An Introduction to Performance Analysis of Sport

1. 1Define each athlete's role and the session's purpose before it starts. The source states that athletes must know their role and understand the purpose and intent of the session. Send a one-line purpose statement with the session invite.
2. 2Ask for the open mindset explicitly. The source names an open mindset, honesty and integrity (Groom et al., 2011) as the qualities athletes bring. Frame the session as a safe space for updating knowledge and behaviours (Kojman, 2022), so athletes admit errors without penalty.
3. 3Match the session format to the athletes. Choose from solo coach delivery, review in units, or a complete player-led session (Middlemas et al., 2018). Place it on the Mosston and Ashworth spectrum, from command (style A) to inclusion (style E), according to the coach's style, the athletes' learning preferences, and their skill level, gender and age.
4. 4Use a hot debrief only for objective data. Emotions are high regardless of the result, so the coach delivers pre-agreed KPI results from the analyst and sets subjective feelings aside. Save reflective, critical discussion for the later video session.
5. 5Design the presentation for engagement. Use video clips, annotations, slide designs, in-session tasks and take-home tasks. Play clips full screen, keep to the scheduled time, and keep the production organised, because messy presentations lose athletes.
6. 6Share clips through an online hosting platform. Players open them on personal devices at times that suit their routines (Butterworth, 2023a). In the netball example, this moves viewing out of squad time and frees more time for on-court drills.
7. 7Build in reflection tasks. Use telestration such as spotlights, magnifier and region graphics to prompt athletes to consider a decision moment. Separate the decision made from the execution of the skill (Lorains et al., 2013) so the athlete reflects on each separately.
8. 8Protect recovery and focus between sessions. When debrief and briefing meetings run back to back, schedule a short break (a drink or a pre-organised activity) so athletes can switch off and refocus. On congested match schedules, give athletes the next day's timetable before the current day ends.

Butterworth's (2023a) netball work builds four highlights videos from the match analysis: good attacking play, attacking areas to improve, good defensive play, and defensive areas to improve. The two improvement videos typically hold six clips each. The videos are hosted on internet-based platforms and shared with the squad, so players watch on personal devices at times and places that fit their routines. Watching alone moves viewing out of squad time and leaves more of the shared time for on-court drills. Schedule shapes the engagement window. At the 2023 Netball World Cup some teams played seven matches in nine days, which limits time for debriefs and briefings. Players also need ice baths, food and sleep, so squad management sets each day's timetable and hands it out before the day ends. In British National Superleague netball there is typically a week between matches, which leaves room for fuller reflection.

• If emotions are high immediately after the performance, run a hot debrief with pre-agreed KPI results only, and delay critical reflection to a later session.
• If matches are on consecutive days, shift video viewing to online platforms, keep sessions short, and publish the next day's timetable before the current day ends.
• If debrief and briefing sessions run back to back, insert a short break so athletes can switch off and refocus.
• If athletes are passive in a coach-led (command) format, move toward a more player-led style (review in units or player-led), matched to their skill level, age and learning preferences.
• If the focus is a decision moment, use telestration to show the options and separate the decision made from its execution before asking the athlete to reflect.

• Athletes do not know their role or the session purpose. The tell: they sit through clips without contributing and ask what the session is for.
• Video is misused, which damages engagement and development. The tell: athletes become defensive or disengage after sessions.
• Messy, unorganised or overlong presentations. The tell: attention drifts, clips are not played full screen, and the session overruns its scheduled time.
• Subjective emotion is allowed to cloud the hot debrief. The tell: discussion centres on blame or feelings rather than the agreed KPIs.
• Online clips are shared with no task attached. The tell: platform views are low and athletes arrive at drills unable to discuss the clips.

• ▸Session type: hot debrief, team debrief, or online clip review, to match purpose to setting.
• ▸Purpose and intent statement: one sentence the athletes receive before the session.
• ▸Athlete roles: what each athlete is expected to contribute.
• ▸Teaching style: command, review in units, player-led, or another point on the spectrum, with the reason.
• ▸Open-mindset brief: how you will signal a safe space and invite honesty.
• ▸Content set: KPIs or clips (for example, six improvement clips per theme), with the number and source of each.
• ▸Platform and access: where clips are hosted and when athletes view them.
• ▸Reflection task: the in-session or take-home question, separating decision made from execution.
• ▸Schedule and break: session length, break between debrief and briefing, and when the next day's timetable is issued.

1. 1Start from the premise that recall is unreliable, and test it before you trust it. Before the match, agree a definition of 'critical incident' with the coach (for example set pieces, goals, turnovers). After the match, have the coach list the incidents they remember without any record in front of them.
2. 2Build the reference record at the same time. Use video, match statistics or both, because the source names both as the accurate record. Code the same critical incidents from it so you have something to compare against.
3. 3Score recall accuracy: the share of critical incidents the coach recalled accurately, out of those in the record. Compare it with the published benchmarks: 42% for novice coaches (Franks and Miller, 1986) and 59% for qualified soccer coaches (Laird and Waters, 2008). Expect the score to vary by event type, since set pieces were recalled better than other events.
4. 4Diagnose why recall fell short using the sources' causes: bias and favouritism, fatigue, attention drawn elsewhere (a coach looking down to write in a notebook can miss a key moment), and emotional involvement. Note which cause applies, because each calls for a different fix.
5. 5Close the delay between match and feedback. Jones et al. (2004) suggest recall accuracy decreases over time and that delayed feedback can be inaccurate. Aim to capture the record live or post-match and run the cycle promptly: game, select events, edit tape, analyse, prepare the presentation, present to coach, modify (Franks, 1997).
6. 6Select events from the record, not from memory. Cut the video clips and statistics for the incidents that matter, so the feedback session works from what happened and not from what was remembered.
7. 7Use telestration to build tactical understanding within those clips. Direct attention with spotlights, player tags and numbers, directional arrows, region graphics, linked cursors and a magnifier that greys out the rest of the frame. Each graphic should point at one tactical idea, such as space, options or defensive structure.
8. 8Re-measure. Repeat the recall test in later matches and track whether accuracy moves relative to the 42% and 59% benchmarks.

