Investigation report templates
Investigation Report Examples: The Pre-Issue Review Record
Last updated 2026-07-30
A finished draft is the point at which the investigator stops being able to see it. The reasoning is sound, the file is closed, and what remains are defects invisible from the inside: the second standard of proof that crept into one finding, the corroboration obtained after the respondent's last interview, the summary sentence that asserts more than the body established. Every one of those is cheap to fix on the Friday before issue and expensive afterward, because after issue the only honest remedy is a dated addendum that announces the defect to everyone who received the report.
Every other document page in this library carries a filled-in example rather than a blank template, most of them written against one fictional matter so they can be read against each other: the full workplace investigation report, the compressed short investigation report, the credibility assessment that leaves one conflict deliberately unresolved, and the allegation matrix shown partway through the matter with two rows determined and the third closed out against an unavailable camera. Read them for the reasoning against a specific evidential record, not for the register. Borrowing a confident tone onto weaker evidence is how a good example produces a bad report.
This page is the other half. The sections below are a review record, not a report — a document the reviewer fills in and files, recording which checks were run, what each one found, what was done about it, and what the review did not cover. That last part matters: a reviewer who cannot say what they were and were not checking offers no protection later. Under Faragher and Ellerth the employer's affirmative defense turns on whether its response to harassment was reasonable, and the report is the artifact by which that response gets judged. A recorded review is evidence of care; an unrecorded one is a claim about it.
Jurisdiction
Written primarily against US federal standards — EEOC guidance, NLRB decisions and Title VII case law — and against general investigative practice. State law, sector rules (public employers, education, regulated industries) and collective agreements impose additional requirements this page does not cover. Where a page borrows a concept from another jurisdiction it says so explicitly. Outside the US, treat the structure as transferable and the legal references as not.
Investigation Report Examples: The Pre-Issue Review Record template
Free Word download, no email required. Or open it in Professional Drafter — it lands in your template library, ready to draft against.
When to use this document
- A draft report is complete and needs a structured review before it is signed and issued.
- The report will be read externally — by an arbitrator, an agency, or opposing counsel — and the cost of a defect has just risen.
- You supervise internal investigators and need one quality bar their drafts are reviewed against.
- A matter has been reassigned and the incoming investigator has to satisfy themselves about a draft they did not write.
- You are working from the library's filled-in examples and want the questions to put to your own draft once it reads like them.
Worked example
Fictional scenario, continuing the dispatch-team matter used across this library — matter RW-2026-038, complainant Nadia Okafor, respondent Sean Whelan, complaint of 29 May 2026. This is not another report. It is the review of one: R. Hall's draft was complete on 10 July 2026 and reviewed on 11 July by D. Whitfield, a Senior HR Business Partner with no reporting line into dispatch, three days before the report date of 14 July. Five defects. Four were fixed by editing. One was not fixable by editing at all.
Review record header
- Matter:
- RW-2026-038.
- Draft reviewed:
- v3, 10 July 2026.
- Reviewer:
- D. Whitfield, Senior HR Business Partner. No prior involvement in the matter, no reporting line to the dispatch team or to the investigator.
- Brief, agreed in writing on 8 July:
- completeness against the complaint, traceability of every finding to a source, internal consistency across the report and the draft outcome letters, and exposure created by wording. Not the merits — the reviewer does not second-guess a finding the evidence supports.
- Date of review:
- 11 July 2026. Three working days before issue.
The slack was deliberate. A review scheduled on the issue date can only produce edits, because anything requiring a person to be interviewed again is by then impossible. On this draft that margin was the difference between a defect and a defect nobody could fix.
Checks 1 and 2 — two standards, and three words for one outcome
The scope section stated the standard correctly: "I have applied the preponderance of the evidence: whether each allegation is more likely than not to have occurred."
The search for stricter language returned one hit, in the draft finding on Allegation 3: "There is no conclusive evidence either way, and I am therefore unable to prove this allegation." Sixteen words, two defects. "Conclusive" imports a standard the report expressly disclaimed. "Unable to prove" casts the investigator as a prosecutor carrying a burden, which is not what a workplace fact-finder does.
