CoachOS v1.4.2 — Monthly Review & Training Sequence Semantics Refinement

CoachOS v1.4.2

CoachOS v1.4.2 is a refinement release for the Workout Sequence Intelligence Product Layer.

What This Release Means

This release clarifies the boundary between monthly adaptation judgment and workout-sequence structure.

Monthly Review owns the overall month-level position. Workout Sequence Intelligence provides supporting context about the roles individual workouts played and which roles carried the available sequence load.

The product no longer treats the most common recovery mission by workout count as proof that the entire month was an absorption month. This keeps normal build months with many easy or recovery runs from being mislabeled.

中文

CoachOS v1.4.2 是 Workout Sequence Intelligence Product Layer 的語意收斂版本。

這一版主要修正月回顧與訓練序列理解之間的判讀邊界。月回顧負責回答「這個月整體走到哪裡」，訓練序列理解則負責補充「各堂課在前後訓練中扮演什麼角色，以及哪些角色承擔了主要的可用負荷」。

過去如果恢復課在堂數上最多，頁面可能直接把整個月說成「以吸收為主的調整月」。但對多數跑者來說，恢復跑與輕鬆跑本來就常常是多數；真正推動月度負荷的，可能是少數建立能力的課程。因此這一版不再用恢復課堂數單獨決定月度位置。

Highlights

• 月回顧主判讀與訓練結構分開
• 以各序列角色的可用訓練負荷補充堂數分布
• 月度吸收判讀優先於局部訓練序列結果
• 6 月等資料不完整的月份會明確提示序列證據不足
• 月回顧與訓練序列理解的可見文字統一為中文
• 更新架構文件，正式記錄月度聚合邊界

English Summary

v1.4.2 keeps the monthly verdict owned by Monthly Adaptation and turns Workout Sequence Intelligence into supporting structure rather than a competing month-level verdict.

The release also adds available sequence-load contribution, makes incomplete sequence evidence explicit, and removes visible English role labels from the monthly sequence surface.

Release Readiness

This release is considered ready when:

• a build month with many recovery runs is not mislabeled as an absorption month
• an absorption month is not overridden by partial Build evidence
• monthly verdict and training structure are visibly distinct
• the monthly sequence surface is readable in Chinese
• existing dashboard and Narrative Engine tests continue to pass

CoachOS v1.4.1 - Repair-focused release for Workout Sequence Intelligence

CoachOS v1.4.1

CoachOS v1.4.1 is a repair and data-completeness release for Workout Sequence Intelligence.

What This Release Means

This release does not introduce a new product theme.

Instead, it makes the existing Workout Sequence Intelligence layer more trustworthy by repairing the data that feeds it:

• activity names are now recovered from FIT workout names and written back into Excel and SQLite
• average power is now backfilled from the official activity-level FIT field instead of being inferred from split values
• existing manual annotations are preserved during batch refresh and backfill flows
• older SQLite databases that were missing avg_power_w are handled safely

In other words:

v1.4.1 makes CoachOS safer to backfill, safer to refresh, and more faithful to the source FIT file.

中文

CoachOS v1.4.1 不是新的產品主題版本，而是一個以資料修補與回補完整性為核心的修正版。

這一版主要在處理兩件事。第一，過去部分活動雖然在 FIT 檔裡有活動名稱，但轉檔後沒有完整回寫到 Excel 與 SQLite，導致活動頁在切換日期時，只有部分日期會顯示名稱。現在活動名稱會優先從 FIT 裡的 workout name 回補，並寫入 Excel 與 SQLite，所以舊資料也能恢復成可辨識的活動名稱。

第二，平均功率現在改成以活動層級的官方欄位為準。FIT 檔如果有 avg_power，就會寫入 Excel 與 SQLite 的 avg_power_w；若舊資料庫原本還沒有這個欄位，平台也會先補欄位再讀取，避免因為欄位缺失而造成顯示錯誤或頁面當掉。

同時，這一版也強化了批次回補時的保護邏輯。過去已經透過標註功能填過的課表類型、訓練目的、鞋款等資訊，不應該因為 backfill 而被清掉，所以現在修補流程會採用補缺值策略，保留既有手動標註，只把缺的欄位補回來。

最後，AI 延伸分析與活動頁的文案也一起對齊。平均功率只會使用活動層級的官方值，不會把分段資料自行代算成官方平均；而原本單堂連續跑步的描述，也改成更精確的 連續跑步起點／中段／收尾，讓用詞跟資料結構一致。

如果說 v1.4.0 是把 Workout Sequence Intelligence 正式放進產品，v1.4.1 就是把這套產品真正接到可信的來源資料上。

English

CoachOS v1.4.1 is a repair-focused release for Workout Sequence Intelligence.