The source gives a real case of emotion distorting recall. The author watched a professional boxing match live at the venue and felt his supported boxer was losing until he won by stoppage. He later watched the full video, knowing the result, and felt the boxer had led on points from the end of the first round. The same contest produced two different judgements, one from emotional involvement and one from the record. Franks and Miller (1986) found that novice coaches accurately recalled only 42% of critical incidents, and Laird and Waters (2008) found 59% for qualified soccer coaches. So even the better figure leaves about four in ten incidents inaccurate or missing. In practice you code the incidents, ask the coach to recall them, and compare. Then you use the recorded clips for feedback. In the rugby sequence the source describes (Figure 7.7), the fly half is spotlighted and the four attacking options are marked with arrows and numbers. A magnifier shows the winger's raised hand while the rest of the image is greyed out. A region graphic marks the space behind the defensive line, and a 3D arrow shows the kick option and trajectory. The coach sees the detail they would otherwise have missed.

• If a coach is about to make a decision from memory of the match alone, then check it against the video or statistics first, because even qualified coaches recalled only 59% of critical incidents.
• If the critical incident is a set piece, then expect better recall than for open play, and spend your verification effort on the other event types.
• If feedback cannot be given soon after the match, then rely on the recorded data and not on recollection, since recall accuracy decreases over time.
• If the key action is subtle (a raised hand, a gap in the defensive line), then telestrate it with a magnifier, region or arrow so the viewer cannot miss it.
• If the coach or players are emotionally invested in the result, then treat their live impressions as suspect and review the full record, as the boxing example shows.

• Treating the coach's memory as the data. The tell: decisions cite incidents that the video does not show, or the coach's account of who was winning changes after viewing the tape.
• Delaying feedback until the following week. The tell: details are vague and the coach disputes the video because they remember it differently.
• Letting the coach take notes during play at the cost of watching. The tell: key events occur while the coach is looking down at the notebook.
• Overloading clips with graphics. The tell: players cannot say what the single tactical point of the clip was; use only the tools needed to direct attention to one idea.
• Assuming the record fixes bias. The tell: only favourable incidents are selected for the tape, so the review repeats the coach's favouritism.

• ▸Match / opponent / date: identify the contest being audited.
• ▸Critical incident definition: list the event types counted (for example set pieces, goals, turnovers).
• ▸Coach's unaided recall list: record the incidents recalled before seeing any data.
• ▸Record source: note whether video, match statistics or both are used as the reference.
• ▸Incidents in record vs. accurately recalled: enter both counts and compute the recall percentage.
• ▸Benchmark comparison: compare your percentage with 42% (Franks and Miller, 1986) and 59% (Laird and Waters, 2008).
• ▸Likely distortion causes: tick bias/favouritism, fatigue, distraction, emotion, delay.
• ▸Feedback delay (hours/days): record the gap between the match and the feedback session.
• ▸Telestration plan per clip: name the one tactical point and the tools used (spotlight, arrows, region, magnifier, linked cursor).

1. 1Define the decision moment before you code anything. Name the trigger (for example, a possession starting, a counterattack beginning, a serve) and the options a player could choose at that point. The source treats analysis as information that supports coaching decisions, so you need an explicit unit of 'a decision' to count.
2. 2Record from video or match data, never from memory. The source cites eyewitness research and Jones et al. (2004): recall is limited by bias, emotion and fatigue, and its accuracy falls as time passes. A coach's feeling that 'we kept choosing badly' is not evidence until it is tagged.
3. 3Tag the decision and the execution as two separate descriptors on the same event. Give the decision a descriptor such as option chosen, and give the execution its own success value. Cross-tabulate them in the matrix feature of your video analysis package, as the source describes. This gives you four cells: good decision/good execution, good decision/poor execution, poor decision/good execution, poor decision/poor execution.
4. 4Add a pressure descriptor and an options-available descriptor to each event, using categories you define in advance. The definition of decision quality depends on what was available and under what pressure, so a decision cannot be judged without them. The source gives no fixed categories, so write and fix your own before tagging.
5. 5Test reliability before you trust the numbers. The source stresses trustworthy, reliable processes and testing reliability. Decision quality is a judgment call, so have a second coder tag the same sequences and compare. Tighten the category definitions wherever you disagree.
6. 6Export the event list or matrix to a spreadsheet and compute frequencies and percentage rates. Use one template across matches, and compare against norms for the level of competition, as the source describes. Use percentages of decisions of each type, such as possessions leading to scoring opportunities, not raw counts.
7. 7Choose the video sequences for feedback by expert judgment, not from the statistics alone. The source says this is a separate process from the quantitative analysis. Pick clips that show the pattern the numbers flagged. Use telestration (cursors, directional arrows, linked cursors, region tools) to show the options available at the decision frame.
8. 8Deliver feedback quickly. Jones et al. (2004) suggest delays make feedback less accurate. Use the source's concurrent or live-data options where the set-up allows, and use coaching management systems so players review clips before they meet at training.