Replaced with: "The evidence does not permit a determination on the preponderance of the evidence. Allegation 3 is accordingly inconclusive. This is a statement about the evidence and not a finding that the incident did not occur."
The same pass caught vocabulary drift across documents. Allegation 3 was "inconclusive" in the findings section, "unproven" in the executive summary, and "not established" in the draft letter to Ms Okafor. Three words for one outcome, in the three places a party compares. Standardized to "inconclusive", defined once at the head of the findings section, and both letters redrafted from the final wording rather than from the version they were written against.
Check 3 — the counts matched and the sets did not
Complaint document of 29 May: three discrete assertions. Report: three findings. On a count, that reconciles. Item by item, it does not.
The complaint raised (a) repeated comments about her appearance, (b) the corridor incident in April, and (c) that after she mentioned the comments to the Shift Supervisor in March, Mr Whelan moved her onto a later shift for two weeks in April. The report's three allegations are the comments, the comments continuing after she asked him to stop, and the corridor incident. Allegation 2 arrived legitimately: it emerged in her interview of 9 June, the extension was requested on 15 June and agreed in writing by the Director of Human Resources on 16 June, and the plan amendment records all of it. Assertion (c) simply stopped existing. Two changes in opposite directions, netting to zero, and the arithmetic hid both.
The descoping was also wrong on the merits. It had been treated at the planning stage as a scheduling grievance. It is not: under *Burlington Northern & Santa Fe Railway Co. v. White*, 548 U.S. 53 (2006), a materially adverse action is one that "well might have dissuaded a reasonable worker from making or supporting a charge of discrimination", and a two-week schedule change imposed by the person complained about, following a report to a member of management, sits squarely inside that. The matter's own interim measures memo of 3 June had already identified Mr Whelan's control of shift allocation as the specific risk, and transferred that function to the Shift Supervisor with effect from 4 June. The risk was visible on 30 May, when the complaint was received. The allegation built on it was not.
Disposition: not a rewrite of this report. Assertion (c) was referred on 11 July to the Director of Human Resources as a potential retaliation allegation, and opened as a separate matter, RW-2026-046, on 13 July. Matters Outside Scope now records that it was raised in the complaint of 29 May, that it was not investigated in RW-2026-038, the date and reason it was referred, and where it went. Three allegations were investigated, and the report now says why it was three and not four.
A decision that is recorded is a judgment. The identical decision, unrecorded, is an omission.
Checks 4, 5, 6 and 11 — sourcing, migration, demeanor and one adverb
Two sentences in the findings carried no source. "The respondent was aware the comments were unwelcome" is an inference in the grammar of an observed fact, and the work is done anyway by Allegation 2, which establishes the request to stop and the comments that followed it. Deleted. "Witness B saw nothing relevant" is true but reads as an assumption; corrected to "Witness B was interviewed on 12 June 2026 and had no relevant knowledge of any allegation." Coverage is a proposition about thoroughness and needs a date like any other.
The evidence summary contained one migrated conclusion: "The respondent was dismissive when asked about the corridor incident." No note records dismissiveness. What the note records is that he answered in four words and did not elaborate, so that is what the evidence summary now says. The judgment went nowhere, because the credibility analysis never needed it.
The demeanor deletion test was run on both determinations. The draft credibility paragraph on Allegation 2 included: "The complainant presented as visibly upset and entirely genuine throughout." Strike it and the determination is untouched — it rests on two things, her text to Witness A of 8 April referring to having "said something to him yesterday", created before any complaint existed, and Witness A's independent recollection of being told at the time. The respondent's failure to deny that a request was made removes nothing and adds nothing. The sentence therefore supported no part of the finding while offering an obvious line of attack, that the outcome turned on how she came across. Deleted, and replaced with: "I have placed no weight on the manner in which either party gave their account. The determination rests on the contemporaneous message of 8 April and Witness A's independent recollection." On the corridor conflict the test was moot: nothing there rested on demeanor because nothing there rested on anything, which is why it is inconclusive.
The language pass removed one word — "understandably", from "understandably distressed". Distress is observable and can be sourced. "Understandably" is the investigator agreeing with a party in the section where the investigator is supposed to be describing.
Check 8 — the defect editing could not cure
Every adverse fact was listed against the date it was put to Mr Whelan in his interview of 19 June. One had no date against it.
