Lesson 4.4.3 — Communicating the Unknown Unknowns

Week 43

Prerequisite lessons: 4.4.2 (proposing concrete improvements), 4.4.1 (architecting the defense), 4.3.3 (long-horizon robustness), 4.3.2 (diagnosing sim-to-real failures), 4.2.4 (safety and recovery), 3.3.4 (safety and constraints on LLM outputs).

Prior results used in this lesson. From 4.4.2: the improvement’s five fields, the kill criterion, and the registry’s eight fields, restated in §L.A.3 and §L.D.2. From 4.4.1: the claim-evidence-warrant table, the vulnerability map, and the four-element claim, restated in §L.A.1 and §L.B.2. From 4.3.3: the horizon’s failure’s law, the extrapolation’s contract, and the register’s horizon entries, restated in §L.B.3 and §L.C.2. From 4.3.2: the diagnosis’s evidence chain and the four failure modes, restated in §L.C.1. From 4.2.4: the register’s six entries, the safety case, and the residual’s acceptance, restated in §L.B.1 and §L.E.3. From 3.3.4: the four-layer stack and the register’s form, restated in §L.A.2. Each is restated at the point of use.

What this lesson hands forward. The four quadrants, the structural sources of the unknown unknowns, the reversal test, the three audiences’ communication forms, and the handoff’s artefacts. This is the curriculum’s final lesson. Its exit test is the capstone’s defence, and the defence’s closing argument is the lesson’s content.

The running case in one sentence. The running arm, gripper, eye-in-hand camera, table, cube, bin, divider, and three distractors are unchanged, and the full-stack agent’s acceptance has been measured, the safety case written, the diagnosis chained, and the proposals priced; this lesson asks how to communicate what the agent’s evidence does not cover — the unknown unknowns — to the three audiences who must act on the communication: the operator who accepts the risk, the committee who judges the argument, and the successor who inherits the system.

§L.0 — Lesson map, objectives, and misconceptions

Observable outcomes. By the end of this lesson you can, without notes: (1) distinguish the four quadrants and state what each one’s communication form is; (2) name the five structural sources of the unknown unknowns and give one running-case instance of each; (3) match the four forms to the three audiences and state why the match is not arbitrary; (4) apply the reversal test to a claim and state the falsifier’s three properties; (5) write the “if I’m wrong, here’s how I’d know” statement for the capstone’s central claim; (6) distinguish the honest uncertainty from the hedge and state the hedge’s signature; (7) write the closing’s three sentences — the claim, the residual, the invitation; (8) name the handoff’s four artefacts and state what each carries that the code does not; (9) assemble the running case’s final defence with the quadrants, the reversal, and the closing; (10) state the curriculum’s exit test as the defence’s passing.

Misconception inventory.

Everything is self-contained. Where a prerequisite lesson is used, the result is restated at the point of use.

§L.0.5 — Retrieval warm-up

Q1. In 4.4.1 §L.A.1, the defensive claim’s four elements were the claim, the evidence, the warrant, and the residual. What does the residual’s element add to the claim’s honesty, and what does it not add?

Answer. The residual adds the named limitations — the register’s entries and the CI’s upper bound — and it does not add the limitations the model cannot name, which are the unknown unknowns. The register’s residual column is the known unknowns’ record; the unknown unknowns are outside its frame. Used in §L.A.1 to define the four quadrants’ boundary.

Q2. In 4.4.2 §L.C.3, the kill criterion was the proposal’s falsifiability’s pre-commitment. What is the communication’s analogue of the kill criterion, and why is it the claim’s falsifier rather than the proposal’s threshold?

Answer. The analogue is the “what would change my mind” statement — the claim’s falsifier — and it is the claim’s rather than the proposal’s because the claim is what the committee accepts, and the proposal is what the operator funds. The falsifier’s is the claim’s honesty’s, and the threshold’s is the proposal’s discipline’s. Used in §L.C.1 to define the reversal test.

Q3. In 4.3.3 §L.B.4, the extrapolation’s contract had three clauses — stationarity, recovery’s independence, and reset’s completeness. What is the communication’s relationship to the contract’s clauses?

Answer. The contract’s clauses are the known unknowns’ — they are named, and their violation is measurable — and the unknown unknowns are the contract’s clauses the analysis did not think to state: the phenomena the model’s form cannot represent. The communication’s is the clauses’ statement and the enumeration of what the clauses’ frame omits. Used in §L.A.3 to define the structural sources’ relationship to the contract.

Q4. In 4.2.4 §L.E.3, the register’s six entries were the known unknowns’ record, and the safety case’s strength was the residuals’ acceptance. What does the committee’s evaluation add to the operator’s acceptance, and why are the two different audiences?

Answer. The operator’s acceptance is the risk’s acceptance — the residual’s is what the operator lives with; the committee’s evaluation is the argument’s validity — the residual’s is what the committee tests. The two audiences’ needs differ: the operator needs the residual’s magnitude and the mitigation’s plan; the committee needs the falsifier and the structural sources. Used in §L.B.1 to define the three audiences.

Part A — The four quadrants and the structural sources

§L.A.1 — The four quadrants

The four quadrants are the classification of what the agent’s evidence does and does not cover, and each quadrant’s communication form is different.

Known knowns. The measured metrics — the 4.1.3 suite’s success rate, the 4.2.4 register’s violation counts, the 4.3.3 horizon’s survival. Their communication form is the evidence’s table — the number, the CI, the suite’s version, and the date. The known knowns’ are the argument’s load-bearing evidence, and their communication’s discipline is the 4.3.1 §L.C.4’s lower’s bound.

Known unknowns. The failure modes the model can name — the register’s residuals, the CIs’ width, the extrapolation’s contract’s clauses. Their communication form is the register’s entry — the failure, the probability, the consequence, the mitigation, and the residual. The known unknowns’ are the argument’s honest limits, and their communication’s discipline is the 3.3.4 §L.E.3’s five fields.

Unknown unknowns. The failure modes the model’s form cannot represent — the structural blind spots. Their communication form is the reversal statement — the “if I’m wrong, here’s how I’d know” and the structural sources’ enumeration. The unknown unknowns’ are the argument’s boundary, and their communication’s discipline is §L.C’s.

Unknown knowns. The knowledge the team has but has not written — the operator’s tacit skill, the engineer’s hard-won intuitions, the debugging’s folklore. Their communication form is the handoff’s narrative — the runbook, the decision log, the “what the code doesn’t say” document. The unknown knowns’ are the argument’s succession, and their communication’s discipline is §L.D.2’s.

The four quadrants’ discipline’s is the match’s: each quadrant’s communication form is the form’s that exposes what the quadrant covers and what it omits. The known knowns’ table exposes the measured; the known unknowns’ register exposes the named; the unknown unknowns’ reversal exposes the unmodelled; the unknown knowns’ narrative exposes the unwritten.

Self-explanation 1. Why is the unknown knowns’ quadrant the one most often omitted from a defence, and why is the omission costly?

Answer. Because the unknown knowns are the team’s tacit knowledge, and the team does not experience it as knowledge — it experiences it as obvious. The omission is costly because the successor’s onboarding is the reconstruction of the tacit knowledge from the artefacts, and the reconstruction’s cost is the hidden part of the handoff.

§L.A.2 — The register and the unknown unknowns’ boundary

The register’s residual column is the known unknowns’ record, and the unknown unknowns’ boundary is the register’s frame’s limit.

