A clear record of the product outcomes moving deal intelligence forward, one considered improvement at a time.
November 2026
Duplicate uploads lead straight to the existing file
When an upload is refused as a duplicate, a Choose it from the library action now opens the existing file, highlighted and pre-selected for the category being filed.
Product noteTurning a refusal into a next step resolves the user's goal rather than just reporting a problem.
Dashboard fits small phones and live updates connect reliably
Dashboard help targets no longer steal taps from neighbouring metrics, long names no longer widen the 320px layout, and realtime channels now authenticate before subscribing.
Product noteFixing touch and layout edges at the smallest supported width keeps the product dependable for people working on the move.
Kewl Ventures updates file themselves to the right deal
Update emails from linked, KV-granted senders now attach to their exact deal automatically, or create it when none exists, instead of waiting in Review as weak matches.
Product noteCalibrating the new-round guard before widening automation lets routing grow more capable without giving up control.
CI browser tests run in a pinned, self-contained image
Playwright jobs now run in the pinned official browser image, removing the runtime dependency on Ubuntu mirrors that had stalled builds for up to ten minutes.
Product noteRemoving fragile external dependencies from the delivery pipeline keeps feedback loops fast and predictable.
Write stays disabled until the IC prep writer is released
While the writer is unreleased, the Write action is disabled and stage-move auto runs are skipped, so no failed runs are recorded and the help text stays accurate.
Product noteInterfaces that only offer what the system can deliver avoid eroding user confidence.
IC prep drafts use formatted figures, full fund names and evidenced roles
IC prep now formats deal-field inputs, spells out full fund names and attributes only roles backed by evidence, fixing three content defects found on the first authorized write.
Product noteInspecting the first real output closely and correcting it quickly is how a new capability earns trust.
IC prep downloads as PDF in the partners' template
IC prep versions can now be downloaded as Short, Long or combined PDFs laid out in the partners' template, rendered from the stored version without re-running the model.
Product noteMeeting users in their established formats removes the last step between a draft and a decision.
Missing IC prep versions show a clear in-page message
Requesting an IC prep version that does not exist now shows a Version not found state with a route back to the deal, instead of a misleading deal-not-found page.
Product notePrecise error states keep users oriented and save support time.
Local test setup works on every engineering machine
The test:local machine setup, hosts-block lifecycle and recovery helpers now ship with the repository, so protected local test runs work beyond the machine they were built on.
Product noteCodifying environment setup in the repository makes safety controls repeatable rather than tribal.
Emails re-judged after a settings change refresh their suggestions and scores but always remain in Needs review, never attaching or creating deals on their own.
Product noteFailing closed on automated re-evaluation keeps people in control of consequential changes.
A committed SECURITY_MODEL.md now maps trust boundaries, components, entry points and accepted risks, giving every security scan a shared starting point.
Product noteWriting down the threat model focuses security effort on the risks that matter most.
Local test runs are isolated to their own disposable stack
The test:local wrapper now runs against a run-owned local Supabase stack with an allowlisted environment and failure-safe cleanup, so a test run cannot touch shared data.
Product noteAnswering an incident with structural isolation prevents recurrence instead of relying on care.
Rematch checks for a new funding round before auto-attaching
The deal rematch job now runs the same round relationship check as normal routing, so an email about a possible new round is not attached to the prior deal.
Product noteOne shared rule across every path prevents inconsistent outcomes.
Key results can now be flagged as watched KPIs that appear under their objective but never count toward objective or quarter scores, with scoring computed in one place.
Product noteSeparating what is monitored from what is measured keeps goals focused.
The pre-fill panel now shows how long a run has actually been going instead of fixed time promises, with clearer dashboard flag text and better spacing on small screens.
Product noteReporting what is known rather than what is hoped sets accurate expectations.
The pre-fill forecast now separates no-call-needed, nothing-to-do and held-back deals and explains each with a tooltip, alongside a clear proposal count.
Product noteExplaining why nothing will happen is as valuable as explaining what will.
Fund permissions drawer passes accessibility checks reliably
Accessibility checks on the Edit access drawer now run with reduced motion, removing an intermittent false contrast failure caused by mid-animation frames.
Product noteStable quality gates keep attention on real defects rather than flaky signals.
A deal-edit rematch can now raise an outside-your-funds hold but never lower one, so admins keep the hold, the sender record and the right-fund prefill.
Product noteEnforcing safety rules on the server makes the safe outcome the only possible outcome.
Accuracy figures count only rejects that still stand
The Accuracy chart, calibration counter and calibration sample now ignore rejects that were undone or restored, so routing quality is measured on decisions people actually kept.
Product noteMetrics stay trustworthy when they reflect the current state of a decision, not its history.
Missing document files are now flagged, not hidden
Checksum verification now records its outcome on each document version and marks files that have gone missing, turning a silent failure into a visible state.
Product noteSurfacing failures where people look is the fastest route to fixing them.
Processing, AI Settings and the Animal explainer now describe Gmail signals in plain language instead of internal jargon, while every technical name stays stable.
Product noteSeparating user-facing language from system identifiers lets copy improve without risking the plumbing.
Rejecting an email in Match Review now shows a toast with Undo for about ten seconds, also available via Ctrl or Cmd+Z, returning the email to Needs review with a full audit record.
Product noteOffering undo instead of confirmation dialogs keeps triage fast while making mistakes cheap.
The field popover and Deal history panel now show Cleared by and the person's note whenever a value was cleared, making every human correction self-explanatory.
Product noteSurfacing the reasoning behind edits next to the edit itself makes the audit trail useful day to day.
Enrichment logs filled, nothing to fill, or failed
Each enrichment run now records whether it filled fields, found nothing to fill, or failed, so incomplete runs are retried and the three-attempt cap works as intended.
Product noteDistinguishing success from silent skips keeps automated work queues honest.
Processing and AI Settings now show the mailbox state with the last email, last Gmail signal and watch renewal times, replacing a blanket health verdict with verifiable facts.