The goal of this release is not to expand the product theme, but to make the existing coaching layer more reliable by repairing the data pipeline underneath it.

The key improvements are:

• activity names are restored from FIT workout names and written into both Excel and SQLite
• average power is written from the official activity-level FIT field instead of being inferred from split data
• existing manual metadata is preserved during backfill and batch refresh flows
• older SQLite databases missing avg_power_w are handled safely

This makes CoachOS more faithful to the source FIT file and more stable when you revisit older activities.

Highlights

• FIT workout names now backfill activity_name into Excel and SQLite
• activity selector now shows named workouts for older dates instead of falling back to generic 跑步
• activity-level average power is now written to avg_power_w
• backfill jobs preserve existing 課表類型、訓練目的、鞋款 and other manually annotated values
• older SQLite databases are automatically extended with avg_power_w when needed
• AI handoff text now treats average power as an official activity-level field
• continuous-run wording in the activity review is now 連續跑步起點／中段／收尾
• the import and repair flow no longer clears previously saved metadata while filling missing fields

Notes

• v1.4.1 is the first repair release after Workout Sequence Intelligence Product Layer
• Workbook schema remains v1.1
• Excel remains intentionally simple; the new work is about making the existing fields trustworthy
• Average power should be read from the official activity-level field, not re-derived from split segments
• Existing manual annotations remain the source of truth when the backfill only needs to fill blanks

Release Readiness

This release is considered ready when:

• older activities show their real activity names again
• average power appears in Excel and SQLite when the FIT file provides it
• manual annotations survive backfill and refresh operations
• the dashboard no longer crashes on older databases missing avg_power_w
• AI handoff uses the same average-power convention as the activity page

CoachOS v1.4.0 CoachOS v1.4.0 — Workout Sequence Intelligence Product Layer

CoachOS v1.4.0 Release Summary

Short Summary

CoachOS v1.4.0 introduces the first product-facing Workout Sequence Intelligence layer.

This version brings sequence-aware coaching into Activity, Weekly, Monthly, and AI handoff. Instead of only asking what happened inside one run, CoachOS now explains where a workout sits in the training sequence, what the week was organized around, and what the month ultimately became.

Highlights

• Activity 新增 訓練序列理解，回答今天這堂課在整段訓練中的位置
• Weekly 新增 本週訓練重點，用教練語言整理本週序列角色
• Monthly 新增月層級 Hero，回答這個月最後建立了什麼
• AI 交棒納入 WSI，讓外部 AI 先理解序列，再延伸分析
• 資料匯入工具可批次重算 SQLite 裡的 WSI
• Excel 維持單純，不匯出 WSI
• 低信心 WSI 不再主導 Weekly / Monthly 的主敘事
• Long Run / LSD 在 rule-v1.1 中回到 建立能力

CoachOS v1.3.6 Calibration Release for Connected Coach Knowledge

CoachOS v1.3.6

CoachOS v1.3.6 是 Connected Coach Knowledge 線上的一次校準版本。這一版不在擴張新的產品主題，而是把日常使用裡最容易失真的幾個地方修正乾淨，讓資料更可信、修補更容易、介面語言更一致。

這次更新的重點

• 修正 Activity 頁最大心率顯示邏輯，明確改為 活動最高心率，不再混用個人最大心率設定
• 新增既有資料修補能力，可批次回補 SQLite 與 Excel 中的心率與欄位標籤
• 將 Critical Power (W) 明確整理為 個人臨界功率
• 統一有氧訓練效果、無氧訓練效果、訓練負荷、恢復時間、體力等欄位的繁體中文顯示
• 收斂 Activity、AI 交棒、圖卡 prompt、首頁與批次工具中的產品語言
• 將 split 相關可見文字統一整理為 分段 / 原始分段

這個版本代表什麼

如果前幾個版本是在把知識接進 CoachOS，那 v1.3.6 做的事情就是把這條鏈路校準乾淨。它不改變產品方向，但會讓每天使用這套系統時，更容易相信畫面上看到的內容，就是你真正該依靠的內容。

備註

• 這個版本仍屬於 Connected Coach Knowledge
• Workbook schema 維持 v1.1
• 這次更新重點是資料真實性、修補流程與語言一致性，不涉及新的長期 roadmap 變更