The boundary’s structure. The register’s entries are the failure modes the 4.2.4 §L.A.1’s hazard analysis named — the six hazards and their residuals — and the register’s frame is the hazard analysis’s completeness’s assumption. The unknown unknowns are the hazards the analysis did not name, and the analysis’s completeness is the assumption’s that the communication must state.

The boundary’s five structural sources. The unknown unknowns are not random; they come from five identifiable structural sources, and each’s communication is the source’s enumeration.

The structure’s assumptions. The model’s form — the 2.3.3 §L.B.2’s contact law, the 3.4.1 §L.A.1’s dynamics’ form, the 4.2.1 §L.B.2’s Gaussian’s noise — encodes assumptions the data did not test. A phenomenon the form cannot represent — the joint’s flexibility, the contact’s adhesion, the camera’s rolling shutter — is the source’s instance.

The interaction’s effects. The subsystems’ composition produces failure modes no subsystem’s test covers — the 4.2.2 §L.B.3’s handoff’s discontinuity, the 4.2.3 §L.D.2’s re-query’s oscillation, the 4.3.2 §L.D.2’s policy-sim pair’s confounding. The interaction’s source is the composition’s emergent behaviour.

The distribution’s shift. The deployment’s states outside the suite’s distribution — the 4.1.3 §L.C.3’s distribution’s statement’s edge, the 4.3.3 §L.B.4’s contract’s clauses’ violation, the 3.4.4 §L.D.1’s MMD’s rise. The source’s is the unvisited states.

The specification’s gap. The operator’s intent beyond the stated instruction — the 4.2.3 §L.C.1’s success predicate’s incompleteness, the 3.3.3 §L.D.2’s semantic gap’s residual. The source’s is the unstated intent.

The scale’s emergence. The behaviour that appears only at the deployment’s volume — the 4.3.3 §L.A.2’s drift’s accumulation, the 4.2.1 §L.C.3’s check’s false alarm’s rate’s saturation, the 4.2.4 §L.B.3’s hysteresis’s effectiveness’s decay at the deployment’s cadence. The source’s is the quantity’s effect.

The five sources are the communication’s content. The unknown unknowns’ statement is not “there are things we don’t know”; it is “the five structural sources are X, Y, Z, W, V, and the running case’s instances of each are A, B, C, D, E”.

§L.A.3 — The running case’s five sources

The running case’s instances, one per source.

The structure’s assumption. The 2.3.3 §L.B.2’s contact law is a penalty’s model with a linear’s damping; the real contact’s adhesion — the rubber pad’s tackiness, which the model’s form omits — is the phenomenon the form cannot represent. The instance’s is the adhesion’s at the 5 N grip’s long’s dwell.

The interaction’s effect. The 4.2.2 §L.B.3’s task-to-motion handoff and the 4.2.3 §L.D.2’s re-query’s loop: when the handoff fires at the 4.2.2 §L.B.4’s threshold, the re-query’s 4.2.3 §L.D.2’s monitor can fire concurrently, and the two’s sequence’s is the emergent’s — the 4.2.2 §L.C.3’s arbitration’s rules did not anticipate the concurrent’s trigger’s. The instance’s is the handoff’s during the re-query’s dwell.

The distribution’s shift. The 4.1.3 §L.C.3’s suite includes’s the running case’s table’s surface’s and the distractors’s; the deployment’s other table’s — a glossy surface the suite did not include — is the unvisited’s state. The instance’s is the reflective table’s pose’s failure.

The specification’s gap. The 4.2.3 §L.C.1’s success predicate’s is the cube’s pose’s and the orientation’s within the bin’s opening; the operator’s intent’s “don’t knock the bin” is not in the predicate. The instance’s is the bin’s displacement’s during the place’s.

The scale’s emergence. The 4.2.4 §L.B.3’s hysteresis’s two-cycle’s false alarm’s rate is the per-invariant’s; at the deployment’s parallel’s monitors — the four invariants’s running’s concurrently’s — the joint’s false alarm’s rate is the union’s, and the operator’s trust’s erosion’s is the emergent’s. The instance’s is the operator’s disabling’s of the I2’s monitor at the deployment’s scale.

The five instances’ communication’s is the reversal statement’s content, and each instance’s falsifier’s is the observation that would confirm’s it.

Concept check. A defence’s unknown unknowns’ statement is “we can’t anticipate everything”. Is the statement a communication of the unknown unknowns?

Answer. No; the statement is a platitude’s — it names no source, no instance, and no falsifier. The communication’s is the five sources’ enumeration with the running case’s instances and the falsifiers, per §L.A.3. [displaces M1]

Worked-example ladder.

WE-L.A.a (fully worked). The running case’s structure’s assumption’s instance is the adhesion’s at the 5 N dwell. Write the falsifier’s observation.

Solution. The falsifier’s is “if the gripper’s release’s requires’s a peel’s force above’s the 5 N grip’s friction’s budget’s on a dwell’s above’s 30 s’s, the adhesion’s is the unmodelled’s phenomenon, and the contact’s law’s form’s is the gap’s”. The observation’s is the release’s failure’s at the dwell’s, and the observation’s confirms’s the source’s instance.

WE-L.A.b (completion). The distribution’s shift’s instance is the reflective’s table. Write the falsifier’s and the monitoring’s channel.

Completion. The falsifier’s is “if the pose’s failure’s rate’s on the reflective’s table’s exceeds’s the 4.1.3 §L.C.3’s threshold’s by 3בs, the shift’s is the unvisited’s state’s”. The monitoring’s channel’s is the 4.2.1 §L.B.3’s cross-modal’s check’s and the 4.1.3 §L.D.2’s drift’s monitor’s.

WE-L.A.c (faded). Write the scale’s emergence’s instance’s falsifier’s and the mitigation’s.

Faded step. The falsifier’s is “if the four’s invariants’s joint’s false alarm’s rate’s at the deployment’s cadence’s exceeds’s the operator’s tolerance’s, the hysteresis’s per-invariant’s rate’s is the emergent’s”. The mitigation’s is the 4.4.2 §L.A.1’s proposal’s — the joint’s false alarm’s rate’s monitor’s.

Completion. The mitigation’s is the joint’s rate’s monitor’s and the hysteresis’s joint’s design’s, and the register’s entry’s is the 4.4.2’s portfolio’s.

WE-L.A.d (contrastive near-miss). A team’s unknown unknowns’ statement names’s the five sources but no instances’s and no falsifiers’s. The committee’s asks’s “which’s instance’s, and what’s would confirm’s it?”. The team’s cannot’s answer’s. Why it fails: the sources’ enumeration’s without the instances’s and the falsifiers’s is the category’s without the content’s. The fix: the five instances and the five falsifiers, per §L.A.3.

Comparative table: the four quadrants.

Design principle P62.1. Communicate the four quadrants by their forms; the unknown unknowns are the five structural sources’ instances and falsifiers, not the platitude of unanticipatable risk.

Consolidation checkpoint 1

(a) Three capabilities. You can name the four quadrants and their communication forms. You can name the five structural sources and give a running-case instance of each. You can write the falsifier’s observation for an instance.

(b) Transfer probe. The interaction’s effect’s instance is the handoff’s during the re-query’s dwell. What is the falsifier’s? Answer: the falsifier is “if the concurrent’s triggers’s produce’s the 4.2.2 §L.C.3’s arbitration’s deadlock’s or the 4.2.3 §L.D.2’s re-query’s oscillation’s at a rate above the 4.1.3 §L.C.3’s threshold’s, the interaction’s is the emergent’s”. The observation’s confirms’s the instance.