Product noteShowing evidence rather than conclusions lets operators judge system health for themselves.
The access-denied page now states plainly that a link points to no deal or to one outside the viewer's access, and guides admins to the Show test deals switch when relevant.
Product notePrecise copy that admits uncertainty earns more trust than a confident but wrong explanation.
Signed-in users no longer hold any Data API privileges on the AI call log and circuit-state tables, completing the least-privilege cleanup with no change to app behavior.
Product noteRemoving access that policies already deny closes the gap between intent and configuration.
Match Review cards, reassign search results and queue rows now show each fund by its name, and members see real fund names even for funds outside their own grants.
Product noteSpeaking the user's language instead of the system's removes a translation step from every review decision.
The Processing tab's routed-to-review count is now keyed on mailbox and Gmail ID together, so the same message arriving in two mailboxes no longer skews the number, with new tests covering degraded mode and dry runs.
Product noteClosing review follow-ups promptly keeps small inaccuracies from compounding into distrust of the numbers.
The MCP gateway now reports every stage by its board name and accepts either the board name or the stored key when updating a deal, so agent conversations read exactly like the pipeline.
Product noteDeriving every label from a single mapping keeps people and agents working from one vocabulary.
A new lint rule requires every new public table, view or sequence to carry explicit Data API grants, ahead of Supabase ending automatic grants on 30 October.
Product noteEncoding an upcoming platform change as an automated check turns a deadline into a non-event.
The Ignored tab now names its action Reclaim everywhere, from buttons to toasts, and empty states say which filter is active and how many emails sit under All.
Product noteConsistent naming across buttons, tooltips and messages lets users learn an action once.
Animal Gross can no longer be assigned as Investment Lead, enforced in the database and removed from every lead picker, while it remains available as an Owner.
Product noteEnforcing an accountability rule at the data layer, not just the interface, makes it hold for every entry point.
AI Settings now reads the same mailbox health as the Processing tab, using the shared 72-hour quiet rule and keeping a genuine expired watch distinct from a quiet inbox.
Product noteGiving two screens one source of truth removes contradictory signals before they cost anyone a decision.
AI summaries show when they were written and flag stale stage claims
The summary panel now states when a summary was written and the deal's stage at that time, highlights deals that have moved since, and marks any stage claim the record contradicts.
Product noteLabeling AI output with its context lets users judge freshness instead of taking it on faith.
The deal page's compact Move control is now fully inert while its header is hidden and shows a Checking... reason while gates are evaluated, so no disabled button goes unexplained.
Product noteExplaining why a control is unavailable turns a confusing dead end into a clear next step.
The deal edit page now shows No changes yet, Unsaved changes, Saving..., or Saved at a specific time, and a failed save keeps the user's edits in place with the reason.
Product noteHonest status feedback protects users from losing work to a reassuring but false label.
Unknown review links land on a clear not-found page
Malformed Match Review URLs now return a proper 404 with a review-specific not-found page instead of a server error, while missing and denied items still look identical.
Product noteHandling bad input at the edge keeps errors meaningful and access boundaries intact.
Reclaims from the Ignored tab and ground-truth reassignments now carry manual provenance, so the routing history clearly separates a person's judgment from automated decisions.
Kanban column, header and swimlane counts now exclude test deals, showing test cards as a separate tally when visible, and search results mark test deals with a Test badge.
Product noteKeeping test data out of headline numbers preserves confidence in every pipeline metric.
Every table's existing Data API privileges are now declared explicitly in a migration, so fresh environments match production exactly once Supabase stops granting defaults.
Product noteMaking implicit configuration explicit removes a class of environment drift before it can surface.
Provider and team domains never pose as deal contacts
Service provider, fund and team domains, including their subdomains, no longer count as identity evidence or fill a deal's Contact, while portfolio companies remain eligible for matching.
Product noteValidating the change against 90 days of production decisions first kept the improvement free of side effects.
The OKR Command Center runbook now routes every board update through the token-authenticated agent API with a recorded log entry, retiring direct database writes.
Product noteFunneling writes through a single audited interface keeps progress tracking accountable.
Operations gained guarded clear and restore functions for unverified AI-filled deal fields that refuse to act if a person has edited the value or it has changed since review.
Product noteBuilding the undo path alongside the corrective action makes data cleanups low-risk by design.
The OKR runbook now states accurately that the former automation is pending switch-off rather than already paused, with the agent API confirmed as the only writer.
Product noteCorrecting documentation as soon as it drifts from reality keeps operational guidance reliable.
Deals created from email now skip provider, fund and team addresses when seeding the Contact, choosing the first genuine counterparty along with the matching phone number.
Product noteApplying a proven rule to every intake path keeps data quality consistent from the first record.
Guard and Score badges explain themselves on any device
Match Review Guard and Score badges now reveal their explanation on keyboard focus, tap or hover, with touch-friendly targets, rather than relying on a mouse-only tooltip.
Product noteDesigning explanations for keyboard and touch users widens access without adding new patterns.
Unneeded Data API privileges were revoked from signed-out and signed-in roles, leaving anonymous visitors read access to the public changelog only, with an automated guard against regressions.
Product notePairing least-privilege cleanup with a regression test keeps the security posture from drifting back.
Continuity matching ignores provider and team domains
Email continuity matching no longer links threads through service provider or team domains, so shared advisors and colleagues cannot tie unrelated deals together.
Product noteVersioning the matcher with each behavior change keeps accuracy tracking comparable over time.
New AI summaries no longer state or infer a deal's stage from email text, relying on the record instead, with the updated prompt certified at a perfect score on the golden set.
Product noteCertifying every prompt change against a fixed evaluation set keeps AI quality measurable rather than anecdotal.
Security allowlist restructured for conflict-free reviews
The SECURITY DEFINER allowlist now lives as one file per function, with an exact-set check in CI, so parallel database changes no longer collide on a shared list.