The source's telestration figure (7.5) shows a soccer counterattack. Davies initiates the counterattack and a teammate moves towards the centre of the pitch. Davies's forward run is tracked with a cursor trail, the wingers' positions are highlighted, and the attacking player and a defender are connected with a linked cursor. Potential player movements are then drawn as directional arrows. Used for decision review, this sequence works as follows. The analyst tags the event 'counterattack initiated' and records the option chosen (for example, run forward rather than pass). The analyst also records the pressure (a defender linked to the attacker) and the options available (the winger positions and arrows). A separate descriptor records whether the execution succeeded. In the review clip, the coach freezes the frame at the decision point and shows the arrows for the alternatives. The discussion is then about choice, not about whether the pass reached its target. The source does not give the outcome of this sequence, so treat it as a template for how to build the review, not as a finding.

• If a decision led to a poor outcome, check the execution descriptor before you criticise the choice. If the option was sound and the execution failed, coach technique. If the option was poor, coach recognition and choice.
• If two coders disagree on a decision category, redefine the category before you collect more data. Do not average the disagreement away.
• If a percentage rate looks high or low, compare it with norms for the same level of competition before you act on it.
• If the numbers flag a decision pattern, choose video clips that show the pattern and annotate the options available at the decision frame. Do not show the statistics alone.
• If feedback will be delayed, anchor it to a recorded clip, not to anyone's recollection, because recall accuracy decreases over time.

• Judging the decision by its outcome. The tell: every failed pass or shot is logged as a bad decision, and the execution descriptor does not exist or is never cross-tabulated.
• Relying on the coach's memory of the match. The tell: feedback cites incidents without timestamps, and different staff recall different events.
• Coding decisions without recording options available or pressure. The tell: identical choices are rated differently by the same coder across clips, and you cannot explain why.
• Skipping the reliability check. The tell: a second coder produces noticeably different decision counts on the same footage.
• Letting the statistics replace clip selection. The tell: players receive tables of percentages but never see the frame where the choice was made.

• ▸Event/trigger: describe the moment where a choice is required (for example, counterattack initiated).
• ▸Timestamp/clip ID: link the event to the video so it can be reviewed later.
• ▸Options available: list the realistic alternatives at that frame.
• ▸Pressure level: record your pre-defined category (for example, defender linked to the attacker).
• ▸Option chosen: record which alternative the player selected.
• ▸Decision rating: appropriate or inappropriate, judged against the options and pressure, not the result.
• ▸Execution result: successful or unsuccessful, recorded separately from the decision rating.
• ▸Second coder rating: record an independent decision rating to check reliability.
• ▸Feedback clip note: record what to freeze on and which telestration tools to use.

1. 1State the decision the analysis serves before you collect anything. The passages define performance analysis as a support activity within coaching: it produces information so coaches and players make more informed preparation decisions. Write one sentence per cycle, e.g. 'This data decides what we train on Tuesday.' If you cannot name the decision, do not collect the data.
2. 2Choose the unit of analysis: individual, team, opponent, referee, or training versus match. The passages state all of these can be analysed. Pick one per cycle so your outputs do not blur together.
3. 3Build an input system that matches the question. The book lists sequential systems (tennis serve and rally outcome, boxing punches, cricket bowling), frequency tables (netball), scatter diagrams (soccer shot locations), and combined sequential-plus-scatter input (doubles squash). Use sequential for order and outcome, a frequency table for counts, and a scatter diagram for location.
4. 4Validate the data before you trust it. Set up data validation lists so every variable option is listed under its column heading, which prevents free-text entries. Then test reliability, which the book names as a core part of building a system, before any result goes to a coach.
5. 5Process the data into KPIs. Use a pivot table with multiple variables selected to generate detailed KPI results. Report percentages as well as totals, for example the share of possessions that lead to scoring opportunities, as in the Euro 2022 comparison, so teams with different possession counts compare fairly.
6. 6Link the numbers to video. Tag events in a coding tool so every KPI result can be replayed as a clip. The passages cite coach-behaviour tagging that lets coaches view their own behaviour interactively and reflect, so the same approach applies to players.
7. 7Deliver the feedback as a service, not a verdict. Frame it as management information for coaches and players. Present areas of good performance alongside areas needing attention in training, and use telestration on the video to direct attention to one point per clip.
8. 8Close the loop. Turn each finding into a training or tactical action, then re-analyse the next match or session with the same system and definitions. Improvement in performance or development is judged only by comparing like with like.