(c) Remediation rule. If you cannot state why the unknown knowns are the most often omitted, re-read §L.A.1 before Part B.

Part B — The three audiences and the four forms

§L.B.1 — The three audiences’ needs

The communication has three audiences, and each one’s need is different.

The operator. The operator lives with the system’s risk, and the operator’s need is the residual’s acceptance — the residual’s magnitude, the mitigation’s plan, and the escalation’s path. The operator’s question is “what is the risk I am accepting, and what happens when the residual fires?”. The operator’s communication form is the register’s entry and the safety case’s residual column.

The committee. The committee judges the argument’s validity, and the committee’s need is the falsifiability’s argument — the claim, the evidence, the warrant, and the falsifier. The committee’s question is “what would make this claim wrong, and would you know?”. The committee’s communication form is the claim-evidence-warrant table and the reversal statement.

The successor. The successor inherits the system, and the successor’s need is the handoff’s map — the runbook, the decision log, and the tacit knowledge’s narrative. The successor’s question is “what do I need to know that is not in the code?”. The successor’s communication form is the handoff’s narrative.

The three audiences’ needs are not substitutable. The register’s entry that satisfies the operator’s is insufficient for the committee’s — the committee needs the falsifier’s — and the reversal statement that satisfies the committee’s is insufficient for the successor’s — the successor needs the runbook’s. The communication’s design is the three forms’ joint, and each audience’s form is the quadrant’s that the audience’s need matches.

§L.B.2 — The four forms’ match to the audiences

The four forms’ match to the three audiences is the communication’s structure, and the match is not arbitrary.

The evidence’s table → the committee’s validity’s question. The evidence’s table’s is the known knowns’, and the committee’s validity’s question’s is the table’s provenance’s — the suite’s version’s, the CI’s, the confounds’ exclusion’s. The table’s is the committee’s evidence’s, and its form’s is the 4.4.1 §L.B.2’s hierarchy’s primary’s.

The register’s entry → the operator’s acceptance’s question. The register’s entry’s is the known unknowns’, and the operator’s acceptance’s question’s is the entry’s five fields’ — the failure’s, the probability’s, the consequence’s, the mitigation’s, the residual’s. The entry’s is the operator’s risk’s statement’s, and its form’s is the 3.3.4 §L.E.3’s.

The reversal statement → both the committee’s and the operator’s. The reversal’s is the unknown unknowns’, and it serves the committee’s falsifiability’s question’s and the operator’s residual’s question’s. The reversal’s is the bridge between the two audiences, and its form’s is the §L.C’s.

The handoff’s narrative → the successor’s. The handoff’s is the unknown knowns’, and it serves the successor’s inheritance’s question’s. The narrative’s is the successor’s map’s, and its form’s is the §L.D.2’s.

The match’s discipline’s. Each form’s is the audience’s need’s expression’s, and the communication’s completeness’s is the four forms’ presence’s — a defence’s with the evidence’s table’s and the register’s entry’s but no reversal’s is the committee’s falsifiability’s gap’s, and a defence’s with the reversal’s but no handoff’s is the successor’s inheritance’s gap’s.

§L.B.3 — The audience’s question as the form’s design

The audience’s question is the form’s design’s criterion, and the question’s specificity’s is the form’s content’s.

The operator’s question’s form’s. “What is the residual’s, and what happens’s when the residual’s fires’s?” The register’s entry’s answer’s is the residual’s magnitude’s, the mitigation’s plan’s, and the escalation’s path’s — the 4.2.4 §L.C.2’s operator’s notification’s. The entry’s completeness’s is the escalation’s path’s presence’s.

The committee’s question’s form’s. “What would make’s the claim’s wrong’s, and would you know’s?” The reversal’s answer’s is the falsifier’s observation’s, the monitoring’s channel’s, and the consequence’s — the 4.3.2 §L.C.3’s stopping’s rule’s. The answer’s completeness’s is the monitoring’s channel’s presence’s.

The successor’s question’s form’s. “What do I need’s to know’s that’s not’s in the code’s?” The handoff’s answer’s is the runbook’s — the failures’s the code’s does not handle’s — the decision’s log’s — the choices’s the code’s embodies’s — and the tacit’s narrative’s. The answer’s completeness’s is the decision’s log’s rationale’s presence’s.

The questions’ role’s. The questions’ are the audiences’ entry’s into the communication’s, and the forms’ are the answers’. The communication’s design’s is the questions’ enumeration’s and the forms’ matching’s, and the test’s is the audience’s — the operator’s acceptance’s, the committee’s judgment’s, and the successor’s independence’s.

Self-explanation 2. Why is the committee’s question “what would make the claim wrong, and would you know?” the hardest of the three?

Answer. Because the question’s second’s clause’s — “would you know’s?” — is the monitoring’s channel’s test’s, and the channel’s is the 4.3.2 §L.D.2’s confound’s detection’s and the 4.2.1 §L.B.3’s checks’s. The first’s clause’s is the falsifier’s statement’s; the second’s is the empirical’s machinery’s, and the machinery’s is the hard’s part’s.

Concept check. A defence’s operator’s communication’s is the register’s six entries, and the committee’s communication’s is the same’s six entries. Which audience’s need’s is unmet’s?

Answer. The committee’s — the entries are the known unknowns’ and the committee’s question’s is the falsifiability’s, which requires the reversal’s statement’s, not the register’s entries’s. The mechanism’s is the audience’s question’s difference’s, and the fix’s is the reversal’s statement’s. [displaces M2]

Worked-example ladder.

WE-L.B.a (fully worked). Write the operator’s communication’s for the H3’s drop’s residual’s.

Solution. The register’s entry’s: the H3’s drop’s, probability’s moderate, consequence’s the cube’s fall’s, mitigation’s the 2.3.3 §L.E.2’s adaptive grip’s and the 4.2.4 §L.C.1’s abort’s, residual’s the slip’s onset’s 12 ms’s excess’s. The escalation’s path’s: the 4.2.4 §L.C.2’s operator’s notification’s and the safe’s state’s. The operator’s acceptance’s is the residual’s 12 ms’s and the mitigation’s plan’s.

WE-L.B.b (completion). Write the committee’s communication’s for the same’s H3’s residual’s.

Completion. The claim’s: the H3’s drop’s residual’s is 12 ms’s at the 4.2.4 §L.B.1’s I3’s threshold’s. The evidence’s: the 4.2.4 §L.D.2’s injection’s suite’s recovery’s rate’s and the 3.4.1 §L.C.1’s twin-track’s. The warrant’s: the momentum’s observer’s 3 ms’s detection’s plus the abort’s 5 ms’s first’s action’s plus the degrade’s 50 ms’s is 58 ms’s, within’s the 100 ms’s hazard’s. The falsifier’s: “if the drop’s rate’s in the field’s exceeds’s the 4.2.4 §L.C.3’s composed’s rate’s prediction’s by 2בs, the residual’s is the slip’s onset’s not’s the mitigation’s, and the 4.2.4 §L.B.1’s I3’s is the gap’s”.

WE-L.B.c (faded). Write the successor’s communication’s for the H3’s residual’s.