Product noteShaping review artifacts to avoid merge conflicts keeps security discipline from slowing delivery.
Enrichment skips model calls with nothing to write
Enrichment no longer calls the AI model when the only empty fields are contact fields that are skipped for the run, eliminating spend that could never produce a result.
Product noteTracing cost back to outcomes surfaces waste that would otherwise hide in aggregate spend.
Decision history guard covered by the reader allowlist
The routing reader allowlist now includes the guard that keeps decision history append-only, returning the database security suite to green across all 4,543 checks.
Product noteRestoring a red security check immediately keeps the safety net trustworthy for every change that follows.
Every count of waiting or routed emails now reads one current decision per email, so the queue reports the 19 emails actually awaiting review rather than 1,373 historical decision rows.
Product noteDefining a single source of truth for a metric keeps every screen telling the same story.
Rejecting a deal that is already in Rejected Investments is now refused before anything is written, so its recorded reason and note stay intact and the person is asked to reload.
Product noteRefusing an action that would silently overwrite history keeps the decision record trustworthy.
The email routing decision history now refuses edits and stray deletes at the database level and keeps its record even when a rematch job disappears mid-run, so no pass drops out of the audit trail.
Product noteHardening an audit trail right after it ships keeps its guarantees from eroding under later changes.
The Match Review sender header now skips blank profile names in favour of the email's own name or address, and an email with no sender record describes its fund signal accurately instead of calling the sender an outsider.
Product noteFixing small copy defects promptly protects the credibility of the reasoning shown beside them.
Each pending Match Review row and card now carries a reason chip, and the detail card opens with a plain-language Why this email is waiting line backed by the email's decision history.
Product noteExplaining the system's reasoning at the point of decision builds the trust that review-first automation depends on.
Rejecting a deal into a lane the person cannot use now names the lane and points to an admin instead of suggesting a retry, and the reason and note stay in place whenever the server refuses the move.
Product noteReplacing a generic retry with the actual cause stops people repeating an action that cannot succeed.
New database tests prove a deal leaves Rejected Investments only through a genuine Reopen, local and CI databases now pin Postgres 17 to match production, and the assistant's deal update returns the same Reopen guidance when a race is caught.
Product noteMatching test infrastructure to production removes a whole class of surprises before they can reach users.
Archive runs execute only the plan someone reviewed
Every merge, consolidate and revert run now requires the plan hash printed by its reviewed dry run and stops if the plan has shifted, while a revert leaves alone any deal changed since the original run.
Product noteBinding execution to the reviewed plan keeps human sign-off meaningful rather than ceremonial.
Every tab and navigation count now uses one shared badge, white bold on a deeper red or dark text on a neutral pill, so counts meet accessible contrast on every screen.
Product noteConsolidating repeated UI into one component lets an accessibility fix land everywhere at once.
Archive scripts check their own work before committing
The duplicate merge and fragment consolidation scripts now confirm, inside the transaction, that they archived exactly the records their dry run listed, and a revert clears the archive reason together with the archive date.
Product noteMaking a bulk data operation verify its own count before commit turns a silent mismatch into a refused change.
Editing a deal no longer rewrites unrelated review decisions
A deal edit now changes a pending Match Review decision only when that deal becomes the email's top candidate, and every rematch pass is recorded in an append-only history so earlier reasoning is never lost.
Product noteNarrowing an automated rewrite to the cases that warrant it protects the work reviewers have already done.
The agentic development changelog gained a 9 to 11 October supplement and full entries for the week of 12 October, delivered as one guarded migration that leaves product rows untouched.
Product noteKeeping the public record on cadence turns the changelog into evidence rather than recollection.
Unused weekly brief helpers are gone, the email now sends only well-formed deal ids to its access check, and the security allowlist cites the correct evidence commit.
Product noteClearing review follow-ups promptly keeps a security-sensitive path small enough to reason about.
Rejection dialog keeps your reason when a save fails
The rejection dialog now stays open while its save runs and, if the save fails, keeps the category and note with a clear message so one tap retries, while a refused drop leaves the card in its lane.
Product noteHolding state until the server confirms success makes failure recoverable instead of costly.
Leaving Rejected Investments always goes through Reopen
The database now refuses any move out of Rejected Investments that does not come from Reopen, so the why-now and reopen record cannot be skipped by any path.
Product noteBacking a product rule with a database guarantee means no future screen can quietly break it.
The board's Move Blocked success toast no longer offers Undo when the deal moved to Rejected Investments, so leaving that stage always goes through Reopen with its why-now.
Product noteRemoving a shortcut that bypasses a deliberate process protects the process without adding friction elsewhere.
The archive reason control now tells people exactly who may change it, and if the permission check fails it asks for a reload rather than showing a refusal.
Product noteDistinguishing denied from unknown keeps an error from being mistaken for a decision.
Plain-text attachments are never hidden from the note count
The inline-image rule now separates plain-text emails from undecoded HTML, so a PDF sent inline from a plain-text mail client always counts while signature logos still stay out.
Product noteSplitting an edge case on evidence avoids trading one wrong count for another.
Ignored rows tell viewers without access to ask an admin
Viewers without Gross Labs access now see Ask an admin to reclaim on reclaimable Ignored rows instead of a Route as deal button that would always refuse them.
Product noteReplacing a dead-end button with a clear next step respects people's time as much as any new feature.
A Can bypass stage gates tick in the Edit access drawer saves with role and funds, adds an Access history line, and shows a Gate bypass chip in the People table, with excluded profiles shown as unavailable.
Product notePlacing a sensitive grant next to existing access controls, with an audit line, keeps the decision visible and reviewable.
When a gated move into Active or Rejected Investments is refused, people holding the bypass permission now get a Move anyway dialog, and every bypass is recorded on the deal with the checks it skipped, while everyone else sees Move Blocked as before.
Product noteLetting the server's refusal drive the dialog keeps the exception explicit without a second copy of the rules.