The passages give the rationale for this construct. Franks and Miller (1986) found novice coaches accurately recalled only 42% of critical incidents from matches. Laird and Waters (2008) found qualified soccer coaches recalled 59%. Even the better figure leaves about four in ten critical incidents wrong or missing. Recall also varies by event type: set pieces were recalled more accurately than other events. Bias, fatigue, distraction (even looking down to write in a notebook) and emotional involvement all distort the coach's picture. The first edition's author felt his boxer was losing a fight until a stoppage win; on video, knowing the result, he judged it differently. Apply this to a netball or soccer coach who 'feels' the team struggles to create chances. You code possessions with a frequency table or sequential system, validate the entries, and produce the percentage of possessions ending in a scoring opportunity. The result replaces the impression with a figure you can clip on video and set against a training target. Next match, you recode with the same definitions and check whether the figure moved.

• If a coach's claim rests on memory of the match, then verify it with coded data and video before changing training. The recall studies show 42% to 59% accuracy.
• If teams or matches have different numbers of possessions or events, then report percentages (e.g. possessions leading to scoring opportunities) rather than raw counts.
• If your question concerns order and outcome (serve then rally result), then use a sequential system. If it concerns where events happen (shot locations), then use a scatter diagram. If it concerns counts only, then use a frequency table.
• If data entry is done by hand or by several people, then lock variable options with validation lists and test reliability before reporting.
• If the audience is players or coaches who may feel threatened, then present the analysis as an information service for their decisions, with good performance shown beside areas for attention.

• Collecting everything, with no stated decision behind it. The tell: large datasets and no change in a training session.
• Trusting unvalidated data. The tell: inconsistent labels for the same variable, or two analysts getting different counts for the same clip.
• Giving feedback as a judgement of players or coaches. The tell: defensive reactions and reduced use of the information, which breaks the intended service role.
• Comparing raw totals across unequal samples. The tell: a team appears better only because it had more possessions.
• Using coach recall to fill gaps in the data. The tell: coaches overrule the analysis with a 'felt' account, though recall studies show it misses a large share of critical incidents.

• ▸Decision to inform: one sentence naming the coaching or preparation decision this cycle serves.
• ▸Unit of analysis: individual, team, opponent, referee, or training/match; pick one.
• ▸Input method: sequential, frequency table, scatter diagram, or combined; justify in one line.
• ▸KPI definitions: list each KPI with its exact inclusion rule and whether it is reported as a count or a percentage.
• ▸Reliability check: record who re-coded what, when, and the agreement result before reporting.
• ▸Coach recall vs data: note the coach's pre-analysis impression next to the coded result and the gap between them.
• ▸Video clips: list the tagged clips (good performance and areas for attention) with one telestration point each.
• ▸Training action: the specific drill or tactical change that follows from the finding.
• ▸Re-analysis target: the figure you expect in the next match or session and the date you will recode with the same definitions.

1. 1Start from the coaching question, not the software. Write one sentence naming the performance issue (e.g., serve effectiveness into the advantage court) and the decision the output must inform. The source's six-step model for gathering quality information on sports performances starts from what to analyse, and system design follows from that.
2. 2List the variables the question needs and nothing more. For each one, write every permissible option in advance. The tennis serve example splits the advantage-court target into wide, body and T zones. Each variable should be a closed list, not free text.
3. 3Choose the data-entry format that fits the event. The source shows several manual formats: a sequential system (tennis serve and rally outcome, boxing punches, cricket bowling), a frequency table (netball), and a scatter diagram (soccer shot locations). Use a sequential layout when order matters, a frequency table when only counts matter, and a scatter diagram when location matters. Combine formats when one is not enough, as in the doubles squash system that pairs sequential entry with a scatter diagram.
4. 4Make entry error-proof. In a spreadsheet, build data validation lists with all variable options listed under column headings, so the operator selects from a list instead of typing. In a computerised package (Hudl SportsCode, Angles, Nacsport), build the code template or input panel the same way. The event buttons or options should mirror your variable list one to one.
5. 5Decide the link type before you code. Decide whether you store events as a sequence (e.g., serve followed by rally outcome), as x and y coordinates (soccer ball events), or as matrix data (basketball player). The link type fixes which questions you can answer later, so choose it from the output you need.
6. 6Build the output at the same time as the input. In a manual system, set up a pivot table with multiple variables selected to produce detailed KPI results, or a totalled frequency table. In a computerised system, build the summary view (the source shows a netball summary analysis and a manual summary sheet for the same system). Define derived measures such as passes per possession, or the percentage of possessions leading to scoring opportunities, as formulas up front.
7. 7Pilot the system on a real performance, then test reliability before you collect the full dataset. The source places reliability testing between development and use: develop, test for reliability, then collect data. Fix any variable or option the operator hesitates on or the second coding disagrees on.
8. 8Format the final output for the coach's use: a summary sheet or view the coach can read quickly. Keep the raw data available behind it.

The source's tennis material gives the skeleton for a serve system. It shows a sequential system for the tennis serve and rally outcome (Fig. 4.1), and it measures technical effectiveness by success into the advantage court split into wide, body and T zones (Fig. 3.1). The coaching question is where first serves into the advantage court succeed. The columns are: serve number, zone (wide / body / T), serve outcome, and rally outcome. Each column gets a data validation list, so the operator picks from fixed options and cannot type variants such as 'T' versus 'tee'. Each serve is one row, entered in sequence. A pivot table then crosses zone with serve outcome and rally outcome to produce the KPIs, with no extra data entry. If the coach later asks where the serves landed, a scatter diagram could be added, as the squash system does by combining sequential and scatter input. The passages supply only captions for these figures, so the column names and options above are a design reading of them, not quoted content.

