If a practitioner decides that the manager evidence is meaningful despite the wider pattern, the chain can retain that decision and its rationale. The report can then present the judgement as a deliberate decision with a visible basis.

A claim needs a state before it needs polished prose

Return to the original evidence. It supports two narrow observations: one manager described a delayed performance conversation, and the participant received a lower manager rating on constructive challenge. The statement that the participant avoids difficult conversations adds a further proposition about stable behaviour.

That proposition needs support of its own. The chain should test whether each material part of the claim is covered and whether an approved rule permits the transformation from evidence to wording. A missing link leaves the claim unready, however careful the sentence sounds.

The response can remain bounded. The language may be narrowed to the observed situation. The claim may be marked as qualified while conflicting evidence is retained. Missing context may be requested. An unresolved claim may be excluded from generation altogether.

This makes hallucination prevention a property of the reporting method, beyond any instruction placed in a prompt. A model may still propose an unsupported inference. The workflow creates a recognisable state for that failure and a route that does not lead directly into the deliverable report.

The extra structure has a cost. For a small number of low-risk reports, an experienced practitioner may be able to reconstruct the reasoning from a source table and a careful reading. Claim-level traceability adds recording and maintenance work.

My practical boundary is materiality. A transition sentence can stand without a miniature audit file. A conclusion that changes how a participant is understood needs its reasoning recorded outside a model's generation step or one consultant's memory. That boundary determines where the extra structure earns its place—and where AI has useful work to do.

Use AI where evidence has to be synthesised

In the hypothetical report, the lower manager rating does not need AI to calculate or display it. Neither do respondent-group averages, bounded score labels or the established assessment presentation. Conventional automation can handle those parts consistently.

The difficult work begins when the comments and scores have to be read together. One manager described a delayed conversation. Peers and direct reports described candour. The combined evidence does not support the original claim that the participant avoids difficult conversations.

This is where AI can add value in Areas of Strength and Areas for Development: reading the full comment set alongside the assessment data to help form subject-specific insights. Its value is not another generated report section. It is helping to structure evidence that varies from one participant to the next.

The AI path still should not jump from raw comments to a polished paragraph. Each source comment first becomes a distinct, source-linked observation. Candidate insights are linked to the observations supporting them, while disagreements and isolated signals remain visible. By the time narrative drafting begins, the system already has a record of why the claim is qualified.

The value of that record becomes clearer at the next handoff, when someone else has to inspect the conclusion.

Traceability is how the reasoning travels

Suppose the qualified difficult-conversations paragraph now moves from the consultant who prepared it to a reviewer. The reviewer sees the finished wording, but the wording alone does not show why the manager evidence was retained, why the peer and direct-report evidence qualified it, or which interpretation rule allowed the claim to proceed.

Without the chain, the reviewer has to return to the scores, comments and reporting rules and reconstruct those decisions. With the chain, the claim arrives with its supporting sources, interpretation boundary, conflicts and decision state. The report changes hands, but the reasoning does not stay behind.

This is the connection to The Copilot Ceiling. Faster drafting does not increase reporting capacity if each handoff creates another reconstruction task. The Evidence Chain addresses one cause of that work by allowing the reasoning to travel with the claim. That is an architectural contribution, not evidence of increased capacity, reduced review time or improved report quality. Those outcomes still require measurement across the full path to delivery.

The immediate test is whether the reasoning survives a handoff. A harder test comes later, when new context changes what the same evidence can support.

The chain has to carry new context

A traceable first draft can still become an untraceable final report.

Return to the delayed-conversation example. After speaking with the participant or their manager, the consultant learns that the conversation happened in unusual circumstances. The original rating and comment still stand, but the new context changes what conclusion they can support.

That context belongs in the same reasoning record. In a complete Evidence Chain, it remains connected to the source evidence and identifies the insights and narrative that depend on the earlier reading.

This is where traceability contributes to scale. The professional context is captured once and can move through the report instead of being reconstructed from every score and comment or applied manually to each paragraph. Whether an implementation does this reliably still has to be tested.

The same structure gives a later read-only audit something specific to inspect: the evidence behind each conclusion, the rule applied, any unresolved disagreement and the effect of new context. Exact links can be checked deterministically; interpretation still calls for specialist judgement.

This does not make the Evidence Chain a complete assurance system. Independent assurance requires its own criteria, tests and failure path. The chain has a narrower job: keep a change in the reasoning from disappearing between the consulting conversation and the finished prose.

In the hypothetical example, the qualified difficult-conversations paragraph can no longer remain current once the unusual circumstances are attached. The chain returns the paragraph and its candidate insight to the reasoning path with the new context visible. It does not invent the replacement conclusion. It prevents the earlier conclusion from surviving unchanged after its basis has changed.

Preventing hallucinations in 360-degree feedback means keeping unsupported movement visible, including when the context changes after the first draft.

Traceability does not prove the conclusion. It makes the conclusion answerable to its evidence.