The never-owner rule now has dedicated refusal tests on real deals that check the exact error message, and the surrounding code comments describe the rule as it actually behaves.
Product noteLocking a guard's exact behaviour in tests keeps a later change from loosening it unnoticed.
Archiving a deal now asks whether it left the pipeline or was never a deal, and dashboard window metrics leave out the Not a deal archives so the numbers reflect real pipeline movement.
Product noteCapturing intent at the moment of action is what makes the metrics built on it honest.
An AI-filled company name, website or contact email that nobody has verified can still help rank a candidate deal, but it can no longer carry an automatic attach or a plain Accept on its own.
Product noteRequiring human confirmation before machine-written data drives automation keeps the review-first posture intact.
Moving a deal to Rejected Investments now checks the stage gate first, so the rejection dialog is the final step and its confirm is what moves the deal.
Product noteAsking for the reason only once the move can succeed spares people from writing context that would be thrown away.
Direct reads of stored weekly briefs are now admin only, so members see brief content exclusively through the access-aware view introduced alongside it.
Product noteSequencing the lock after the replacement path was live made a security change invisible to the people it protects.
Bulk accept now treats a match whose hold status cannot be read as held and skips it, and a same-company duplicate resting on an unchecked company name now asks for Accept and confirm instead of a plain Accept.
Product noteFailing closed when a safety lookup errors keeps an automation's worst case harmless.
Forwarded email bodies no longer pose as attachments
Text bodies that Gmail passes by attachment reference are no longer saved as attachments named by part id, and older rows of that kind no longer count as lost notes.
Product noteFixing both the ingest and the reader of historical rows resolves a data issue fully rather than only going forward.
The weekly brief's facts now leave out deals archived as Not a deal, and scripted archives record that reason, so departures, baselines and bullets agree with the dashboard.
Product noteApplying a new rule everywhere the data is read keeps every summary telling the same story.
When a rejection is refused by the stage gate, Save and move in Move Blocked now carries the category and note already entered, and says so, instead of dropping them.
Product noteNever asking someone to re-enter what they already gave is a basic test of a well-designed recovery path.
Members read the brief through an access-aware view
The dashboard now loads members' weekly briefs through a function that returns only their funds and trims every bullet to deals they are allowed to open.
Product noteShipping the new read path before tightening the old one keeps every member's card working throughout the change.
A new per-person permission and server function let selected people move a deal past an unmet stage gate, with exclusions checked both when the grant is made and when it is used.
Product noteLanding the permission model in the database first means every later screen inherits enforcement it cannot skip.
Admins can see, preview and resend the Monday email
Under the weekly brief card, admins now get a delivery line for every outcome, a delivery log with provider reasons, Send again for failed sends and Preview my email.
Product noteMaking delivery observable from the product itself turns an email job into something admins can trust and act on.
The weekly brief card now drops any bullet about a deal the reader cannot open, using the same rule and wording as the Monday email, while admins still see the full brief.
Product noteOne shared rule behind two surfaces prevents the card and the email from ever disagreeing about access.
When someone forwards a Monday brief to Animal with a note written above it, the Ignored row now says so, while the brief itself is still ignored as a whole.
Product noteSurfacing the human signal inside an ignored email keeps automation strict without losing what people meant to say.
The On and Off word beside each Monday brief checkbox in System Config > Users is gone, so the tick alone shows the state and no longer looks like a fixed label.
Product noteRemoving redundant copy is often the fastest way to make a control's purpose obvious.
Members now see only the Ignored emails about their funds or sent to their own mailbox, with counts, sender lists and reclaim all following the same server rule, while admins still see everything.
Product noteEnforcing scope in a single server-side rule keeps every view and count in agreement.
A forwarded or bounced weekly brief in the Ignored tab is now visible only to admins and the owner of the mailbox it reached, so its subject and snippet never surface to other members.
Product noteLocking down visibility before the email feature ships means the first send lands on a surface that is already safe.
Every opted-in recipient now receives the weekly brief by email at 08:00 New York time each Monday, cut to the funds they can see and built from the stored brief with no fresh model call.
Product noteDelivering already-reviewed content, rather than regenerating it per send, keeps what people read consistent with what was checked.
The agentic development changelog gained an 18 to 20 September supplement and full entries for the weeks of 21 September, 28 September and 5 October, delivered as one guarded migration that leaves product rows untouched.
Product noteCatching up missed runs in a single reviewable change keeps the public record complete without rewriting history.
A locked OKR period can no longer be deleted or edited, protecting every objective, key result and log beneath it, and the dial now shows unscored work honestly.
Product noteLayering safeguards at each level of a record means one gap never becomes data loss.
Stale invites can be replaced, and sign-in explains why
Admins can now replace a pending invite that has gone stale, and someone signing in with an unaccepted invite is told exactly what is waiting for them.
Product noteTreating the dead ends in an onboarding flow as product work removes friction before it reaches support.
Emails from linked team senders open deals more reliably
An email from a team member's linked address that names a real company within their fund grants can now open a new deal, instead of being held back only for lacking a deal keyword.
Product noteShipping a routing change behind a new settings version keeps its activation explicit and reversible.
An email filed on the Velocity Opportunity Fund through a service-provider contact is back in Match Review, with the deal's dates and contact details left exactly as they were.
Product noteReusing a proven correction pattern, with end checks that refuse any unintended change, keeps data fixes predictable.
Eight internal tables now carry an explicit deny policy for signed-in and anonymous users, and the last mutable search path warning on audit helpers is cleared.
Product noteWriting down an existing rule turns an implicit safeguard into one that can be reviewed and tested.
Every privileged database function a signed-in user can call is now documented with the check it runs and a verdict, giving the lint gate an exact allowlist to enforce.
Product noteAn explicit, reviewable inventory makes privileged access a deliberate decision rather than an accumulation.
Attaching a deal document to a note is now limited to the person who wrote that note, matching how notes behave everywhere else, with no change for anyone using the app as intended.