• If the question depends on the order of events (serve then rally outcome, punch sequences), then use a sequential layout; if only totals matter (netball counts), then use a frequency table.
• If location is part of the question (shot locations in soccer), then add a scatter diagram or x/y coordinates; if one input method cannot capture both order and place, then combine two, as in the doubles squash system.
• If a variable can be typed in free text, then convert it to a validation list or a button with a fixed option set before any data is collected.
• If you cannot name the KPI a variable feeds, then drop the variable; if a KPI needs a derived measure (passes per possession, % of possessions leading to scoring opportunities), then build the formula before coding.
• If the coach needs a view during or straight after the match, then choose the format that produces the summary fastest (a pre-built pivot table or summary sheet) over the one that records the most detail.

• Variable overload: the system records everything possible and the coach cannot find the answer. The tell: you cannot name the coaching decision each column serves.
• Free-text entry: the same event appears under several spellings and the pivot tables split. The tell: pivot rows with near-duplicate labels.
• Output built last: the system collects data in a form that cannot be summarised into the KPIs the coach wants. The tell: you are reformatting raw data by hand after every match.
• Wrong link type: events are stored as isolated counts when the question is about sequences, or as sequences when only totals are needed. The tell: the coach asks 'what happened before that?' and the data cannot answer.
• Skipping the reliability check: the system is used for full data collection before testing. The tell: a second coding of the same performance gives different results.

• ▸Coaching question: one sentence naming the performance issue and the decision the output informs.
• ▸Variables: each variable the question needs, one per line, with its reason for inclusion.
• ▸Option lists: every permissible value per variable (e.g., wide / body / T), written out in full.
• ▸Entry format: sequential, frequency table, scatter diagram, coordinates, matrix, or a named combination, with the reason.
• ▸Platform and validation: spreadsheet validation lists or package template (e.g., SportsCode, Angles, Nacsport) and how options are constrained.
• ▸Link type: how events connect (sequence, location, player matrix) and which questions that allows.
• ▸Output and KPIs: the pivot table, frequency totals or summary view, with the formula for each derived KPI.
• ▸Reliability plan: who recodes, which performance, and the agreement level you require before full collection.
• ▸Coach sign-off: the coach's check that the output answers the question, with the date.

1. 1Classify every variable by how it is recorded. Automated measures (ball and player tracking, heart rate, some biomechanical variables) are objective at the recording level. Any indicator derived from them can be assumed objective. Tagged video variables need an objectivity test.
2. 2Define only the action variables (event type, outcome, possession) and build every performance variable from them. For example, % of long passes reaching a teammate = successful long passes ÷ all long passes, computed per team. Once the action variables are defined and made objective, the derived variables inherit that objectivity.
3. 3Devolve judgement to the officials wherever you can. Record the official's decision even if you disagree, because it is what happened in the match. This removes subjective calls on rule infringements, scores and who last touched the ball before it went out.
4. 4Write guidance for the remaining judgement calls, such as 'any ball transfer from a player to a teammate is a pass'. Then train the users, because operational definitions do not guarantee objectivity. Avoid thresholds an observer cannot measure by eye: a 10 m long-pass rule fails because a pass of about 10 m cannot be judged as just under or just over.
5. 5For subjective classifications, run an inter-operator agreement study. Two trained observers analyse the same performance, or several performances, independently.
6. 6Align the two event lists before computing agreement. Time-match the events and insert a null event wherever one observer recorded an event the other did not. Then cross-tabulate the two sets and compute kappa, which adjusts agreement for chance.
7. 7Test the numerical performance variables as well as the events. Analyse multiple matches so that both error components show up. Systematic bias is the average difference between observers across matches, where one observer counts a behaviour more readily than the other. Random error is the remaining non-uniform differences between corresponding observations.
8. 8Report the reliability in the methods section of the study that uses the system. Do not split it off as a separate paper, since journals reject unreliable variables and also reject standalone reliability studies of single-use systems.

Take a soccer analyst who sees a player contact the ball and a teammate pick it up. The analyst is unsure whether to tag a pass or a touch, and the choice goes into the event list as a categorical value. The data will then produce counts of passes and touches that look quantitative, but the underlying decision was a subjective perception of behaviour. The fix starts with guidance: any ball transfer from a player to a teammate is recorded as a pass. Two trained observers then code the same match independently. Their event lists are time-matched, and a null event is inserted wherever only one of them tagged something. The two sets are cross-tabulated and kappa is calculated, so agreement is judged after adjusting for chance. To check the variables derived from these events, such as pass counts per match, both observers code several matches. The mean difference in their pass counts across those matches estimates systematic bias, and the leftover scatter estimates random error.

• If the variable comes from automated tracking or heart rate with no human judgement at capture, then treat it as objective and spend your effort elsewhere.
• If a tagged event can be settled by an official's decision (score, infringement, who last touched the ball out of play), then record the official's call, not your own.
• If a definition needs a distance or area the observer cannot measure by eye, then drop the threshold or reword it. Do not rely on a definition the observer cannot apply.
• If an event is subjective and cannot be defined precisely, such as defensive style, then accept subjective classification by experts and demonstrate objectivity with an inter-operator kappa study.
• If the observers' event counts differ, then time-match and insert null events before cross-tabulating for kappa.
• If you will report numerical performance variables, then analyse multiple matches so that systematic bias and random error are both represented.

