Add annotated fixed-horizon floor evaluation

manuscript/main/REVIEWER_CHECKLIST.md

@@ -145,8 +145,11 @@ retrain test-time counterfactuals.
 - [ ] Three seeds, mean ± sample standard deviation.
 - [ ] Paired cluster bootstrap by session/conversation block for 95% intervals.
 - [ ] Operating points at wearer-onset recall 0.7, 0.8, and 0.9; headline 0.8.
-- [ ] False wearer entries per multi-other opportunity minute, onset latency,
-  wearer-yield recall/latency, and original ten-horizon AP/mAP.
+- [ ] Fixed-horizon entry/INT event AP, recall-matched hard-negative
+  opportunity FPR, subtype FPR, and original ten-horizon dense AP/mAP.
+- [ ] Treat one-to-one event matching and signed latency as secondary; report
+  them only after freezing the online threshold, persistence, refractory
+  period, matching interval, and competing-risk rule.
 - [ ] Error strata: visibility, number of visible people, overlap, offscreen
   activity, and transition proxy. Add track age/fragmentation only if a
   track-aware method is selected.

manuscript/main/main.tex

@@ -33,8 +33,8 @@ not explicitly trained to expose sub-second, wearer-relative floor evidence.
 We therefore study a candidate causal local floor adapter that combines the global
 Omni state with short-window mixed audio and an anonymous set of visible-person
 observations, without assuming persistent speaker identity. On EasyCom,
-\todo{report the final three-seed dense and committed-event
-results: wearer-entry and INT-yield recall, false-positive rate, latency,
+\todo{report the final three-seed dense and fixed-horizon behavioral-event
+results: wearer-entry and INT-yield AP, recall-matched false-positive rates,
 confidence intervals, and oracle-gap recovery}. Controlled
 \todo{time-shift/remove-vision/same-rate or capacity-matched} interventions
 show that \todo{state the supported mechanism, without claiming addressee or
@@ -102,9 +102,9 @@ corresponding TODO gates pass:
   \item We evaluate a wearer-relative causal local floor adapter that combines
   long-context Omni information with short-window audio and anonymous
   visible-person evidence.
-  \item We evaluate whether the adapter reduces erroneous wearer commits
-  during multiparty floor transfers under an entry/INT event protocol with
-  explicitly annotated hard-negative regions.
+  \item We evaluate whether the adapter assigns lower erroneous-action scores
+  during multiparty floor transfers under an entry/INT protocol with point
+  positives and explicitly annotated hard-negative opportunity regions.
 \end{itemize}
 
 \begin{figure}[t]
@@ -136,11 +136,13 @@ Wearer--other overlap is allowed. The model outputs
  \in[0,1]^{K\times3}.
  \label{eq:task}
 \end{equation}
-The wearer-activity forecast also provides a causal control score. When the
-wearer is silent, $\hat y^{\mathrm w}_{t,1}$ acts as a near-term take-turn
-score; while the wearer is active, $1-\hat y^{\mathrm w}_{t,1}$ acts as a
-yield score. These scores are trained only from future activity targets.
-Committed wearer-entry and INT-yield events used for behavioral evaluation are
+The wearer-activity forecast also provides causal behavioral scores at every
+horizon. At decision time $t$, if the wearer is silent,
+$\hat y^{\mathrm w}_{t,k}$ acts as a take-turn score for future bin $k$; if
+the wearer is active, $1-\hat y^{\mathrm w}_{t,k}$ acts as a yield score.
+These scores are trained only from future activity targets and are not event
+hazards or committed controller actions.
+Adjudicated wearer-entry and INT-yield events used for behavioral evaluation are
 defined separately from participant-resolved annotations in
 Sec.~\ref{sec:experiments}.
 
@@ -229,7 +231,7 @@ the training/development pool and Sessions 1--3 only for final evaluation
 converted into two parallel representations. First, they are downsampled to
 5~Hz to construct the EgoSpeak-compatible ten-horizon targets for background,
 camera-wearer, and aggregate non-wearer activity. Second, the original
-participant timeline is retained to derive committed interaction events for
+participant timeline is retained to derive adjudicated interaction events for
 behavioral evaluation.
 
 \textbf{Behavioral event annotation.}
@@ -326,24 +328,38 @@ configuration, trainable parameter counts, and inference latency.}
 We first retain dense future-activity evaluation for direct comparison with
 EgoSpeak, reporting AP at the 0.2--2.0~s horizons and their mean mAP.
 
-Behavioral evaluation follows a committed-event protocol. For both
-wearer-entry and INT-yield, each gold event is matched to at most one committed
-prediction and is scored by recall and signed detection latency.
+Behavioral evaluation uses fixed-horizon event classification. For both
+wearer-entry and INT-yield, we align the dense forecast to each adjudicated
+item at each of the ten nominal horizons. A positive remains a single corrected
+5-Hz behavior-anchor bin: its entry score is wearer activity probability and
+its INT-yield score is one minus wearer activity probability. We do not expand
+the positive timestamp into a tolerance window. A hard-negative opportunity is
+one sample whose score is the maximum erroneous-action probability over its
+labeled half-open region at the same fixed horizon. Thus, each positive event
+or hard-negative opportunity contributes exactly one score per horizon.
 