Product noteFixing the one finding an audit surfaces, before widening scope, keeps security work focused on real risk.
Exact matches show their evidence and need corroboration
Match Review shows what produced an Exact match, uncorroborated identity hits go to review instead of filing automatically, and team domains no longer count as identity.
Product noteShipping a routing guard switched off first makes its activation a separate, deliberate decision.
The corroboration guard and team-domain exclusion are now live through a new matching settings version, applied only once the code that reads it was serving.
Product noteSequencing configuration behind the code that reads it eliminates a deploy-order race.
Four emails auto-attached to the wrong deal are back in Match Review for the right call, and the documents they filed are reversibly taken off the affected deals.
Product noteCorrecting past errors through the normal product path keeps the fix auditable and reversible.
Removal now locks the document and marks every related record in one call, completing the safeguards required before any public way to remove documents ships.
Product noteSetting a hard rule that safeguards land before exposure keeps risky capabilities contained.
Emails from a linked address now read via the address, linked to the person, across five sender surfaces, and matching uses the linked person as the sender.
Product noteResolving identity once and reusing it everywhere keeps attribution consistent.
Linked sender addresses and holds for unknown senders
Admins can link extra sender addresses to a person, and email from unlinked senders now waits in Match Review instead of filing itself or creating a deal.
Product noteHolding uncertain input for review rather than acting on it is a sound default for automated routing.
When a later email states a fact word for word, an empty deal field now fills automatically, marked unverified with Confirm and Undo on the field and in For You.
Product noteAutomating only verbatim evidence and keeping every fill reversible lets automation earn trust incrementally.
A batch of security-scan findings is resolved with a patched Next.js release, bind parameters for script SQL, dependency bumps and tighter checkout credentials.
Product noteClearing quick security wins in one focused batch keeps the findings backlog from going stale.
Admins get a Users tab with an Edit access drawer for fund grants, and members see honest denied states wherever they reach something outside their funds.
Product noteShipping the tool before anyone needs it lets access changes happen deliberately rather than urgently.
Admins can now review matching calibration proposals on AI Settings and Apply, Dismiss, Undo or run a check, while everyone else sees the card read-only.
Product notePutting a human approval step in front of automated tuning keeps control where accountability lives.
Path and file-read findings reviewed and documented
The remaining file-inclusion and path-traversal findings were confirmed unreachable from any request, job or tool, and the reasoning now sits beside the code.
Product noteDocumenting why a finding is a false positive is as valuable as fixing a real one.
Invite end-to-end tests now create ordinary, non-test invites with a local-only guard, so they assert real behaviour while the live admin rule stays unchanged.
Product noteFixing the test rather than loosening the product rule protects the guarantee the tests exist to check.
Roadmap and Updates become one page with a month rail
Roadmap and both changelogs move into a single signed-in Roadmap + Updates page with a month side nav, legacy links redirect cleanly, and OKRs becomes admins only.
Product noteConsolidating related surfaces while preserving old links lets navigation improve without breaking anyone's bookmarks.
Fund card rings stack one per row with clearer copy
Rings now sit one per full-width row with their legends beside them, grow to 96 px, and carry refined copy that invites selecting a slice to see its count.
Product noteIterating layout and copy as a follow-up keeps the data change and the presentation change independently reviewable.
Fund grants backfilled alongside the strict fund rule
Every real member is granted every real fund in the same release that switches on the strict fund rule, so no one loses access to a deal they could already see.
Product notePairing a backfill with the rule it enables removes any window where users are worse off.
Twelve OKR and delivery tables now allow reads by admins only, closing direct API access to spend, invoices and audits that the pages already restricted.
Product noteEnforcing access in the database, not just on the page, is what makes a permission real.
The Activity person view and the Team table now share one SQL definition of a person's field activity, so a filtered person no longer shows No activity while Team counts their edits.
Product noteWhen two panels report the same fact, one definition should produce both.
Delivery Performance accuracy entries now appear as accessible accordion rows with date, title and outcome, and new audits are written with human-readable titles.
Product noteFixing both the display and the data written behind it keeps the improvement from fading as new entries arrive.
The calibration regression gate now counts rows rather than totals and records its knob audits cleanly, while still only proposing changes for a person to act on.
Product noteTreating review feedback on an automated decision engine as must-fix guards against silent drift.
Signed-in users following the old Updates address land on the matching tab of Roadmap + Updates, while signed-out visitors still see the public changelog.
Product noteRedirecting by audience lets one address serve both members and the public without compromise.
Fund grants gain provenance and an append-only audit
Every fund grant and revoke now records who, how and when in an append-only audit, laying the groundwork for fund-scoped permissions without changing anyone's visible deals.
Product noteLanding the additive model before any rule change means each step is safe to ship on its own.
Matching calibration proposes settings from review outcomes
A weekly check replays human Match Review decisions under candidate settings, backtests each one, and records either a proposal or an honest nothing to suggest, without changing live matching.
Product noteLetting automation propose while people decide keeps learning loops fast and accountable.
Every document version can now get a first-page picture drawn in the background and served only to people who can already see the document, with the engine merged switched off.
Product noteShipping dark behind one switch separates deploying code from releasing behaviour.
Filter by sender on Extraction and Email Intelligence
A From select on the Extraction tab and all four Email Intelligence tabs narrows the loaded rows to one sender, with bulk selection scoped to what is visible.
Product noteFiltering client-side over rows already loaded delivers the feature without widening a single query.
Under 360 px the five tab labels shrink to fit their columns, so DASHBOARD no longer loses its first letter on the narrowest phones while tap targets stay at 48 px.
Product noteTesting at the smallest supported viewport, not the designer's phone, is where layout defects live.
Test-created admin accounts are locked with cap, audit and undo
A migration locks admin accounts created by tests, with a dry run, a hard cap on how many it touches, an audit row per account and an undo while the ban is in place.