Witness A, Priya Anand, was interviewed on 12 June and recalled Ms Okafor telling her in early April that she had "told him to knock it off". That recollection is one of the two supports for the finding on Allegation 2. The notes of 19 June show the allegation put to Mr Whelan in substance — that a request to stop had been made — but do not show him being told that a colleague recalled being told about it at the time, or being given the chance to answer that specific corroboration. The interview order was correct when it was set; the corroboration was simply weighed after he had gone.
This cannot be repaired in the draft. A finding resting on evidence the respondent has never seen does not survive review, and softening it to fit the gap produces a hedged conclusion nobody can act on.
A supplementary interview was held on 12 July, twenty-two minutes, with the corroboration put to him in terms. He maintained that he did not recall a request being made and did not dispute that one might have been. The finding on Allegation 2 is unchanged. What changed is that it now rests on evidence he had the opportunity to answer, and the methodology section records the supplementary interview and the reason for it.
The report date remains 14 July. Finding this on 11 July cost one afternoon. Finding it at an arbitration would have cost the finding.
Disposition and sign-off
Defects and disposition:
1. Two standards of proof in one document — edited, 11 July.
2. Three words for one outcome across report, summary and letters — standardized, letters redrafted from final wording, 11 July.
3. Assertion in the complaint of 29 May neither investigated nor recorded — referred to the Director of Human Resources 11 July; opened as RW-2026-046 on 13 July; Matters Outside Scope updated.
4. Unsourced findings, one migrated evaluation, one demeanor sentence, one adverb — edited, 11 July.
5. Corroboration relied on but never put to the respondent — supplementary interview held 12 July; methodology updated.
One further sentence was deleted rather than edited, from the executive summary read cold at the end: "The respondent's conduct made the complainant's working environment intolerable." Nothing in the findings establishes it, no allegation was framed in those terms, and "intolerable" reaches toward a legal characterization that was expressly outside the mandate. It arrived the way summary sentences do, written last from memory of the file rather than from the findings.
Materials reviewed: draft report v3 and its appendix index; the allegation matrix; both draft outcome letters. Not reviewed: the interview notes in full, beyond the notes of 19 June checked against defect 5, and no witness was re-contacted by the reviewer.
Five defects, seventy minutes, one supplementary interview. Four of the five were in language rather than in fact-finding: the investigation was sound and the draft misrepresented it in small, quotable ways. A report that is right and a report that reads as right are different achievements, and only the second one is ever seen.
D. Whitfield, Senior HR Business Partner — 12 July 2026. This review is of draft v3 dated 10 July 2026 and does not extend to any later version.
Section-by-section guide
What each section is for, what a good one contains, and where they typically go wrong.
1.Review record header
The record is a file document in its own right, so it starts with identifiers. Name the reviewer, their role, and their distance from the matter — a reviewer inside the respondent's reporting line has a problem no amount of care fixes. Record the draft version reviewed and its date, because a review is only ever a review of the version in front of you, and state the brief in the terms it was given. Define the brief before the review starts rather than describing it afterward, when it will be described as whatever the review happened to catch.
2.Check 1 — Standard of proof stated once, and not undercut elsewhere
Presence is the easy half. The harder half is whether a stricter standard crept in somewhere else, because a reader who finds two standards will hold you to the stricter one. Search the draft for "conclusive", "clear evidence", "beyond doubt", "insufficient proof" and "unable to prove", then test each hit against the standard claimed in the scope section. "Unable to prove" also puts the investigator in the position of a prosecutor carrying a burden, which is not the position the preponderance of the evidence describes.
3.Check 2 — Outcome vocabulary defined once and used consistently
"Not substantiated" means the evidence does not support the allegation. "Inconclusive" means the evidence does not permit a determination either way. Parties rely on the difference and will argue about it, so the risk is not the definitions — it is drift. A finding that is inconclusive in the findings section and "unproven" in the executive summary invites the argument that two different things were meant. Check the outcome words across the report, the matrix and both draft outcome letters in one pass, because the letters are what the parties actually keep.