Faded step. The runbook’s: the slip’s detection’s is the observer’s residual’s above the 4 Nm’s, and the abort’s is the 4.2.4 §L.C.1’s release’s script’s. The decision’s log’s: why the adaptive grip’s was chosen’s over the fixed 5 N’s — the 2.3.3 §L.E.2’s friction’s interval’s. The tacit’s narrative’s: the operator’s observation’s that the slip’s is audible’s before the observer’s fires’s.

Completion. The handoff’s tacit’s narrative’s is the operator’s audible’s slip’s cue’s — the knowledge’s that’s not’s in the code’s — and the runbook’s is the observable’s trigger’s. The two’s are the handoff’s content’s.

WE-L.B.d (contrastive near-miss). A defence’s operator’s communication’s is the evidence’s table’s — the success’s 0.898’s — and the operator’s acceptance’s is the register’s need’s. Why it fails: the evidence’s table’s is the known knowns’ and the operator’s question’s is the known unknowns’ — the residual’s and the escalation’s. The table’s is insufficient’s for the operator’s. The fix: the register’s entries’s and the escalation’s paths’s, per §L.B.2.

Comparative table: the audiences and the forms.

Design principle P62.2. Match the form to the audience’s question; the evidence’s table, the register’s entry, the reversal’s statement, and the handoff’s narrative are the four forms, and each audience’s need is the form’s completeness’s test.

Consolidation checkpoint 2

(a) Three capabilities. You can name the three audiences and their questions. You can match the four forms to the audiences. You can write the operator’s, the committee’s, and the successor’s communications for a residual.

(b) Transfer probe. The successor’s question is “what do I need to know that is not in the code?”. Which form answers it, and what is the completeness’s test? Answer: the handoff’s narrative, and the completeness’s test is the decision’s log’s rationale — the why of the choices the code embodies.

(c) Remediation rule. If you cannot state why the committee’s question is the hardest, re-read §L.B.3 before Part C.

Part C — The reversal test and the narrative

§L.C.1 — The reversal test

The reversal test is the claim’s falsifiability’s operational’s form: the claim is honestly communicated only if the speaker can state the observation that would make them abandon it.

The test’s structure.

where $F$ is the observation’s that contradicts $C$, $M$ is the mechanism’s by which the observation’s is collected, and $K$ is the action’s the observation’s triggers’s — the claim’s abandonment’s or the re-scope’s.

The test’s three’s properties’s. A well-formed falsifier’s is:

Observable. The falsifier’s is an observation’s — a measurement’s, not a speculation’s. The observation’s is the 4.3.2 §L.D.2’s confound’s check’s or the 4.3.2 §L.C.3’s stopping’s rule’s.

Discriminating. The falsifier’s separates’s the claim’s from the alternatives’s — the 4.3.2 §L.B.1’s swap’s. A falsifier’s that’s consistent’s with the alternatives’s is not’s discriminating’s.

Pre-committed’s. The falsifier’s threshold’s is the 4.4.2 §L.C.3’s kill’s — stated’s before’s the observation’s. A falsifier’s chosen’s after’s the observation’s is the post hoc’s and is not’s the test’s.

The reversal test’s distinction’s from the hedge’s. The hedge’s is “there may’s be some risk’s” — the unfalsifiable’s. The reversal’s is “if the observation’s is X’s, the claim’s is wrong’s, and the channel’s is Y’s, and the action’s is Z’s” — the falsifiable’s. The hedge’s hides’s the uncertainty’s; the reversal’s exposes’s it’s.

§L.C.2 — The running case’s central claim’s reversal

The running case’s central claim’s — the 4.4.1 §L.E.1’s success’s rate’s at 0.88’s — with the reversal’s four’s elements.

The claim’s $C$. “The success’s rate’s is at least 0.88’s at the 95% lower’s bound’s on the 4.1.3 §L.E.1’s seen-task’s axis’s at the 400-episode’s suite’s.”

The falsifier’s $F$. The observation’s that the paired’s deployment’s episodes’s — the 4.2.4 §L.C.2’s shadow’s deployment’s — give a success’s rate’s below’s the 0.88’s lower’s bound’s at the same’s suite’s version’s and the same’s CI’s level’s.

The monitoring’s channel’s $M$. The 3.1.4 §L.A.1’s shadow’s deployment’s and the 4.1.3 §L.D.2’s drift’s monitor’s. The channel’s is the deployment’s measurement’s, not’s the suite’s.

The consequence’s $K$. The claim’s re-scope’s to the deployment’s measured’s rate’s, and the 4.4.2 §L.E.3’s fallback’s — the 4.4.2 §L.D.3’s improvement’s is the next’s action’s.

The reversal’s statement’s. “The claim’s is the 0.88’s at the suite’s; the falsifier’s is the deployment’s paired’s measurement’s below’s the 0.88’s at the same’s CI’s; the channel’s is the 3.1.4 §L.A.1’s shadow’s deployment’s with the 4.1.3 §L.D.2’s drift’s monitor’s; the consequence’s is the re-scope’s to the measured’s rate’s and the 4.4.2 §L.E.3’s fallback’s.”

The reversal’s test’s. The claim’s is falsifiable’s — the deployment’s measurement’s is observable’s, the channel’s is named’s, and the consequence’s is pre-committed’s. The reversal’s is the claim’s honesty’s.

§L.C.3 — The honest uncertainty and the hedge

The honest uncertainty and the hedge are different communication’s, and the difference’s is the specificity’s.

The hedge’s signature’s. The vague’s quantifier’s — “some’s”, “may’s”, “possibly’s” — the unnamed’s assumption’s, the absent’s falsifier’s, the absent’s channel’s. The hedge’s is the unfalsifiable’s dressed’s as caution’s.

The honest’s signature’s. The interval’s — “0.873’s to 0.923’s at the 95% level’s” — the named’s assumption’s — “the suite’s is the deployment’s distribution’s sample’s, conditional’s on the 4.1.3 §L.C.3’s three preconditions’s” — the stated’s falsifier’s — “if the deployment’s paired’s measurement’s is below’s 0.88’s” — and the named’s channel’s.

The two’s consequence’s. The hedge’s is the committee’s distrust’s — the hedge’s hides’s the uncertainty’s, and the committee’s discovers’s the hiding’s in the questions’s. The honest’s is the committee’s trust’s — the honest’s exposes’s the uncertainty’s, and the committee’s rewards’s the exposure’s.

The honest’s discipline’s. The honest’s is not’s the pessimism’s — the interval’s upper’s bound’s is stated’s — and it is not’s the false’s modesty’s — the evidence’s is stated’s. The honest’s is the complete’s statement’s: the point’s, the interval’s, the assumption’s, the falsifier’s, and the channel’s.

Self-explanation 3. Why is the honest uncertainty’s interval’s upper’s bound’s part of the honesty’s?

Answer. Because the honesty’s is the complete’s statement’s, and the upper’s bound’s is the evidence’s — the claim’s is not’s stronger’s than’s the evidence’s. The hedge’s omits’s the upper’s; the honest’s states’s it’s, and the committee’s evaluates’s the claim’s at the lower’s bound’s and credits’s the upper’s as the evidence’s ceiling’s.

Concept check. A defence’s uncertainty’s statement’s is “the success’s rate’s is around’s 0.90’s, give or take’s”. Is the statement’s the honest’s or the hedge’s?

Answer. The hedge’s — “around’s” and “give or take’s” are the vague’s quantifiers’s, and the interval’s, the assumption’s, the falsifier’s, and the channel’s are absent’s. The fix’s is the interval’s $[0.873, 0.923]$‘s, the assumption’s, the falsifier’s, and the channel’s. [displaces M3]

Worked-example ladder.