Product noteSecurity clean-up shipped with a cap and an undo is how a one-way change becomes a reversible one.
An account flagged as a test user can no longer pass any admin check, with the rule held in one database function that 56 policies and 12 functions already inherit.
Product notePutting a security rule in one place means every surface inherits the fix instead of waiting for its own.
A person filter on Dashboard Activity narrows the 7-day feed to Animal or any teammate, running on the server across the whole window rather than the loaded page.
Product noteFiltering on structured actor data instead of display text keeps the answer correct as names and copy change.
Tab pills show the true server count instead of 99+, match the footer total of the list they count, and explain themselves in a plain-language tooltip.
Product noteA count capped for layout convenience hides exactly the backlog a team needs to see.
Every unmet requirement for the next gate now comes from one ordered list, so the header count, tab dots, section strip and For You always agree on what is left to do.
Product noteDeriving every indicator from a single source of truth removes a whole class of contradictory signals.
Matching weights and routing thresholds now load from an append-only versioned table at the start of every routing run, and AI Settings shows which version is live.
Product noteVersioning configuration makes every change to automated decisions auditable and reversible.
Each Dashboard fund card gains two donut rings, deals by Gross Labs class and by deal type, computed on the server from the query the cards already use.
Product noteReusing the existing query keeps the new view consistent with the numbers beside it.
E2E guard refuses preview and production site addresses
The Playwright guard now checks the browser target as well as the database, so a suite pointed at a Vercel preview or the production site stops before its first click.
Product noteA test suite that can only reach disposable environments is the cheapest safety net there is.
Equal-score routing ties go to review, never auto-attach
When two candidate deals clear the auto-attach threshold with the same score, the router sends the email to review instead of letting a tie-break rule pick a winner.
Product noteAutomation that knows when to stop earns more trust than automation that always decides.
Moving a deal and the For You blocker list no longer ask for an Investment lead the deal already has, because the gate now reads the current profile column rather than the retired one.
Product noteWriting the failing test on main before the fix turned a report into evidence.
Deal pages replay correctly now that entrance animations no longer freeze at their first frame, and console and network capture are enabled for every recording.
Product noteObservability tooling that shows a blank page is worse than none, because it looks like the product failed.
OKRs tab gains a period rail, progress card and collapsible objectives
A year and quarter rail, a progress card and collapsible objectives replace the flat OKR list, with the tab following the URL and the rail opening the selected year.
Product noteBuilding the rail from the data that exists, rather than a hard-coded list of quarters, means next year needs no code change.
The queue row, Match Review header score, Top badge and Accept target now all come from a single pure ranking function, so they can never name different deals for the same email.
Product noteWhen four surfaces show the same fact, one function should compute it.
A Needed to move row for a field with a shared control now opens a fill-in modal built from the same editor the Move Blocked modal uses, so the gap can be closed without leaving the rail.
Product noteReusing an existing save path for a new entry point keeps validation identical everywhere the field is edited.
Review ranking ties break on stage before corroboration
At exactly equal score the review ranker now prefers the earlier-stage deal before considering corroboration, applying the ruling to the single shared ranking function.
Product noteRecording a ruling in the one function every surface reads makes the rule visible and reversible.
Dashboard cards become surfaces for the same Animal engine the board uses, with one provider, one timer and one planner, and every number and tooltip painted above him.
Product noteReusing one engine across surfaces means a personality feature never forks into two behaviours.
Dashboard tooltips, motion, accessibility and analytics
Every metric label, chip and footer on the dashboard carries a tooltip explaining what it is, why it matters and what good looks like, with reduced-motion support, keyboard access and an axe check in the e2e suite.
Product noteA metric a reader cannot explain is a metric they cannot act on.
Funds compared emails load for every signed-in user
Linked emails on the Funds compared tab now come through a definer function, fixing a permission error that showed Not loaded for every user while granting no new access to the raw email table.
Product noteFixing a permission failure without widening a grant keeps the least-privilege model intact.
Weekly brief sections run concurrently within a measured budget
Fund sections of the weekly brief generate in parallel with a measured time budget, so a slow model response for one fund no longer times out the others.
Product noteSizing timeouts from measured behaviour rather than assumptions is what keeps a scheduled job reliable.
The stable OKR identifier function now runs with an empty search path and fully qualified calls, clearing the last security advisor warning outside the accepted allowlist without changing behaviour.
Product noteDriving the advisor warning count to a known, accepted set makes any new warning immediately visible.
The set of deals behind any dashboard number is now exactly the set the same viewer sees on the board, fixed at the single predicate every dashboard RPC joins through.
Product noteTwo screens that disagree about what exists erode trust faster than any missing feature.
The brief now says Active Investments where every screen does and formats allocations compactly, with both applied at the fact-pack boundary before the model sees them.
Product noteNormalising inputs before generation is more reliable than correcting outputs afterwards.
Weekly brief card with sources panel and admin actions
The stored weekly brief appears on the dashboard with every state handled, a Sources panel that shows the facts behind each line, a history drawer and admin Regenerate and Retry.
Product noteShowing the sources beside the summary turns AI output into something a reader can check in seconds.
Sign-in mirroring, presence tracking, the Team rows function and team metrics land as the data layer for the Team tab, with idempotent migrations and no UI change.
Product noteSeparating the data layer from the screen makes each half easier to review and safer to ship.
Every section 7 metric the weekly dashboard owns is now computed in one module from the aggregate RPCs, backed by fixtures, unit tests and a proof query.
Product noteProving the numbers before drawing a single chart separates a dashboard from a decoration.
Pipeline card age badge counts days in the current stage
The age badge now reads stage_changed_at rather than the last edit, so a routine update no longer resets a deal's apparent age and the card agrees with the dashboard.
Product noteWhen two surfaces disagree on a number, one shared helper is the fix.
Board fund and stage parameters with a shared fund cookie
The board accepts fund and stage parameters, remembers the fund filter in a shared cookie and gains the flag gate and navigation the dashboard will use for drill-downs.