4.Check 3 — Allegations reconciled item by item against the complaint
Counting is not reconciling. Line the assertions in the original complaint against the rows of the matrix and the findings in the report, and match them one to one by content, because a dropped assertion and an added allegation cancel out in a count and equal numbers then read as coverage. Expect the sets to differ legitimately: an allegation added by a dated, authorized scope extension belongs in the report and not in the complaint. What is never legitimate is a difference with no record. An allegation raised in writing and absent from the report, with nothing explaining its absence, is the most visible defect a reader can find.
5.Check 4 — Every factual proposition traceable to a source
Read the findings section alone, sentence by sentence, and ask of each one where it was learned. If the answer is "from the file generally", it needs a reference: an interview date, a witness name, an appendix number, an evidence log item. A fact with no source reference is an assertion, and it is the first thing a careful reader tests. Watch particularly for propositions about coverage — that a witness had nothing relevant to add is a finding about the thoroughness of the investigation, and it needs an interview date behind it like anything else.
6.Check 5 — Evidence and analysis genuinely separated
Run the deletion test in both directions. Delete every analytical section: what is left should still be a coherent account of what each person said and each document showed. Then read the evidence summary for "clearly", "obviously", "evasive", "unconvincing" and "of course". Those are conclusions that migrated upstream, and they contaminate the one section a reviewer needs clean. Most of them are also unsourceable, which is the practical tell — a note can record that a witness answered in four words, but no note records that he was dismissive.
7.Check 6 — Demeanor deletion test on each credibility determination
Do not look for a disclaimer; look for load. Strike the demeanor sentence from each determination and read what remains. If the determination still stands on plausibility, corroboration, consistency and motive, the sentence was decorative and should go anyway, because it adds an attack surface and no support. If the determination collapses, it was resting on demeanor whatever the surrounding caveats said, and it has to be rebuilt on other evidence or the conflict recorded as unresolved.
8.Check 7 — Scope honored outward and inward, with every change dated
Outward: does the report determine anything nobody asked for — a legal characterization, a disciplinary recommendation, a conclusion about a person who was never a respondent? Inward: is anything inside the mandate left uninvestigated with no explanation, including evidence sought and never obtained? Where evidence was requested and refused, the entry needs what was asked, from whom, on what date, the answer, and what the absence does to the analysis. Then check the changes: a scope extension needs a date, a requester and written authorization from the commissioning party. An extension that happened by drift and appears nowhere is worse than no extension at all.
9.Check 8 — Every adverse fact put to the person it counts against
The most expensive defect on this list, because editing cannot cure it. A finding resting on evidence the respondent was never shown does not survive challenge, and the remedy is a supplementary interview. List each fact adverse to the respondent and find the interview date at which it was put. The recurring failure is corroboration obtained after the respondent's interview: the allegation was put in substance, the witness who later corroborated it was not, and the interview order that made sense in June leaves a gap in July.
10.Check 9 — Policy quoted in the version in force, and applied after the facts
Two failures that travel together. Version: quote the provision as it stood when the conduct occurred, with its effective date, and cite the evidence the respondent was on notice of it — an acknowledgment, a training record, a handbook receipt. Conduct measured against a policy the employee never received, or one postdating the conduct, is not repaired by careful fact-finding. Order: the policy analysis belongs after the findings of fact. Where it comes first, the report reads as though the conclusion selected the facts.
11.Check 10 — Executive summary reconciled to the body
Read the summary alone, cold, after every other check. Mark each claim in it and locate that claim in the body. It is the section most readers read and the section written last, usually at the end of a long day, which is why reasoning appears there that was never tested anywhere else. Check the reverse too: a summary that hedges more than the findings do — "appears to have", "may have" — quietly withdraws findings that were actually made to the standard.
12.Check 11 — Adjectives applied to people, and anything you would not read aloud
Strike descriptive adjectives applied to people rather than to conduct. "Aggressive", "defensive", "emotional" and "difficult" characterize a person and cannot be sourced; "raised his voice across the bay" describes conduct and can. Then read the draft as though the person it treats least favorably is sitting opposite. Sarcasm, exasperation and small editorial asides survive into final drafts more often than investigators expect, and they get quoted back with far more effect than any finding does.