• Assuming an operational definition makes data objective. The tell is that definitions exist but observers have not been trained or tested against each other, so their tags diverge.
• Using precise-sounding thresholds the observer cannot judge, such as the 10 m long pass. The tell is borderline passes tagged inconsistently between observers.
• Computing kappa without time-matching the events. The tell is observers with different event counts and misaligned rows in the cross-tabulation.
• Establishing event-level agreement only and assuming the variables are reliable. The tell is a kappa result with no multi-match check of systematic bias and random error in the counts.
• Choosing an attractive but hard-to-judge variable, such as forced versus unforced errors. The tell is that distinguishing them needs sport knowledge and is hard at elite level, so the values are compromised by subjective collection and can be invalid, whereas winner versus error is distinguished with high reliability.

• ▸Variable and action definition: write the exact operational definition and any guidance rule (for example, any ball transfer to a teammate = pass).
• ▸Collection type: automated, official's decision, or observer judgement. Only observer-judgement items need the agreement study.
• ▸Observer training completed: record who was trained, on what, and when.
• ▸Performances analysed: list the matches coded independently by both observers.
• ▸Event alignment: note how events were time-matched and how many null events were inserted.
• ▸Kappa result: enter the kappa from the cross-tabulation of the two event sets, adjusted for chance agreement.
• ▸Systematic bias: enter the average difference in variable values between observers across matches.
• ▸Random error: describe the remaining non-uniform differences between corresponding observations.
• ▸Corrections and decision: list the definitions revised or retraining done, and state whether the system is accepted for use.

1. 1Name the decision first. Write the specific preparation or tactical decision the variable will inform (e.g., how to prepare for an opponent's defensive style). The source states a valid variable 'must be useful' and support decisions about preparation and tactics. If you cannot name the decision, drop the variable as a 'so what?' or 'who cares?' variable.
2. 2Run the logical (face) validity check. Ask whether the variable is obviously the thing it claims to measure, as the 10,000 m time is for a 10,000 m runner. If it is not obvious, move to the criterion or construct checks below.
3. 3Run the criterion validity check. Compare the variable against a gold standard such as match result, winning margin, time or distance achieved. Check whether it is associated with the margin, or whether winners and losers have clearly different values.
4. 4If the variable is a created measure (not directly observable, like the Eagle rating), apply the construct validity test. Does it clearly distinguish performers of known different ability levels? If not, it is not valid as a construct.
5. 5Classify each indicator as effectiveness or style. Tactical indicators (e.g., pressing) represent style irrespective of success, so they need not separate winners from losers. Test them on whether they separate teams adopting different styles, and keep them if the style matters for opposition preparation.
6. 6Test the validity-reliability trade-off before adopting. Ask whether two observers would record the same value. Prefer variables decided by officials or automatic tracking. Where the judgement needs sport knowledge (e.g., forced vs unforced errors), accept that subjective collection can compromise validity.
7. 7Check content validity at profile level. List the relevant aspects of performance and confirm your set of indicators covers all of them. Add indicators for any aspect left uncovered.
8. 8Apply the quality-over-quantity rule with the recipients. Confirm with the coach and athletes that the information is relevant to their philosophy and the environment (in-game, post-game or training), and cut anything they would deem inadequate.

Take the tennis error variables from the source. You want to report errors in a match. Winners versus errors are easy to separate with high reliability: a ball that lands in the opponent's court and cannot be played before a second bounce is a winner, and a ball that lands outside or hits the net is an error. Forced versus unforced errors are more relevant to practitioners, and media coverage reports unforced errors too. But classifying an error as unforced is hard when the shot it answers was well placed, fast and spun, especially at elite level, and it requires knowledge of the sport from the person collecting data. Face validity is high, but subjective collection can compromise the recorded values and so the variable's validity. The decision: report total errors with confidence, and report unforced errors only with a defined rule and a stated caveat. This is the validity-to-reliability trade-off in practice.

• If a variable is hyper-accurate but supports no preparation or tactical decision, then drop it as a 'so what?' variable.
• If a variable is clearly relevant but recorded values depend on the observer's opinion, then treat its validity as compromised and tighten the operational definition or replace it with an official-decision or automatically measured variable.
• If an indicator is tactical (style, such as pressing), then do not reject it for failing to separate top-half from bottom-half teams; judge it on whether it separates teams using different styles.
• If a measure is created rather than directly observed, then require it to distinguish performers of known different ability before using it.
• If you assemble a performance profile, then check that the indicators cover all relevant aspects of performance; otherwise content validity fails.

• Collecting accurate but irrelevant variables because they are easy to count. The tell: the coach asks 'so what?' and no decision changes.
• Reporting a subjective variable as if it were objective (e.g., unforced errors). The tell: two analysts tagging the same match give different counts.
• Discarding a tactical indicator because it does not separate winners from losers. The tell: pressing counts show no top-half versus bottom-half difference and you declare the indicator invalid.
• Building a profile with gaps in coverage. The tell: a relevant aspect of performance has no indicator, so a coach cannot discuss it.
• Prioritising technical data volume over what recipients need. The tell: large data sets that coaches and athletes deem inadequate to the environment.

