The only evidence collected before students touch a bot is the Pre-Task Diagnostic. It's not graded — its job is to give you a per-concept starting line so growth in the post-task is real and not retrofitted.
Every student's individual Pre-Task Diagnostic responses, then the per-concept breakdown on the Diagnostic Growth Dashboard (proportional reasoning, slope, unit rate, ratio, proportional scaling, coordinate geometry, linear decay, distance).
Growth Dashboard ↗Not the goal here — this is a baseline, not a mastery check. Strong = students filled it out completely without coaching and the dashboard shows a useable per-concept distribution.
Multiple blank answers — usually means a student gave up rather than guessed. Privately redirect with "just put your best guess." Don't react to wrong answers in front of the class; that breaks the no-grade frame the post-diagnostic depends on.
You've reviewed the dashboard and identified the 1–2 concepts with the widest baseline gap. Those become Open Lab scaffolding targets in Week 1 and the re-teach priority if Cold Recall (Day 10) confirms they didn't move.
Use these in a 20-minute pre-program meeting once the diagnostic dashboard has data.
- Looking at this cohort's Pre-Task baseline, which concept's gap surprises you? What would you change about a specific Day 1–5 mini-block in response?
- If two students with very different starting profiles end up on the same team, what's your plan for making both their thinking visible in the team's gap sentences?
Every day produces three reliable artifacts: the Challenge Block exit ticket, the dashboard loadout entry (Days 1–5) or match log (Days 6+), and the ENDS reflection. These are what you act on the next morning.
The exit ticket at the end of every Challenge Block. Three required pieces: the formula stated, the calculation shown with the student's own numbers, and one plain-language sentence interpreting the answer.
Day 5 exit ticket exemplars ↗"Charge% = Power² = 8 × 8 = 64. So at full Power 8, my bot kicks at 64% charge. That's more than double what I had at Power 4 (16%) — the quadratic is why doubling Power quadruples force." Formula + math + plain interpretation.
Writes the answer with no calculation shown. Copies the formula but plugs in someone else's numbers. Skips the sentence. The sentence is the goal-1 evidence — a correct answer with no interpretation is not yet "uses formulas as predictive models."
80% of exit tickets show all three pieces. Below 80%, hold the next morning's warm-up for a 5-minute re-teach naming the missing piece (almost always the sentence). Don't paper over it — the gap compounds across Days 1–5.
Each team's dashboard loadout row: Speed + Endurance + Turning + Power = 20, Data Owner named, and the three-part gap sentence (formula predicted X / bot did Y / gap is Z) in the Notes field.
Pre-Season Dashboard Guide ↗Total sums to 20, every stat has a one-sentence reason citing the previous day's lesson, and the gap sentence uses a specific number ("formula predicted 30%, bot did 22%, gap is 8 percentage points").
Stats sum to 19 or 21 — the equals-sign-as-arithmetic misconception. Stats picked with no reasoning ("we just felt like it"). Gap sentence in feeling-talk ("we did bad", "the bot was off") with no number.
Total = 20, Data Owner named, and the gap sentence contains an actual number. Don't approve a loadout that doesn't balance — the budget constraint is the load-bearing math of the whole program.
One sentence per student in the journal — the prompt varies by day but the structure is always: one specific thing I noticed, one specific number to back it up. Read three or four aloud before dismissal.
"I noticed our Speed 6 bot moved slower than the Speed 8 bot by about 5 cm/s, even though the formula says they should differ by 10%. I think the wheels are sticking on the start." Observation + number + hypothesis.
"It was fun." "I learned a lot." No number, no specific. Or every student in a team writes the same sentence — usually means one student wrote it and the rest copied.
Each student's sentence is different from their teammates' and contains one specific observation. Skip the read-aloud only if you're already past the dismissal bell — never skip the writing.
A revised artifact — exit ticket, loadout, equation, or proposal — that cites or addresses the specific feedback from the previous round. Revision happens in the next morning's 5-minute warm-up, the next Open Lab, or the next match cycle.
"On Tuesday my exit ticket missed the plain-language sentence. Today I added it." Or: "Our Day-3 stat reasoning was off because our Day-2 gap sentence showed Endurance was too low — here's the updated allocation, with the new decay calculation."
Exit tickets show the same gap pattern as yesterday with no acknowledgment of yesterday's feedback. A team writes a Loss Recovery plan, the plan doesn't help, and they don't return to update it. Revision treated as "if I have time" instead of "this is the work."