WE-L.C.a (fully worked). Apply the reversal test to the H1’s force’s residual’s claim’s: “the I1’s force’s limit’s is enforced’s at the 20 N’s”.

Solution. The claim’s is the enforcement’s. The falsifier’s is the observation’s of a force’s above’s 20 N’s at the 2.3.3 §L.E.2’s momentum’s observer’s beyond’s the noise’s 3 σ’s. The channel’s is the 2.3.3 §L.E.2’s observer’s and the 4.2.4 §L.B.1’s I1’s monitor’s. The consequence’s is the 4.2.4 §L.C.1’s abort’s and the 4.4.2 §L.C.1’s proposal’s for the I1’s threshold’s. The reversal’s is well-formed’s — the observation’s, the channel’s, and the consequence’s are named’s.

WE-L.C.b (completion). Apply the reversal test to the unknown unknowns’ structure’s instance’s — the adhesion’s at the 5 N dwell.

Completion. The claim’s is the contact’s law’s form’s sufficiency’s. The falsifier’s is the release’s peel’s force’s above’s the 5 N’s friction’s budget’s on the dwell’s above’s the 30 s’s. The channel’s is the 2.3.3 §L.E.2’s momentum’s observer’s on the release’s, and the consequence’s is the 3.4.3 §L.B.3’s contact’s model’s extension’s and the 4.4.2 §L.C.1’s proposal’s. The completion’s shows the unknown unknowns’ reversal’s is the structural’s claim’s falsifier’s.

WE-L.C.c (faded). Write the honest uncertainty’s statement’s for the recovery’s 0.809’s rate’s.

Faded step. The point’s is 0.809’s; the interval’s is the 4.3.1 §L.C.1’s CI’s at the suite’s size’s; the assumption’s is the 4.2.4 §L.C.3’s composed’s rate’s is the kinds’ independence’s, conditional’s on the 4.3.3 §L.C.3’s per-segment’s compounding’s; the falsifier’s is the field’s recovery’s rate’s below’s the interval’s lower’s bound’s; the channel’s is the 4.3.3 §L.E.2’s horizon’s evidence’s.

Completion. The statement’s is the complete’s — the point’s, the interval’s, the assumption’s, the falsifier’s, and the channel’s — and the hedge’s would’s omit’s the interval’s and the channel’s.

WE-L.C.d (contrastive near-miss). A defence’s falsifier’s is “if the success’s rate’s turns’s out to be lower’s than we think’s”. Why it fails: the falsifier’s is not’s observable’s (no’s threshold’s, no’s measurement’s), not’s discriminating’s (no’s alternative’s excluded’s), and not’s pre-committed’s (no’s threshold’s before’s the observation’s). The fix: the reversal’s four’s elements’s, per §L.C.1.

Comparative table: the honest and the hedge.

Design principle P62.3. Test every claim’s reversal; the well-formed falsifier is observable, discriminating, and pre-committed, and the honest uncertainty is the complete statement whose interval’s upper bound is the evidence’s ceiling.

Consolidation checkpoint 3

(a) Three capabilities. You can apply the reversal test to a claim and state the falsifier’s three properties. You can write the running case’s central claim’s reversal. You can distinguish the honest uncertainty from the hedge by their signatures.

(b) Transfer probe. A claim’s falsifier is “the deployment’s success is below 0.85 at the 800-episode shadow”. Is it observable and discriminating? Answer: observable — the 800-episode shadow’s measurement is a channel; discriminating — the 0.85 threshold separates the claim from a weaker claim; and pre-committed if the 0.85 was stated before the shadow. The three’s properties hold.

(c) Remediation rule. If you cannot state why the hedge hides the uncertainty, re-read §L.C.3 before Part D.

Part D — The closing and the handoff

§L.D.1 — The closing’s three sentences

The defence’s closing has three sentences, and each one’s role’s is the audience’s entry’s.

The first’s sentence: the claim. “The agent achieves’s X, at the scope’s Y, with the evidence’s Z.” The first’s is the claim’s restatement’s — the scope’s and the evidence’s bound’s — and it is the committee’s anchor’s.

The second’s sentence: the residual. “The residual’s is R, and the risk’s is accepted’s by the operator’s, with the mitigation’s M and the escalation’s E.” The second’s is the residual’s statement’s — the operator’s acceptance’s and the committee’s honesty’s — and it is the defence’s most’s distinctive’s sentence’s.

The third’s sentence: the invitation. “The claim’s is falsifiable’s by F, at the channel’s C, and the next’s question’s is Q.” The third’s is the invitation’s — the falsifier’s and the next’s question’s — and it is the committee’s entry’s and the 4.4.2’s proposals’s forward-looking’s.

The three’s discipline’s. The three’s are the defence’s last’s 30 seconds’s, and they are the memory’s the committee’s leaves’s with. The closing’s is not’s the summary’s; it is the claim’s, the residual’s, and the invitation’s — the three’s sentences’s the committee’s repeats’s to itself’s.

Self-explanation 1. Why is the second sentence — the residual — the defence’s most distinctive?

Answer. Because the first’s is the claim’s and the third’s is the invitation’s, and the second’s is the honesty’s — it is the sentence the committee’s repeats’s to itself’s as the evidence’s that the candidate’s knows the limits. The second’s is the sentence that differentiates the defence’s from the demo’s.

§L.D.2 — The handoff’s four artefacts

The handoff’s has four artefacts, and each carries what the code does not.

The runbook. The failures the code does not handle — the 4.2.4 §L.C.2’s escalation’s, the 4.2.1 §L.B.3’s checks’s false’s positives’s, the 3.4.4 §L.C.3’s re-fit’s trigger’s. The runbook’s is the operator’s response’s to the failures the code’s detects’s but does not resolve’s.

The decision’s log. The choices the code embodies — why the 4.2.2 §L.B.1’s four’s authorities’s, why the 2.3.3 §L.C.1’s impedance’s, why the 4.1.2 §L.A.3’s parallel-jaw’s. The log’s is the rationale’s, and its absence’s is the successor’s re-derivation’s cost’s.

The tacit’s narrative. The team’s knowledge’s that’s not’s in the runbook’s or the log’s — the operator’s audible’s slip’s cue’s, the engineer’s smell’s for the Q-shield’s miscalibration’s, the maintainer’s pattern’s for the drift’s. The narrative’s is the successor’s onboarding’s content’s, and its form’s is the story’s.

The register’s final state. The 3.3.4 §L.E.3’s entries’s and the 4.4.2 §L.E.3’s forward-looking’s entries’s. The register’s is the known unknowns’ record’s, and its final’s state’s is the successor’s baseline’s.

The four’s completeness’s. The successor’s independence’s is the test’s: the successor’s can maintain’s the system’s and extend’s it’s without the team’s, and the completeness’s is the successor’s questions’s — the what’s and the why’s — being answerable’s from the artefacts’s.

§L.D.3 — The curriculum’s synthesis

The curriculum’s synthesis is the learner’s self-assessment’s against the capstone’s defence, and its structure’s is the exit test’s.

The synthesis’s three’s questions’s.

The capability’s question. “Can I build the system?” The capability’s is the 4.2.1–4.2.4’s subsystems’s and the 4.1.3’s suite’s; the capability’s is the demonstration’s.