• ▸Variable / indicator: name and operational definition.
• ▸Decision supported: the preparation or tactical decision it informs.
• ▸Logical (face) validity: is it obviously the thing measured? Yes/No and why.
• ▸Criterion check: the gold standard used (result, margin, time, distance) and whether winners and losers differ.
• ▸Construct check (if created measure): does it separate performers of known different ability? Evidence.
• ▸Type: effectiveness or style (tactical). If style, which styles should it separate?
• ▸Objectivity: who or what decides the value (official, tracking, analyst judgement)?
• ▸Content coverage: which aspect of performance in the profile it covers; any aspects uncovered.
• ▸Recipient verdict: coach/athlete view on relevance for in-game, post-game or training use; keep, revise or drop.

1. 1Place the feedback inside the coaching cycle before you build it. In Franks' (1997) model the game is played, data are collected live or post-match, analysed, and feedback informs preparation for the next match. In Irwin et al.'s (2013) model the cycle runs between training or competitive performances. Name the next decision the feedback must inform, and work backwards from it.
2. 2Fix a delivery deadline. Jones et al. (2004) suggest recall accuracy falls over time and delays can make feedback inaccurate. Set the time between performance and feedback in advance, and don't let analysis polish push you past it.
3. 3Pair a record with the story. Bias, emotion, fatigue and limited memory distort how coaches and players remember a contest, so give them video, match statistics or both rather than anyone's recollection. Open with the record when it contradicts the live impression.
4. 4Choose the audience and shape the output for them. Coaches and players use the information directly. High-performance managers use multi-match information for trends and strategic decisions. Strength and conditioning coaches use movement data such as GPS. Sports psychologists may request video sequences of particular periods of play. One output does not serve all four.
5. 5Check reliability before you present anything. Inter-operator agreement on variables (for example possession type, event type, disposal) should be tested first, because a clear dashboard of unreliable data is still wrong.
6. 6Build the dashboard from the KPIs, not from the raw data. Use the frequency tables and pivot tables you already have, and select several variables in the pivot table to produce detailed KPI results. The book gives eight principles for manual notation output dashboards (Table 4.2). The supplied extract does not list them, so apply that table when you design.
7. 7Telestrate one message per clip, using the tool that fits it. The passages' rugby example uses a spotlight on one player, directional arrows with numbers for options, a magnifier with the rest of the frame greyed out for a subtle action, a region graphic for space, and a 3D arrow for trajectory. Add each graphic only if it carries the message.
8. 8Deliver in a way that involves the player. The book shows a player using a touchscreen during a team meeting and tablet control of the analysis software (RT Tactic). Hand over control where the learning objective allows it, rather than presenting passively.
9. 9Review your own delivery. Coach-behaviour tagging (Focus X2) lets coaches reflect on their style. Apply the same logic to analyst feedback by checking what you showed, to whom, and how.

The book's rugby telestration sequence (Figure 7.7) shows a single phase of play built in three stages. First, play develops outside the opposition 22. The fly half (number 10) is spotlighted, and directional arrows with numbers mark the four attacking players or options, so the viewer reads the decision before seeing any outcome. Second, a magnifier tool shows the winger's raised hand, a subtle action. The rest of the image is masked or greyed out, which draws the eye to the circle in the top right corner. Third, a region graphic highlights the space behind the defensive line, and a 3D arrow shows the kicking option and the trajectory a kick in behind the line would need. Each stage has one visual job, and the tool is chosen for that job. The boxing anecdote supplies the reason for showing the footage at all. The author felt his boxer was losing until a stoppage win, yet on video, knowing the result, he judged the boxer to have led on points from the end of round one. Live perception and the record can disagree, so the record goes first in the meeting.

• If the next match or session is close and recall is fading, then send a short, targeted clip package now rather than a full report later, because delay degrades recall (Jones et al., 2004).
• If a coach's or player's recollection conflicts with the data, then show the video or statistics before discussing, because emotional involvement and fatigue distort recall.
• If the audience is a high-performance manager, then aggregate across multiple matches to show trends. If it is a coach or player, then use a few clips and KPIs tied to the next performance.
• If a clip contains a subtle action, then use the magnifier with masking. If the issue is space or trajectory, then use a region graphic or 3D arrow. If it is a player's options, then use a spotlight and numbered arrows.
• If inter-operator agreement on a variable has not been checked, then keep it off the dashboard or label it provisional.

• Late feedback delivered from memory or after a long delay. The tell: players dispute the account, or your summary differs from what the footage shows.
• Telestration overload, with every tool on one clip. The tell: viewers ask what they are meant to look at, or recall the graphics but not the point.
• A dashboard that reports raw counts with no KPI or context. The tell: coaches ask for the same figure to be re-cut every week, or ignore the sheet.
• One output sent to every audience. The tell: managers receive clip-level detail while players receive trend tables, and neither acts on it.
• Presenting unverified data with confidence. The tell: two operators' disposal or event codings disagree when someone checks, and the meeting loses trust.