13.Check 12 — Document set agreement and file hygiene
The report is one artifact in a set and the set has to agree. Allegation wording must match across the complaint, the matrix, the report and both outcome letters; dates, name spellings and the matter reference must match everywhere; the appendix index must match the numbers cited in the body, because a late-inserted appendix renumbers everything after it and no cross-reference looks wrong when it breaks. Then control the file itself. Accepted revisions, resolved comments and the author field travel inside a .docx and are recoverable after it is sent.
14.Disposition and sign-off
Close the record with what was found, what was done, and what was not looked at. Each defect gets a disposition — edited, rebuilt, referred, or accepted with a reason — and anything requiring a step outside the document, such as a supplementary interview, gets a date. Then state the limits plainly: which materials the reviewer read, which they did not, and that the review is of a named draft version. Write it as a professional quality record, because it is discoverable and will be read by people looking for either diligence or an admission.
Common mistakes
The failure modes that show up most often when this document is reviewed later.
Reviewing your own draft in reading order
Read start to finish, you re-encounter your own reasoning in the sequence that already persuaded you, and it persuades you again. Review by check rather than by section: one pass for sourcing across every finding, a separate pass for demeanor across every determination, the summary read cold and last.
Letting the reviewer return comments as tracked changes inside the report file
Their wording ends up in the investigator's findings and nobody can afterward say which sentences the fact-finder wrote. It also puts a second author's name in the document properties of a report that attests to one person's analysis. Comments belong in a separate record; the investigator makes every change.
Running the review before the appendices are assembled
Cross-references cannot be checked against an index that does not exist yet, and the check is almost never rerun once the appendices land. The file then holds a completed review record against a document that changed after it was signed.
Scheduling the review for the day the report is due
A review with no slack silently becomes a proofread. The two defects worth finding — an adverse fact never put to the respondent, and an allegation raised and never addressed — both need days rather than edits, so a same-day review is structurally incapable of reporting them.
Drafting the outcome letters from the pre-review findings and not redrafting them after
The letters go out carrying wording the review removed. The party holds one version and the file holds another, and the discrepancy surfaces at the point where a party's representative reads both side by side.
Tidying the interview notes so they match the corrected report
Notes are contemporaneous evidence and their value is that they were made at the time. Editing them after a review destroys that, and the file metadata records the date of the edit, converting a wording fix into an allegation of fabrication.
Questions
- Who should review a draft investigation report before it is issued?
- Someone who was not the fact-finder and has no reporting line to the parties — in practice an experienced investigator, an employee relations lead, or counsel. Agree the brief in writing before they start: completeness, traceability, internal consistency, legal exposure. Then record who reviewed it, when, and what the review did not cover. A reviewer who cannot say what they were and were not checking is not much protection later.
- Can in-house or outside counsel change an investigator's findings?
- They can and should flag findings that are unsupported, ambiguous, outside scope, or worded in a way that creates avoidable exposure. Rewriting one is a different act: it makes the lawyer a fact-finder, undercuts the investigator's independence, and complicates any later privilege position. Where a review changes a finding, the change comes from the investigator, rests on identified evidence, and is recorded — not applied silently to the document.
- The review found a finding that was never put to the respondent and the report is due tomorrow. Do we issue and supplement later?
- No. You have two defensible options: move the issue date and hold the supplementary interview, or issue without that finding and address the allegation once the respondent has answered the evidence. Issuing a finding on evidence the respondent has not seen, intending to fix it afterward, produces a report whose central finding is already vulnerable — and the supplementary interview then looks like it was arranged to shore up a conclusion already published.
- How do I review a report properly without re-reading the whole evidence file?
- Most of these checks are internal and run against the report alone: sourcing, allegation reconciliation, outcome vocabulary, evidence-versus-analysis, the summary read cold. Three need the file — whether every adverse fact was put to the respondent, whether the policy version quoted was in force at the time, and whether the recorded gaps match what was actually requested. Run the internal passes first. They produce the short, specific list of things worth pulling the file for.
- Is the review record itself disclosable?
- Assume yes. It is a document about the investigation, held on the file, and it will be read by someone looking either for evidence of care or for an admission. That is a reason to write it well rather than to skip it: check, finding, action, date, in neutral terms. What creates exposure is editorializing about the investigator or speculating about outcomes, neither of which the record needs.