Product noteLaying plumbing in its own release keeps the visible feature that follows small and reviewable.
Weekly dashboard screen with widgets and drill-down drawer
The /dashboard screen renders every metric from the signed-in client with RLS applied, with widgets, a drill-down drawer and loading, empty and error states.
Product noteReading through the caller's permissions rather than a service role means the dashboard can never show more than the board would.
Weekly brief job with grounded fact pack and validator
A Monday job builds a fact pack from the dashboard metrics, asks Gemini for a structured brief and rejects any sentence not grounded in the facts before storing it append-only.
Product noteValidating AI output against the source facts is what makes a generated brief safe to put in front of a partner.
The pure metrics that compare funds side by side, the grouped gate-context reads and fixtures land ahead of the tab, with test and archived deals excluded at the source.
Product noteShipping the calculation before the chart lets the numbers be reviewed on their own.
Money, dates and names in the For You check rows now render with the same formatters the confirm-term rows use, so $25,000,000 reads the same wherever it appears.
Product noteOne formatter per data type keeps a product consistent without a style guide anyone has to remember.
Dashboard foundation: tables, RLS and aggregate RPCs
Three dashboard tables, an audit table and five read-only aggregate RPCs land with the universal dashboard predicate baked into SQL, so no caller can forget to exclude test or archived deals.
Product noteEncoding the data rules once at the database layer makes every future widget correct by construction.
A numbered tour of the deal page, a screenshot freshness check and an accessibility pass complete the training story, all built from a test fixture deal so no real data appears.
Product noteBuilding help material from test fixtures keeps it safe to publish and easy to refresh.
Twelve pgTAP assertions lock in the exact reported case of flipping a deal to test data with test deals hidden, covering the admin non-owner, the owner and a viewer without rights.
Product noteTurning a reproduced bug into a permanent database test is how a fix stays fixed.
Once a retry parses a term sheet successfully, the earlier failure moves to history instead of leading the panel, so the deal team sees the parsed terms first.
Product noteRead models should reflect the current state of the work, not every step it took to get there.
Parsed term-sheet values awaiting a decision now show as confirm-or-reject rows in the For You panel, so reviewers act from one place instead of hunting through the Terms section.
Product noteRouting every pending decision to a single queue shortens the loop between extraction and confirmation.
The admin-only Animal panel becomes a /training route with navigation links, generated field tiles and a section sub-nav, giving the whole team one place to learn the product.
Product noteShipping the frame first, with content sections to follow, lets the team start using help material while it is still being written.
Sign-in becomes invite-only at the app and database
An uninvited external address can no longer request a magic link or receive a profile; eligibility is checked at request time and a database guard backs it, so the hosted signup flag is no longer the single control.
Product noteDefence in depth means a security setting is never the only thing standing between the outside and the team's data.
Any active, verified team member can now attach files by text to a live deal they can read, with the size cap, type allowlist and archived-deal refusal unchanged.
Product noteAligning a new channel with the existing permission model removes a special case nobody needed to remember.
Product Coalition recognised by name, not by sender domain
The newsletter stays on the known non-deal list, but matching by its sender domain is removed so a real deal forwarded from the same address is never withheld.
Product noteA filter is only as safe as its worst false positive, so the rule is scoped to the narrowest signal that still works.
Training page gains content sections, FAQ and glossary
Approved copy for every training section, a twelve-item FAQ and a glossary now live on /training, with each sentence footnoted to the code it was checked against.
Product noteDocumentation that cites the code it describes stays honest as the code changes.
Saving a deal as test data with test deals hidden now returns to the board with a confirming toast instead of a full-screen 404, and every flag flip writes an audit row.
Product noteFollowing the user past the save button reveals failures a green request log never shows.
Ticking the test-deal flag no longer fails on a row-level security refusal; the save goes through a dedicated RPC, the banner states the reason when a save is blocked, and Product Coalition joins the known non-deal list.
Product noteDiagnosing a failure down to the exact policy error before changing anything avoids fixing the wrong layer.
When an incoming email matches an existing deal, the router now judges whether it describes the same money or a new round, and anything that looks like a new round or is unclear goes to review instead of attaching quietly.
Product notePausing on ambiguity costs a reviewer a minute; misfiling a new round costs the pipeline its accuracy.
A verified texter can send a supported file to the deal agent and have it land on the right deal through a reserved, signed single-PUT upload into private storage, with the original send time and sender preserved as provenance and a sweeper reclaiming abandoned reservations.
Product noteTreating an inbound file as evidence with provenance, not just a blob, is what makes a text channel auditable.
Addepar asset classes derive from GL at every gate
The Addepar class and subclass are now inherited from the GL pair at every stage gate and never require a human to set them, with gate rules relabelled rather than removed so per-stage counts stay exactly as they were.
Product noteRemoving a field nobody should have to fill is a gate simplification that costs no rigour.
Pre-fill Execute unlocks after a completed dry run
Execute is now enabled once a dry run has reached every open deal, not only after a separate Estimate, and requests that stop early on a lost request or cap halt are explained on screen instead of silently leaving Execute disabled.
Product noteWhen a safety gate blocks the intended path, the fix is to honour its purpose, not to remove it.
The peers switch is now part of the enrichment run identity, so releasing peers re-offers every previously enriched deal for a fresh run instead of leaving the field empty until the next inbound email happened to arrive.
Product noteVersioning a run by everything that shaped its output is what makes a feature release actually reach existing records.
The changelog gained the 11 to 13 September supplement and six new rows for the week of 14 September in a single guarded migration, leaving product rows untouched and safe to re-run.
Product noteA public record that updates on cadence, not on memory, is one the team can rely on.
A text that arrives with a file and no words is now accepted by the gateway and shown as Attachment only, with the body recorded honestly as empty rather than filled from earlier messages. Every attachment check still runs unchanged.