• ▸Audience: coach, player, high-performance manager, S&C or psychologist. This sets the format and depth.
• ▸Decision it informs: the specific next-match or next-session decision.
• ▸Delivery deadline: the time between performance and feedback, set in advance to limit recall decay.
• ▸Evidence type: video, statistics or both, plus the one headline KPI.
• ▸Reliability status: inter-operator agreement checked (yes or no) and any provisional variables.
• ▸Dashboard contents: the KPIs shown, from the pivot or frequency table, and the manual-notation dashboard principle each follows.
• ▸Clip list: for each clip, the single message and the telestration tool used (spotlight, arrows, magnifier, region graphic, 3D arrow).
• ▸Delivery mode: presented, or player-operated via touchscreen or tablet, and who controls it.
• ▸Follow-up check: how you will confirm the message landed and was applied in the next performance.

1. 1Treat relational skill as part of the job, not a side issue. The source says soft skills (interacting with colleagues, communicating and listening, managing time to deadlines, solving problems alone or in a team) are often overlooked. Before each feedback cycle, write down who you must listen to first (coach, player, support staff) and what each needs from the data.
2. 2Profile the performer before choosing the feedback. Record anthropometric characteristics, attentional style, psychological factors, physical ability and experience, because athletes in the same sport may need to perform differently. Then set the standard: for beginners, analyse against common faults; for athletes approaching expert level, compare against the ideal performance so subtle technique differences show.
3. 3Separate the decision from the outcome before you speak to a coach or player. Judge a decision on the perceived probability of success, the consequences if it fails in this situation, and the player's awareness of the options available. Never let the result alone settle whether the choice was right.
4. 4Frame the contested options as questions. If you rule out options the player could not have been aware of from their position or under that pressure, say so openly and check it with the player and coach rather than presenting it as a verdict.
5. 5Choose the delivery route to fit the audience. Present results collectively or individually. Use internet-based coaching management systems (CoachLogic, My Team Performance) or social media so players engage at times and places that suit them. Use online discussion before players meet so training time is spent training.
6. 6Plan the match-day interaction window. Keep the coded video correct and ready to copy onto the coaches' computers before everyone leaves the venue. Corrections take 15–20 minutes while players cool down and change. Tell people in advance why you will look unavailable, since the source notes you may appear antisocial.
7. 7Take coaches and players through the video together using matrices of situations and decision-making parameters. Use telestration to highlight the key areas of frames. Use the discussion to tailor coaching advice and training activities, and align your information with the other support services (physical, psychological, medical, lifestyle) so the athlete gets one coherent picture.

A soccer analyst codes a penetrative pass that was intercepted. Judged on outcome alone, it is a poor decision. Following Lorains et al. (2013), the analyst separates the decision from the execution and checks three things. First, the perceived probability of success: the pass may succeed less than 0.25 of the time, but attempted often enough it creates scoring opportunities, so the question is whether this choice is sound across similar situations in the match. Second, the consequences of failure here: teammates were poorly positioned to deal with a counterattack, which makes the attempt a poor option in this moment. The reverse also holds, since a pass that created a chance could still be a poor decision. Third, awareness: the analyst sees four options but rules out two the player could not have seen from where he stood or under the pressure he faced. The analyst brings this to the player and coach as a sequence of video clips found through the situation and decision-making matrix, and asks what the player saw. The conversation then shapes the coaching advice and training activity rather than ending in blame for the interception.

• If the athlete is a beginner, then frame feedback around common faults; if they are approaching expert level, then compare against the ideal performance to expose subtle differences.
• If a result is good or bad and you are about to judge the decision from it, then stop and assess probability of success, consequences if it fails, and the player's awareness of options first.
• If you rule out an option as one the player could not have known about, then confirm it with the player and coach rather than asserting it.
• If players can engage with feedback remotely and discuss it in advance, then use online systems so contact time is spent on training.
• If you must correct data in the post-match window, then tell the coaches and players beforehand so your absence is not read as being antisocial, and deliver the complete video before everyone leaves.

• Technical-only analyst development: you produce accurate data but poor listening, communication and time management. The tell: coaches and players do not engage with, or ask follow-ups on, what you send.
• Judging decisions by outcomes: you label a failed pass a bad decision, or a lucky one a good decision. The tell: your decision-accuracy variable simply duplicates the outcome variable.
• Treating all athletes alike: you apply one feedback standard to everyone. The tell: experts get fault-finding they have outgrown, or beginners get ideal-model comparisons they cannot use.
• Imposing your own view of the options: you assume the player saw all four options. The tell: the player disagrees about what was visible from their position and you have no answer.
• Being unreachable post-match without explanation: you work silently through the 15–20 minute correction window. The tell: staff and players read you as antisocial and do not trust the delivered video.

• ▸Audience: name the coach, player(s) or staff receiving the feedback and whether it goes collectively or individually.
• ▸Performer profile: note anthropometric characteristics, attentional style, psychological factors, physical ability and experience.
• ▸Standard applied: state common faults (beginner) or ideal performance (near-expert) and why.
• ▸Decision under review: describe the action and the outcome, kept separate.
• ▸Probability and consequence check: give the estimated chance of success and what happens if it fails in this situation.
• ▸Player awareness: list the options available and which the player could realistically have seen from their position and under that pressure.
• ▸Delivery channel and timing: choose live meeting, touchscreen or telestrated clips, or an online system with pre-training discussion.
• ▸Match-day handoff: record the deadline for the corrected video and who has been told about your busy window.
• ▸Agreed follow-up: note the coaching advice or training activity the discussion produced and who owns it.