CoachOS v1.3.5 — Connected Coach Knowledge, Refined

CoachOS v1.3.5 是 Connected Coach Knowledge 的收斂版。這一版沒有改變產品主題，而是把 v1.3 已經建立的方向做得更完整、更穩定、更適合每天使用。現在，Activity 會更直接讀取課表結構來理解 WU / Main / Recovery / CD，不再只靠 raw split heuristic；已確認知識會更穩定地回流到 Weekly / Monthly 與 AI handoff；每日、每週、每月訓練圖卡 prompt 也都已經可以直接從平台交給圖像 AI。另一方面，GPS backfill 與課表結構 backfill 已可回補到 Excel 與 SQLite，Metadata / Settings 的建議值、provenance、鞋款狀態與補標註流程也更完整。整體來說，v1.3.5 的價值不在於宣布一個新方向，而是在讓 CoachOS 從「已經連起來」走向「真的能每天順手使用」。

CoachOS v1.3.5 is a refinement release for Connected Coach Knowledge. It does not introduce a new milestone theme; instead, it makes the v1.3 direction more complete, more reliable, and more usable in everyday workflow. Activity now reads workout structure more directly so WU / Main / Recovery / CD shape the interpretation instead of relying only on raw split heuristics. Confirmed knowledge flows more cleanly into Weekly / Monthly reasoning and AI handoff, and daily, weekly, and monthly training-card prompts can now be handed directly to image AI from the platform. GPS backfill and workout-structure backfill are also available for both Excel and SQLite, while Metadata / Settings has become more stable through better suggested values, provenance, shoe-status handling, and tagging flow improvements. In short, the value of v1.3.5 is not declaring a new product direction, but moving CoachOS from “connected in principle” to “usable every day.”

[superseded]CoachOS v1.3.4 Connected Coach Knowledge Refinement

This release pointed to the wrong target commit and does not reflect the intended v1.3.x product state.

Please use v1.3.5 instead.

v1.3.5 is the correct release for the refined Connected Coach Knowledge milestone, including the Activity reasoning improvements, AI handoff refinements, training-card prompts, GPS/workout-structure backfill, and related dashboard stability fixes.

這個 release 當時指到了錯誤的 target commit，因此不代表原本預期的 v1.3.x 產品狀態。

請改用 v1.3.5。

v1.3.5 才是這一波 Connected Coach Knowledge 收斂版的正確版本，包含 Activity 判讀修正、AI handoff 收斂、訓練圖卡 prompt、GPS / 課表結構 backfill，以及相關的 dashboard 穩定性修正。

CoachOS v1.3.3 — Shoe Metadata Reliability Fix

這個版本修正了鞋款設定的可靠性問題，讓鞋款分類在設定後不會因為頁面重整或同步流程而消失，也整理了重複鞋款的識別邏輯。HOKA Clifton 10 / Clifton 10 這類同鞋異名的資料，現在會收斂到同一筆 canonical shoe record，不再反覆長出重複項目。這一版也補強了鞋款設定頁的儲存流程，讓分類、服役狀態與退役日期可以穩定寫回，避免 runner 明明已經整理過鞋款，系統卻又回到未整理狀態。

This patch improves the reliability of shoe metadata handling across CoachOS. Shoe categories now persist correctly after saving and no longer get cleared by page reloads or metadata sync flows. We also fixed duplicate shoe identity handling so variants such as HOKA Clifton 10 and Clifton 10 are consolidated into a single canonical shoe record instead of being recreated repeatedly. In addition, the shoe settings workflow is now more stable, ensuring category, active status, and retire date updates are saved consistently and reflected correctly across the product.

CoachOS v1.3.2 — Coach Knowledge Reliability Update

v1.3.2 improves the reliability of Coach Knowledge across data import, Activity review, and AI follow-up workflows. This update adds GPS backfill inside the import tool, allowing CoachOS to scan a selected month, detect activities in SQLite that are missing GPS, and recover location data from matching FIT files when available. It also improves location handling in Activity so recovered GPS can be turned into readable place names, and refines the daily training card prompt so location context is more useful for AI-generated visuals without being overly detailed. In addition, this release fixes scroll-position issues across dashboard interactions, including image upload, image deletion, AI reply save flows, and other form-based edits, so returning to the same working context feels stable and uninterrupted.