The understanding’s question. “Can I explain the system’s behaviour?” The understanding’s is the 4.3.1’s ablations’s, the 4.3.2’s diagnosis’s, and the 4.3.3’s horizon’s; the understanding’s is the evidence’s.

The judgment’s question. “Can I communicate the system’s limits and the plan?” The judgment’s is the 4.4.1’s defence’s, the 4.4.2’s proposals’s, and the 4.4.3’s quadrants’s; the judgment’s is the mastery’s.

The three’s joint’s. The curriculum’s mastery’s is the three’s joint’s, and the capstone’s defence’s is the test’s — the committee’s judges’s the three’s joint’s, and the defence’s passing’s is the curriculum’s exit’s.

The synthesis’s discipline’s. The synthesis’s is not’s the recitation’s — the curriculum’s facts’s are not’s the point’s — and it is the judgment’s: the learner’s knows’s what the system’s can’t’s do, why’s, and what’s to do’s about’s it’s. The judgment’s is the mastery’s, and its communication’s is this lesson’s.

Concept check. The learner’s can build the system’s and cannot’s explain’s why’s the 4.3.1’s ablation’s gave’s the observed’s result’s. Which of the synthesis’s three’s questions’s is unmet’s, and what’s is the fix’s?

Answer. The understanding’s — the capability’s is met’s, and the judgment’s cannot’s be met’s without the understanding’s. The fix’s is the 4.3.1’s registry’s re-examination’s and the 4.3.2’s diagnosis’s chain’s, and the committee’s asks’s the understanding’s. [displaces M4]

Worked-example ladder.

WE-L.D.a (fully worked). Write the closing’s three sentences for the running case’s central claim.

Solution. First: “the agent’s achieves’s a success’s rate’s of at least 0.88’s at the 95% lower’s bound’s on the 4.1.3 §L.E.1’s seen-task’s axis’s at the 400-episode’s suite’s.” Second: “the residual’s is the 0.90’s target’s gap’s, accepted’s by the operator’s with the 4.4.2 §L.E.3’s multi-seed’s proposal’s and the 3.3.4 §L.E.3’s escalation’s.” Third: “the claim’s is falsifiable’s by the deployment’s paired’s measurement’s below’s 0.88’s, at the 3.1.4 §L.A.1’s shadow’s deployment’s and the 4.1.3 §L.D.2’s drift’s monitor’s, and the next’s question’s is the multi-segment’s horizon’s evaluation’s.”

WE-L.D.b (completion). The committee’s asks’s “what’s would you do’s with more’s time’s?”. Write the answer’s as the invitation’s extension’s.

Completion. The answer’s is the 4.4.2 §L.E.2’s multi-segment’s horizon’s and the 4.3.3 §L.B.4’s extrapolation’s contract’s testing’s — the deferred’s proposal’s. The completion’s shows the invitation’s is the forward-looking’s and the proposal’s registry’s is its content’s.

WE-L.D.c (faded). Write the handoff’s decision’s log’s entry for the parallel-jaw’s gripper’s choice’s.

Faded step. The decision’s: the parallel-jaw’s over the dexterous’s. The rationale’s: the 4.1.2 §L.A.3’s decisiveness’s criterion’s and the 4.1.2 §L.A.2’s per-episode’s cost’s.

Completion. The rationale’s is the decisiveness’s — the parallel-jaw’s resolution’s at the budget’s is better’s, and the dexterous’s hand’s is the future’s iteration’s.

WE-L.D.d (contrastive near-miss). A defence’s closing’s is the claim’s repetition’s — “so the agent’s achieves’s 0.88’s, thank you’s”. Why it fails: the closing’s is the three’s sentences’s, and the residual’s and the invitation’s are absent’s. The committee’s leaves’s with the claim’s and not’s the honesty’s or the next’s question’s. The fix: the three’s sentences’s, per §L.D.1.

Comparative table: the closing’s three sentences.

Design principle P62.4. Close with the claim, the residual, and the invitation; the residual’s sentence is the defence’s most distinctive, and the handoff’s four artefacts are the successor’s independence’s.

Consolidation checkpoint 4

(a) Three capabilities. You can write the closing’s three sentences. You can name the handoff’s four artefacts and what each carries. You can state the synthesis’s three questions.

(b) Transfer probe. The handoff’s decision’s log is absent and the runbook is present. Which of the synthesis’s questions is at risk, and why? Answer: the judgment’s — the successor’s independence requires the rationale, and the runbook alone is the what’s without the why’s. The successor’s re-derivation of the choices is the cost.

(c) Remediation rule. If you cannot state why the residual’s sentence is the most distinctive, re-read §L.D.1 before Part E.

Part E — The running case

§L.E.1 — The final defence’s closing argument

The running case’s final defence, assembled from the four quadrants, the three audiences, the reversal, and the handoff.

The quadrants’ statement’s.

The three audiences’ communications’s. The operator’s is the register’s thirteen’s entries’s and the escalation’s paths’s. The committee’s is the claim-evidence-warrant’s table’s and the reversal’s statement’s. The successor’s is the runbook’s, the decision’s log’s, and the tacit’s narrative’s.

The reversal’s statement’s. The central’s claim’s is the 0.88’s at the suite’s; the falsifier’s is the deployment’s paired’s measurement’s below’s the 0.88’s; the channel’s is the 3.1.4 §L.A.1’s shadow’s and the 4.1.3 §L.D.2’s drift’s monitor’s; the consequence’s is the claim’s re-scope’s and the 4.4.2 §L.E.3’s fallback’s.

§L.E.2 — The unknown unknowns’ five instances

The running case’s five instances, with the falsifiers and the channels.

The five’s communication’s is the reversal’s statement’s extension’s, and each’s consequence’s is the 4.4.2 §L.E.3’s forward-looking’s entry’s.

§L.E.3 — The closing and the handoff

The closing’s three sentences’s.

“The agent’s achieves’s a success’s rate’s of at least 0.88’s at the 95% lower’s bound’s on the 4.1.3 §L.E.1’s seen-task’s axis’s, with the 4.2.4 register’s six’s residuals’s accepted’s by the operator’s, and the 4.3.3 shift’s survival’s 0.97’s at the 4.3.3 §L.E.1’s maintenance’s schedule’s. The residual’s is the 0.90’s target’s gap’s, accepted’s by the operator’s with the 4.4.2 §L.E.3’s multi-seed’s proposal’s and the 3.3.4 §L.E.3’s escalation’s. The claim’s is falsifiable’s by the deployment’s paired’s measurement’s below’s 0.88’s at the 3.1.4 §L.A.1’s shadow’s and the 4.1.3 §L.D.2’s drift’s monitor’s, and the next’s question’s is the multi-segment’s horizon’s evaluation’s, with the unknown unknowns’ five instances’s and their falsifiers’s in the handoff’s register’s.”

The handoff’s four artefacts’s.

The curriculum’s exit’s test’s. The defence’s passing’s is the three synthesis’s questions’s joint’s — the capability’s, the understanding’s, and the judgment’s — and the committee’s judgment’s is the test’s.

Self-explanation 2. Why is the handoff’s tacit narrative the artefact the successor most needs and the team least wants to write?

Answer. Because the narrative is the knowledge’s the team does not experience as knowledge — the operator’s cue’s, the maintainer’s pattern’s — and writing it requires the team to externalize what feels obvious. The successor’s independence depends on it, and the team’s reluctance is the unknown knowns’ characteristic.

Worked-example ladder.