At least 60% of students show one revised artifact citing earlier feedback by end of Week 2. Below that bar, the next morning's warm-up should explicitly model the revision pattern — pick yesterday's strong example and revise it live so students see what revision looks like.
Use these in a 30-minute weekly debrief, bringing actual student work to the table.
- Bring three exit tickets from this week — one strong, one weak, one mixed. What language pattern distinguishes them, and what would the next morning's 5-minute warm-up teach?
- Pick one team whose dashboard loadout reasoning matured between Day 2 and Day 5. What changed in their writing? Can you name the facilitator move that triggered it?
The match is the test. The prediction, the residual, and the loss-recovery plan are the evidence. None of them are scored — they're tracked, and they make Goals 1 and 3 visible in real time.
A written prediction on every team's data log before the match begins: predicted goals, the formula used to get there, and the input values.
Day 7 prediction exemplars ↗"We predict 2 goals. Charge% = Power² = 36 at our Power 6, so we get about 2 hits per match at full force. Their Endurance is 4, ours is 7 — they'll lose power before round 3."
"We'll win." No number. Or a number with no formula behind it ("we'll get 3 because we feel good"). Or a prediction made during the match — that's a guess pretending to be a prediction.
Every team has a written prediction citing a formula. The match does not start until they do — this is the non-negotiable that converts a match from drama into data.
Predicted − actual = residual, with a + or − sign and one sentence naming a specific cause when the gap is over 20%.
Day 8 residual exemplars ↗"Predicted 3 goals, scored 1, residual −2. Our Endurance equation didn't account for the ping pong ball drag in Chaos mode — we'll re-derive the decay rate for the new ball before the next match."
Residual recorded but no hypothesis. "The bot just had a bad day." Or the cause is a feeling ("we got nervous") instead of a math parameter. Or the cause names a variable not in their equation — a sign they're guessing.
Residual + signed direction + a single-variable hypothesis for any gap > 20%. Below that bar, run Loss Recovery before the team touches the bot for the next match.
The Loss Recovery Protocol output: a written change for next match, citing the residual or a dashboard number, with the changed variable named.
Loss Recovery Protocol ↗"We'll drop Power from 8 to 6 and raise Endurance from 4 to 6. Reasoning: our match-time decay was 35% faster than predicted, and we don't need the extra force because we only used Charge% twice."
Team nods along during the protocol but writes nothing. Or changes a stat without citing data ("let's try more Speed"). Or the team that lost shuts down and you let them skip — that's how the gap between teams widens.
The team has a written, data-cited plan. Don't let them code the bot for the next match until the plan is on paper — that's the rule that converts a loss from "we're worse" into "we have one more data point."
Use these after a match-heavy day, while predictions and residuals are still fresh.
- Compare two teams' residual hypotheses from the same match round. Whose hypothesis was more testable, and what did the less-testable team's language signal about their model?
- After a Loss Recovery cycle: did the team's next-match plan cite the dashboard or their gut? What's the conversation you'd have if it cited their gut?
Three named checkpoints test what's accumulated: the scatter-plot equation on Day 8, the Cold Recall on Day 10, and the Commissioner's Meeting on Day 12. Each surfaces a different layer of evidence.
Each team's worksheet: two real points from the class scatter plot, slope calculated, b solved, equation y = mx + b written, slope interpreted in context, y-intercept interpreted in context.
Day 8 linear equation exemplars ↗"Goals = 0.4 × Speed + 0.5. Slope means 1 extra Speed point gives 0.4 more goals on average. The y-intercept of 0.5 means even a Speed-0 bot might score one goal if the opponent self-destructs."
Writes the y = mx + b template but doesn't compute m from real points. Computes m correctly but says the intercept is meaningless. Or every team writes the same equation — usually means they copied from the board instead of using their own pair of points.
75% of teams have their own equation with a slope interpretation. If not, the next morning's warm-up is class-wide scatter-plot interpretation — Day 13's residual work depends on this being solid.
Cold Recall Check: 5 questions, individual, no notes, no devices, no grade. Tally per-question pass rate. Q4 (Endurance decay) is the load-bearing one — it gates Season 2.
Day 10 Cold Recall ↗Formula written cold + one valid worked example with the student's own numbers. Q4 especially: P(t) = 100 − (20 − E) × 1.5 × t, evaluated correctly at a specific E and t.