v1.3.2 主要補強 Coach Knowledge 在資料匯入、Activity 判讀與 AI 延伸工作流中的穩定性與可用性。這個版本在資料匯入工具中新增了 GPS 補寫流程，讓 CoachOS 可以依月份掃描 SQLite 中缺少 GPS 的活動，並在 FIT 檔仍存在時自動回補定位資訊；若對應 FIT 已不存在，也會明確提示使用者重新下載。同時，Activity 頁現在能更穩定地把已回補的 GPS 轉成可讀地名，而每日訓練圖卡的 prompt 也進一步優化，讓交給 AI 的地點資訊更實用，但不會過度細碎。此外，這次也修正了多個頁面中的捲動位置跳回頂端問題，包含圖檔上傳、圖檔刪除、AI 回覆儲存，以及其他表單操作，讓整體使用流程更連續、更穩定。

CoachOS v1.3.0 Connected Coach Knowledge

CoachOS v1.3.0

CoachOS v1.3.0 is the release that turns confirmed Activity knowledge into connected coaching behavior across the product.

What This Release Means

This release moves CoachOS from a learning loop that only exists at the Activity surface to a connected loop that changes later coaching.

The important product shift is:

• not just storing metadata
• not just showing a confirmation state
• but making confirmed knowledge change later interpretation

In other words:

A confirmed activity changes later coaching.

中文

CoachOS v1.3.0 的主題是 Connected Coach Knowledge。這一版的重點不是再多收一組資料，而是讓跑者在 Activity 裡確認的知識，真的回流到後面的教練判讀。現在，Activity 的確認結果會影響 Weekly / Monthly 的判讀文字，Weekly / Monthly 會讀 confirmed Coach Knowledge，而不只是原始數據；Weekly / Monthly 的 AI handoff 也會一起帶上相同的知識脈絡。另一方面，Activity 標註頁會先提供建議值，讓標註更快，但仍保留明確確認，而且 provenance 會保留，讓來源看得見是手動標註還是 Coach Knowledge 建議。這代表 v1.2.0 讓 CoachOS 學會了，而 v1.3.0 則讓 CoachOS 把學到的東西連起來，真的影響後面的判讀，讓跑者不只是把資料填進系統，而是在實際教 CoachOS。

English

CoachOS v1.3.0 is themed Connected Coach Knowledge. The focus of this release is not collecting another round of data; it is making the knowledge a runner confirms inside Activity flow back into later coaching judgments. Activity confirmations now affect the reasoning text in Weekly and Monthly, which read confirmed Coach Knowledge instead of only the raw data. Weekly and Monthly AI handoff also carry the same knowledge context. On the metadata side, Activity tagging can start from suggested values so the workflow is faster, while still requiring explicit confirmation, and provenance remains visible so the app can tell whether a value came from a manual edit or a Coach Knowledge suggestion. In short, v1.2.0 taught CoachOS how to learn, and v1.3.0 connects what CoachOS learned so later coaching actually changes. That way, the runner is not just filling in fields, but genuinely teaching CoachOS.

Notes

• v1.2.0 remains the released truth for the previous milestone
• v1.3.0 is the milestone that connects confirmed knowledge to Weekly / Monthly reasoning and AI handoff
• Suggested values in metadata editing are hints, not hidden overwrites
• Coach Memory is intentionally not part of this release

Release Readiness

This release is ready when the connected loop feels stable on real data and the runner can clearly see that a confirmation changed later coaching.

CoachOS v1.2.0 – Activity Review, Raw Data, and Coach Settings

CoachOS v1.2.0

這個版本把 CoachOS 從「看活動結果」再往前推進一步，開始更完整地支援活動回顧、原始資料檢視、標注流程，以及設定治理。

這次更新重點

• 新增單堂課的 Raw Data / Garmin 細節檢視區
• 保留 split、chart、tabs 等活動細節探索流程
• 改善單堂標注與批次補標流程
• 強化課表類型、主目的、次目的的標注邏輯
• 新增更集中的設定頁，整理課表、訓練目的與回饋字典管理
• 將核心預設資料與對照來源整合進 SQLite，並支援 seed 初始化
• 改善轉檔頁流程，讓課表選擇能更自然帶入主要 / 次要訓練目的

為什麼這版重要

CoachOS 不再只是把跑步資料整理出來，而是開始支援一個更完整的流程：

回顧活動 → 看懂原始資料 → 完成標注 → 補齊知識 → 讓後續判讀更完整

補充說明

• 預設資料現在可透過 seed 初始化建立
• 歷史活動標注與設定流程已整理為更一致的使用方式
• 這版也為後續的 Coach Knowledge 演進打下基礎