WE-L.E.a (fully worked). The committee’s asks’s “what’s is the biggest unknown unknown’s?”. Write the answer’s with the source and the falsifier.

Solution. The answer’s is “the interaction’s — the 4.2.2 §L.B.3’s handoff’s and the 4.2.3 §L.D.2’s re-query’s concurrent’s trigger’s — and its falsifier’s is the deadlock or the oscillation’s above’s the 4.1.3 §L.C.3’s threshold’s at the 4.2.2 §L.D.3’s trace’s, and the consequence’s is the 4.4.2 §L.C.1’s proposal’s for the arbitration’s joint’s design’s.” The answer’s is the source’s, the falsifier’s, and the consequence’s, and it is not the platitude’s.

WE-L.E.b (completion). The committee’s asks’s “how’s would the successor know’s the 2.3.3 §L.E.2’s adaptive grip’s was chosen’s?”. Write the answer’s as the decision’s log’s entry’s.

Completion. The answer’s is the 2.3.3 §L.E.2’s friction’s interval’s and the 4.4.2 §L.B.4’s fixed 5 N’s insufficiency’s at the interval’s worst case’s. The completion’s shows the decision’s log’s rationale’s is the successor’s answer’s.

WE-L.E.c (faded). Write the operator’s acceptance’s statement’s for the thirteen’s register’s entries’s.

Faded step. The operator’s accepts’s the thirteen’s residuals’s with the 4.2.4 §L.C.2’s escalation’s paths’s and the 4.4.2 §L.E.3’s proposals’s for the four’s highest’s.

Completion. The acceptance’s is the residuals’s magnitude’s and the escalation’s paths’s, and the operator’s signature’s is the 4.2.4 §L.D.3’s safety case’s residual’s column’s.

WE-L.E.d (contrastive near-miss). The defence’s closing’s is the claim’s and the residual’s, and the invitation’s is omitted’s. The committee’s leaves’s without the next’s question’s and the falsifier’s channel’s. Why it fails: the closing’s is the three’s sentences’s, and the invitation’s is the committee’s entry’s. The fix: the third’s sentence’s, per §L.E.3.

Comparative table: the final defence’s items.

Design principle P62.5. Defend with the quadrants, the reversal, and the closing’s three sentences, and hand off with the four artefacts; the defence’s passing is the three synthesis questions’ joint, and the successor’s independence is the handoff’s test.

Consolidation checkpoint 5

(a) Three capabilities. You can state the running case’s quadrants’ statement. You can name the five unknown unknowns’ instances with their falsifiers. You can write the closing’s three sentences.

(b) Transfer probe. The committee’s asks’s “what’s would change’s your mind’s about the interaction’s source’s?”. Write the answer’s with the falsifier and the channel. Answer: the falsifier is the deadlock or the oscillation at the 4.2.2 §L.D.3’s trace above the 4.1.3 §L.C.3’s threshold; the channel is the trace’s and the 4.2.2 §L.B.3’s handoff’s log; the consequence is the 4.4.2 §L.C.1’s proposal for the arbitration’s joint design.

(c) Remediation rule. If you cannot state why the tacit narrative is the most needed and least wanted, re-read §L.E.3 before the lab.

Failure taxonomy

The table covers the three lab corruptions plus three failures not exercised in the lab: the operator’s rejection, the successor’s dependence, and the forgettable close.

The lab

Task. Build the four quadrants’ communication matrix for the running case, apply the reversal test to the five unknown unknowns’ instances, and generate the closing’s three sentences from the register and the claims. Then run a peer’s mock defence with the three audiences’ questions and diagnose each audience’s gap.

Conditions. Python 3.10+, NumPy. The register’s thirteen entries are the 4.2.4 §L.E.3’s and 4.3.3 §L.E.3’s; the five instances are the §L.E.2’s; the audiences’ questions are the §L.B.1’s. Seed all RNGs.

Controls. (i) The full four-form communication; (ii) the communication without the reversal; (iii) the communication without the handoff; (iv) the communication with the hedge instead of the honest statement.

Metrics.

Pre-registered hypotheses.

H1. The full communication’s quadrants’ coverage is 4 of 4, because the four forms are the §L.B.2’s match.

H2. The communication without the reversal has the committee’s questions’s falsifier’s gap and the defence’s plausibility’s falls.

H3. The communication without the handoff has the successor’s questions’s unanswerable and the independence’s loss.

H4. The hedge’s communication has the operator’s acceptance’s lower and the committee’s trust’s lower, because the hedge’s is the unfalsifiable’s.

Break-it sequence.

1. Use the platitude silently. “We can’t anticipate everything” passes the committee’s first question and fails the second. Silent: the sentence sounds humble.
2. Use the hedge silently. “Around 0.90” passes the operator’s first look and fails the acceptance. Silent: the number is approximately right.
3. Omit the handoff silently. The successor’s questions are unanswerable and the dependence on the team is invisible. Silent: the code runs.

Acceptance criteria.

What good looks like. A good artefact has five properties. (1) The four quadrants’ forms are separate artefacts. (2) The five instances each have the three falsifier properties. (3) The three audiences’ questions are answered by their forms. (4) The closing’s three sentences are written and delivered. (5) The break-it results are a table.

A runnable core of the quadrants, the reversal test, and the closing, seeded and annotated:

# lab_4_4_3.py — the four quadrants, the reversal, the closing
import numpy as np
np.random.seed(0)

QUADRANTS = {
    "known_known":   {"content": "0.898 lower bound", "form": "evidence table"},
    "known_unknown": {"content": "13 register residuals", "form": "register entries"},
    "unknown_unknown": {"content": "5 structural sources", "form": "reversal statement"},
    "unknown_known": {"content": "operator's cues", "form": "handoff narrative"},
}

INSTANCES = {
    "structure":     {"falsifier": "peel force > friction budget at 30 s dwell",
                      "channel": "momentum observer release"},
    "interaction":   {"falsifier": "deadlock or oscillation above threshold",
                      "channel": "4.2.2 trace"},
    "distribution":  {"falsifier": "pose failure 3x threshold on reflective table",
                      "channel": "4.2.1 cross-modal check"},
    "specification": {"falsifier": "bin displacement above tolerance",
                      "channel": "4.2.3 success predicate"},
    "scale":         {"falsifier": "joint false alarm above operator tolerance",
                      "channel": "4.2.4 joint monitor"},
}

def reversal_test(claim, falsifier, channel, consequence):
    properties = {
        "observable": bool(falsifier and channel),
        "discriminating": bool(falsifier),
        "pre_committed": bool(consequence),
    }
    return {"claim": claim, "falsifier": falsifier, "channel": channel,
            "consequence": consequence, "properties": properties,
            "well_formed": all(properties.values())}

print("quadrants' communication:")
for q, spec in QUADRANTS.items():
    print(f"  {q:16s} -> {spec['form']}")

print$"\nreversal tests:"$
for source, inst in INSTANCES.items():
    r = reversal_test(f"{source} claim", inst["falsifier"], inst["channel"],
                      f"{source} consequence")
    print(f"  {source:14s} well_formed={r['well_formed']}")

def closing(claim, residual, invitation):
    return [claim, residual, invitation]

print$"\nclosing's three sentences:"$
for i, s in enumerate(closing(
    "success >= 0.88 at 95% lower bound, seen-task axis",
    "residual: 0.90 target gap; accepted by operator; multi-seed proposal",
    "falsifier: deployment paired measurement < 0.88; channel: shadow + drift monitor"
), 1):
    print(f"  {i}. {s}")

def honest_uncertainty(point, ci, assumption, falsifier, channel):
    return {"point": point, "interval": ci, "assumption": assumption,
            "falsifier": falsifier, "channel": channel}

h = honest_uncertainty(0.898, (0.873, 0.923), "suite is the deployment sample",
                       "deployment paired < 0.88", "shadow + drift monitor")
print$"\nhonest statement's fields:", list(h.keys())$

Line 4–9 are the four quadrants’ forms; line 11–21 the five instances’ falsifiers and channels; lines 23–33 the reversal test’s three properties; lines 35–44 the closing’s three sentences; lines 46–53 the honest uncertainty’s five fields.