 \textbf{False-positive scoring rule.}
-A false positive is defined by the model committing inside a labeled
-hard-negative region, regardless of whether the wearer subsequently exhibits
-an onset or offset. Thus, false-positive evaluation includes conversational
-situations in which the correct wearer behavior is to remain silent or retain
-the floor.
+A false positive is a hard-negative opportunity whose region-level score
+exceeds the frozen threshold, regardless of whether the wearer subsequently
+exhibits an onset or offset. Region max-pooling means that multiple high-scoring
+frames still count as one false-positive opportunity. Thus, false-positive
+evaluation includes conversational situations in which the correct wearer
+behavior is to remain silent or retain the floor.
 
-We report the recall--false-positive trade-off and a common operating point at
-matched recall $R=0.8$. The primary behavioral quantities are therefore entry
-false-positive rate, INT false-positive rate, and signed latency at the frozen
-operating point. We additionally report performance on the other-to-other
-subset to test whether local visual evidence specifically reduces erroneous
-wearer turn-taking during multiparty floor transfers. \todo{Freeze the FPR
-denominators and report signed median latency separately for entry and INT.}
+For every horizon we report event AP and the recall--false-positive trade-off.
+The headline behavioral operating point uses the first, nominal 0.2-s forecast
+bin at matched recall $R=0.8$; thresholds are selected independently for entry
+and INT from training-side out-of-fold predictions. We additionally report the
+complete 0.2--2.0-s curves and the other-to-other hard-negative subset to test
+whether local visual evidence specifically reduces erroneous wearer action
+scores during multiparty floor transfers. \todo{Freeze the hard-negative
+region endpoints and verify identical event denominators across methods.}
+
+A deployment-style committed-event analysis with one-to-one matching and
+signed latency is secondary and is omitted unless a threshold, persistence,
+refractory period, matching interval, and competing-risk rule are all frozen
+before test evaluation. The fixed-horizon protocol itself makes no detection-
+latency claim.
 
 
 \section{Results and Analysis}
@@ -361,9 +377,9 @@ frozen before test evaluation. \todo{Freeze all baseline implementations and
 verify identical targets, causal validity, and evaluation.}
 
 \begin{table}[t]
-\caption{Confirmatory EasyCom results. Entry and INT false-positive rates are
-measured at matched recall $R=0.8$. E/I Lat. gives signed median entry/INT
-latency in seconds; values closer to zero are better.}
+\caption{Confirmatory EasyCom results. Entry and INT false-positive rates use
+the nominal 0.2-s forecast bin at matched recall $R=0.8$; E/I AP averages each
+event family over all ten fixed horizons.}
 \label{tab:main}
 \centering
 \small
@@ -371,7 +387,7 @@ latency in seconds; values closer to zero are better.}
 \begin{tabular}{lccccc}
 \hline
 Method & 0.2-s mAP $\uparrow$ & Mean mAP $\uparrow$
-& Entry FPR@$R=.8$ $\downarrow$ & INT FPR@$R=.8$ $\downarrow$ & E/I Lat.\\
+& Entry FPR@$R=.8$ $\downarrow$ & INT FPR@$R=.8$ $\downarrow$ & E/I AP $\uparrow$\\
 \hline
 EgoSpeak reproduction       & \todo{--} & \todo{--} & \todo{--} & \todo{--} & \todo{--/--}\\
 Omni-3B frozen $G$          & \todo{--} & \todo{--} & \todo{--} & \todo{--} & \todo{--/--}\\
@@ -392,13 +408,13 @@ The gain is largest at \todo{short prediction horizons / transition-adjacent
 frames / specify actual observation}, indicating that local audiovisual
 evidence primarily improves \todo{the interpretation supported by the data}.
 
-\textbf{Committed behavior.}
+\textbf{Annotated behavior.}
 At matched wearer-entry recall $R=0.8$, the proposed adapter reduces entry FPR
 from \todo{$X$} to \todo{$Y$}; at matched INT-yield recall, it changes INT FPR
-from \todo{$A$} to \todo{$B$}. Its signed median entry/INT latencies are
-\todo{$+0.XX/+0.XX$~s}. These recall-matched error rates and signed latencies
-test whether fewer erroneous commits arise from better floor decisions rather
-than indefinite waiting.
+from \todo{$A$} to \todo{$B$}. Its ten-horizon mean entry/INT event APs are
+\todo{$E/I$}. These recall-matched opportunity error rates test whether lower
+scores on confusing floor situations are obtained without suppressing
+legitimate entry or yield events.
 
 \textbf{Comparison with adaptation baselines.}
 Matched PEFT yields \todo{result}, whereas adding recent local audio alone
@@ -544,7 +560,7 @@ for causal floor forecasting of an invisible egocentric wearer. The candidate
 method separates long-context Omni understanding from short-window local floor
 evidence without assuming persistent speaker identity. \todo{State the final
 learned-method improvement, matched-recall entry/INT false-positive rates,
-signed latency, causal-control evidence, and main limitation. If no learned
+fixed-horizon event AP, causal-control evidence, and main limitation. If no learned
 variant passes these tests, reframe the submission as an information audit or
 do not make the method claim.}