- Can I use a state or federal agency's published sample report as our template?
- Use it to see how a finished report reads, not as a structure to adopt wholesale. Oregon's Department of Administrative Services publishes one of the more accessible examples, written against that employer's policies, its collective agreements and its procedural rules — so its section list encodes obligations that may not be yours and omits ones that are. Take the reasoning and the level of specificity; rebuild the structure against your own policy.
- Do we keep the earlier drafts of the report?
- Follow your retention schedule and apply it uniformly. Drafts are ordinarily discoverable, and deleting them selectively — or at all, once litigation is reasonably anticipated — is a far larger problem than any inconsistency the drafts contain. Settle the retention position in policy before an investigation starts, rather than deciding it about a particular file.
The blank template
Exactly what is in the Word download, so you can see the structure before deciding whether you want it.
Review record header
Check 1 — Standard of proof stated once, and not undercut elsewhere
Check 2 — Outcome vocabulary defined once and used consistently
- [Terms defined at: ...]
- [Variant phrasings found and corrected, with the document each appeared in]
- ["No evidence of wrongdoing" removed — it conflates "we found nothing" with "it did not happen"]
Check 3 — Allegations reconciled item by item against the complaint
Check 4 — Every factual proposition traceable to a source
- [Unsourced sentence — source added, or removed]
- [Facts resting on the investigator's recollection rather than on a note]
- [Inferences stated in the grammar of observed fact]
Check 5 — Evidence and analysis genuinely separated
- [Evaluative language found in the evidence summary]
- [Evidence appearing for the first time inside the analysis, never described to the reader]
- [Findings that restate evidence rather than resolving it]
Check 6 — Demeanor deletion test on each credibility determination
Check 7 — Scope honored outward and inward, with every change dated
Check 8 — Every adverse fact put to the person it counts against
Check 9 — Policy quoted in the version in force, and applied after the facts
Check 10 — Executive summary reconciled to the body
- [Assertions in the summary with no counterpart in the body]
- [Outcome wording differing from the findings section]
- [Hedging weaker or stronger than the finding it summarizes]
Check 11 — Adjectives applied to people, and anything you would not read aloud
- [Characterizing adjectives replaced with described conduct]
- [Editorial or exasperated phrasing removed]
- [Names, pronouns and titles used consistently throughout]
Check 12 — Document set agreement and file hygiene
Disposition and sign-off
Download this template
Free Word download, no email required. Or open it in Professional Drafter — it lands in your template library, ready to draft against.
Sources
- Faragher v. City of Boca Raton, 524 U.S. 775 (1998) — Cornell LII (employer's affirmative defense turns on reasonable care to prevent and correct)
- Burlington Industries, Inc. v. Ellerth, 524 U.S. 742 (1998) — Cornell LII (companion case establishing the same two-part defense)
- Burlington Northern & Santa Fe Railway Co. v. White, 548 U.S. 53 (2006) — Cornell LII (materially adverse action standard)
- EEOC — Enforcement Guidance: Vicarious Liability for Unlawful Harassment by Supervisors (1999, still operative — adequacy of the employer's investigation)
- Cornell LII — Preponderance of the evidence (definition of the standard)
- Oregon Department of Administrative Services — Sample Investigation Report (a publicly published finished report)
Related templates
Workplace Investigation Report Template & Example
A complete workplace investigation report template with a filled-in example report, plus a section-by-section guide covering scope, evidence, credibility, standard of proof and findings.
Short Investigation Report Template & Sample
A short-form investigation summary report template for straightforward matters — when a full report is more process than the complaint warrants, without dropping what makes a finding reviewable.
Credibility Assessment Template & Example (Workplace Investigation)
A structured credibility assessment template using objective factors — plausibility, corroboration, consistency and motive — with a filled-in worked example showing the reasoning written out.
Allegation Matrix Template & Example
An allegation matrix template that tracks every allegation to its evidence and its finding, so nothing is left unaddressed in the final report.
This page is a documentation aid, not legal advice. Workplace investigation requirements vary by jurisdiction, sector and policy, and a template cannot tell you what your matter requires. The worked example describes a fictional scenario.