Cheat sheet

Exercises with solutions

E 4.4.3.1. Name the four quadrants and their communication forms.

S 4.4.3.1. Known knowns → evidence table; known unknowns → register entry; unknown unknowns → reversal statement; unknown knowns → handoff narrative. The lesson: the four quadrants’ forms are the communication’s structure, and each form exposes what its quadrant covers and omits.

E 4.4.3.2 (transfer). Name the five structural sources of the unknown unknowns and give one running-case instance.

S 4.4.3.2. Structure (the adhesion at the 5 N dwell), interaction (the handoff during the re-query’s dwell), distribution (the reflective table), specification (the bin’s displacement), scale (the joint false alarm). The lesson: the unknown unknowns are identifiable by source, and each source’s instance has a falsifier.

E 4.4.3.3. State the three audiences’ questions and the form each one’s question requires.

S 4.4.3.3. Operator: “what is the residual and what happens when it fires?” → register entry. Committee: “what would make the claim wrong and would you know?” → reversal statement. Successor: “what do I need to know that is not in the code?” → handoff narrative. The lesson: the match is not arbitrary; each audience’s question is the form’s design criterion.

E 4.4.3.4 (diagnostic). A defence’s uncertainty statement is “around 0.90, give or take”. Which signature, and what is the fix?

S 4.4.3.4. The hedge’s signature — the vague quantifier, the unnamed assumption, the absent falsifier and channel. The fix is the honest statement: the interval $[0.873, 0.923]$, the assumption, the falsifier, and the channel. The lesson: the hedge hides the uncertainty; the honest statement exposes it.

E 4.4.3.5. Write the falsifier’s three properties and test the claim “the I1 force limit is enforced at 20 N”.

S 4.4.3.5. Observable (the observer’s force above 20 N beyond 3σ), discriminating (the 20 N separates the enforcement from a weaker claim), pre-committed (the threshold stated before the observation). The falsifier’s channel is the momentum observer and the I1 monitor. The lesson: the well-formed falsifier is observable, discriminating, and pre-committed.

E 4.4.3.6 (transfer). Write the closing’s three sentences for the central claim’s 0.88 success.

S 4.4.3.6. First: the claim’s 0.88 lower bound on the seen-task axis. Second: the residual’s 0.90 gap, accepted by the operator with the multi-seed proposal and the escalation. Third: the falsifier’s deployment paired measurement below 0.88 at the shadow and drift monitor, and the next question’s multi-segment horizon. The lesson: the closing’s three sentences are the committee’s memory, and the residual’s sentence is the most distinctive.

E 4.4.3.7 (diagnostic). The successor’s questions about the parallel-jaw choice are unanswerable from the artefacts. Which artefact is missing, and what is the fix?

S 4.4.3.7. The decision’s log; the fix is the 4.1.2 §L.A.3’s decisiveness’s rationale and the 4.1.2 §L.A.2’s cost’s. The lesson: the successor’s independence is the handoff’s test, and the decision’s log is the rationale’s artefact.

E 4.4.3.8. Write the honest statement’s five fields for the recovery’s 0.809 rate.

S 4.4.3.8. Point 0.809; interval the CI at the suite’s size; assumption the kinds’ independence conditional on the 4.3.3 §L.C.3’s per-segment compounding; falsifier the field’s rate below the interval’s lower bound; channel the 4.3.3 §L.E.2’s horizon evidence. The lesson: the honest statement’s completeness is the five fields, and the hedge omits the interval and the channel.

E 4.4.3.9 (transfer). The learner can build and explain the system but cannot communicate the limits and the plan. Which of the synthesis’s questions is unmet, and what is the fix?

S 4.4.3.9. The judgment’s; the fix is the 4.4.1’s defence’s rehearsal and the 4.4.2’s proposals’s registry. The lesson: the three synthesis questions are the joint, and the judgment’s is the mastery’s.

Exit test

You have completed the curriculum when you can, without notes, defend the capstone — the claim, the evidence, the warrant, the residual, the unknown unknowns, the proposals, and the handoff — before a committee that tests the reasoning, not the demonstration. The exit test is the defence’s passing, and the defence’s passing is the three synthesis questions’ joint: the capability’s, the understanding’s, and the judgment’s.

You are ready when you can:

1. State the four quadrants and their communication forms, and give the running case’s content for each.
2. Name the five structural sources of the unknown unknowns and give an instance and a falsifier for each.
3. Match the four forms to the three audiences and state each audience’s question.
4. Apply the reversal test to a claim and state the falsifier’s three properties.
5. Write the honest uncertainty’s five fields and distinguish it from the hedge.
6. Write the closing’s three sentences — the claim, the residual, the invitation.
7. Name the handoff’s four artefacts and state what each carries that the code does not.
8. State the three synthesis questions and diagnose which one is unmet from a description.
9. Defend the running case’s central claim with the evidence, the warrant, and the residual.
10. Answer the five question categories with the four-part template.
11. State the five unknown unknowns’ instances with their falsifiers and channels.
12. State the operator’s acceptance, the committee’s judgment, and the successor’s independence as the three audiences’ tests.
13. State why the residual’s sentence is the defence’s most distinctive.
14. State why the hedge hides the uncertainty and the reversal exposes it.
15. State the curriculum’s exit test as the defence’s passing.

Cross-references. 4.4.2 supplies the improvement’s five fields, the kill criterion, and the registry’s eight fields restated in §L.A.3 and §L.D.2. 4.4.1 supplies the claim-evidence-warrant table, the vulnerability map, and the four-element claim restated in §L.A.1 and §L.B.2. 4.3.3 supplies the horizon’s failure’s law, the extrapolation’s contract, and the register’s horizon entries restated in §L.B.3 and §L.C.2. 4.3.2 supplies the diagnosis’s evidence chain and the four failure modes restated in §L.C.1. 4.2.4 supplies the register’s six entries, the safety case, and the residual’s acceptance restated in §L.B.1 and §L.E.3. 3.3.4 supplies the four-layer stack and the register’s form restated in §L.A.2. The curriculum’s exit test is the defence’s passing; the synthesis’s three questions are the capability’s, the understanding’s, and the judgment’s, and the defence’s closing argument is the lesson’s content.

The curriculum is complete. You have built the bridge (Phase 0), the robot’s modeling and control (Phase 1), the perception, learning, and manipulation (Phase 2), the foundation models and deployment (Phase 3), and the synthesis and mastery (Phase 4). The capstone’s defence is the exit test, and the defence’s passing is the mastery’s. The next step is your own: to defend the system you built, with the claim’s honesty, the evidence’s rigor, and the unknown unknowns’ communication, before the committee that will judge it. The curriculum ends here; your practice begins.