Blank on Q4. Writes the formula but can't substitute. Q4 is the diagnostic question — anyone who misses it will be lost in Day 13's residual-and-equation work.
Not a class threshold — flag specific students. Anyone who misses Q4 gets the Linear Decay Math Concept page assigned in the next Open Lab, no group work, individual remediation.
Each team's proposal: 2 specific dashboard numbers cited, a percent-change calculation showing impact, the rule's effect on the last-place team named, and confirmation the proposing team's own allocation still satisfies the rule.
Day 12 Commissioner proposal exemplars ↗"Cap Power at 6. Reason: 4 of 4 teams have Power ≥ 7. Charge% at P=7 is 49% vs 36% at P=6 — that's a 36% jump for one stat point. The last-place team is at P=3, so this cap doesn't affect them, but it forces high-Power teams to redistribute."
Argument from feeling ("it's not fair"). One number cited, not two. A rule that helps the proposing team more than the last-place team. A rule whose math says one thing but whose plain-language framing says another. Bounce these at the morning verification, not at the meeting.
Every proposal has 2 cited numbers + percent change + impact on last-place + self-check. The meeting itself is then a math conversation, not a political one.
Use these at the end of Week 3, with Cold Recall tallies and draft Commissioner proposals in hand.
- Cold Recall Q4 — who passed and who didn't? What did the passing students do in Week 1 or Week 2 that we want to make visible for the others?
- Bring three Commissioner proposals. Which one's math survives scrutiny? Which would you bounce at the morning verification, and how would you frame the bounce so the team revises instead of withdrawing?
Two summative measures and one cumulative review: the Post-Task Diagnostic (math you can measure), the Visitor Mini-Challenge tracker (math students can teach), and a final pass against the four Program Goals. All three produce evidence you can report.
Per-concept percent change on the Growth Dashboard: Pre-Task score → Post-Task score across all 8 concepts, anonymized at the class level.
Growth Dashboard ↗Positive percent change on every concept (any growth is success). The largest movement is usually on slope, proportional scaling, and linear decay — concepts the bot work hammers daily.
A concept with zero or negative percent change. That concept is the next cohort's redesign target — usually because its Days 1–5 placement got rushed or its formula never appeared in a match prediction.
You've reviewed the per-concept growth and named one concept that under-performed. That's the artifact you bring to next-cohort planning, alongside the visitor tracker.
Visitor Completion Tracker filled in per visitor per team: did the visitor complete unit-rate, coordinate-distance, and proportional-scaling first-try? Yes / no / partial.
Day 16 Community Exhibition ↗Visitor explains the math back to the student in their own words after completing the challenge. Student adjusts a confused visitor without re-doing the calculation for them.
Student does the calculation for the visitor. Visitor leaves without completing. Student uses bot jargon ("our Motor%") instead of plain math ("speed times five over twenty"). Track these — they're the explanations to revise for the next cohort.
The exhibition is the summative — there is no separate "move forward." Goal 4 is scored as a binary observable: did first-try completions cross 70% of visitors? That's your stake-in-the-ground.
For each goal, the artifact named in the Toolkit Guide's evidence statements: prediction-residual-change cycle (Goal 1), model-comparison entries (Goal 2), gap-sentence-to-improvement chains (Goal 3), revised-explanation records (Goal 4).
Toolkit Guide — Goals + Evidence ↗You can point to at least 2 named teams per goal with the specific artifact dated. Goal 2 stretch: at least one team can trace a keep/change decision to improved match performance.
A goal with no evidence collected — usually Goal 2 (Ownership), because the model-comparison artifact isn't built into a standalone activity. If this is empty, plan a model-comparison block into the next cohort.
You've written one sentence per goal about what the program produced — that's the artifact for stakeholders. Combine with the Growth Dashboard + Visitor Tracker for the full picture.
Use these in the post-program debrief, before the next cohort's planning begins.
- Per-concept growth on the Post-Task — which concept under-performed, and what's the one change you'd make to its Days 1–5 placement for the next cohort?
- Pick one student whose Day 16 visitor explanation worked and one whose didn't. What was different in their ENDS journals across the four weeks?
Each of the four Program Goals is named in at least three evidence moments above. Counts are tag occurrences across all the cards. This is the evidence map a stakeholder can ask for: "show me where Goal N is observable in the program."