Product noteRecording an absence as an absence keeps the correspondence history trustworthy.
A one-command probe sends a single test email and follows it hop by hop from raw email to routed deal with a UTC timeline, every Gmail push receipt is now logged, and reclaimed reserved-domain deals are flagged as test data automatically.
Product noteInstrumenting the real delivery path is the only way to distinguish a quiet mailbox from a broken one.
Sixteen portfolio companies recorded as known non-deals
Following Jay's ruling, the sixteen portfolio company names that the guard replay would still have minted as deals are now on the known non-deals list as a new portfolio kind, added by an idempotent migration with an audit row per entity.
Product notePutting a human ruling into a reviewable migration preserves both the decision and its reasoning.
Term sheet parsing now runs Gemini at a low thinking level, cutting a twelve-page sheet from 235 seconds to under 20, scales its timeout with page count, and offers a Retry parse control on the deal page for anything that still fails.
Product noteMeasuring the same input under both settings before changing a default is what separates tuning from guessing.
Processing tab shows whether the email pipeline is healthy
A read-only Processing tab at the front of Email Intelligence records every scheduled job run and mailbox activity, so the team can tell normal latency from a gated pipeline, a dead intake or a connected-but-quiet mailbox at a glance.
Product noteMaking pipeline health observable to the people who depend on it replaces anxiety with evidence.
Any active team member can edit and verify live deals
Editing and verification are now open to every active team member for any live deal they can read, through a single database helper that every write policy and RPC delegates to. Owner remains the accountability field, archived deals stay read-only, and test or banned accounts never receive the grant.
Product noteCentralising the answer to who may write this in one place makes a policy change a one-line change.
Production migrations unblocked and the Processing tab live
Five migrations from parallel sessions had collided on the same version number, blocking every production migration run since the attachment-only change. Renumbering them restored the deploy pipeline and let the job history tables behind the Processing tab reach production.
Product noteParallel delivery needs a filename validator in CI precisely so a collision fails loudly rather than silently.
Duplicate migration files removed after a merge collision
Two fixes for the same migration version collision landed with different numbering, leaving identical files twice on main; the copies production never recorded are gone and the ledger again matches the repository.
Product noteReconciling the repository to what production actually applied, rather than the other way round, keeps the deployment history the single source of truth.
Every pre-fill control now states its reason when greyed out, and Execute enables in the same render that shows the dry-run result, so a two-deal run no longer leaves the button dark with the wrong hover text.
Product noteA disabled control that explains itself turns a moment of doubt into a clear next step.
Main CI turned green once the duplicated migration copies from two overlapping fixes were removed, restoring the filename check that protects the production migration ledger.
Product noteA red main branch is treated as an incident, not a backlog item.
Reconciliation guard completes: mention positions, thread veto and a withheld-creations tile
Rules two and three of the reconciliation guard reached main: a company mentioned only in passing no longer seeds a deal, a reply inside an existing thread cannot spawn a new one, every routing decision records which guard rule applied, and an admin tile counts creations withheld by the guard.
Product noteRecording why automation declined to act is as valuable as recording what it did.
Empty deal-page sections collapse to a single line
A section with nothing in it now renders as one row with the section name, an Add action and a Show fields toggle, so a deal page shows what is known instead of a wall of blanks. Pre-filled values awaiting verification keep their section open.
Product noteReducing visual noise on sparse records makes the missing information, not the layout, the thing a reader notices.
Owner and Investment Lead names never render blank
A shared display-name helper falls back from full name to invite name to email before ever showing nothing, so the ten legacy deals whose owner profile had an empty name now show a person on every card, chip and detail view. No data was changed.
Product noteFixing the render path rather than patching the data means every future incomplete profile is handled too.
Known non-deal entities stop the router creating phantom deals
An admin-owned list of companies that are structurally never deals, such as our own funds, service providers and co-investors, now causes the router to park matching emails in review with the rule named, instead of minting a new deal. Nothing is dropped and a human can still create the deal.
Product noteEncoding what the team already knows is not a deal is the cheapest form of precision improvement available.
Deal page gains a right rail, side sheet and bottom sheet
Behind the deal_page_layout flag, notes and correspondence move into a right rail on desktop and a bottom sheet on phones, driven by a single notes provider so there is never a second realtime channel. Ungranted users see the current page unchanged.
Product noteShipping a layout change behind a per-user flag lets the product lead review it live before anyone else sees it.
Fill sheet loader accepts real deals in any board column
The legacy fill-pack loader now matches real deals wherever they sit on the board, so the twenty deals in Active and Rejected Investments are no longer rejected as not open when a fill sheet is applied.
Product noteTooling that keys on business identity rather than current stage survives the board reorganisations it was built to support.
External collaborators can complete magic-link sign-in
The confirm route now handles the PKCE code flow the email templates actually send, so external collaborators reach the platform instead of an invalid-link banner. Until this fix no external collaborator had ever completed a production session.
Product noteReading the auth provider's logs before touching code turned a vague login complaint into a precise fix.
The Managing Partner now holds the admin role through a committed, audited data migration, opening the admin area and system configuration to him. The migration path was chosen because no UI action edits another person's role, by design.
Product notePrivilege changes that travel through version control leave the same audit trail as any other change.
At phone widths the deal page no longer scrolls sideways on long company names, every control meets a 44 pixel tap target, tables render as cards, long text gets a Read more, and sheets open on tap. These are defect fixes at every width with no feature flag.
Product noteTesting at 390 pixels finds the failures that a desktop review never will.
The admin flags page gains an On for everyone audience control backed by an audited RPC, and e2e grant cleanup now runs in a finally block, so a test failure can no longer leave a stray grant behind.
Product noteA flag rollout path that ends in a single audited click is how a staged release becomes a general release.
Deal-page panels follow the section navigation order
Under the deal_page_layout flag, panels are grouped in the same order as the section nav chips and anchors land cleanly below the sticky header, so tapping Summary reaches the AI summary rather than a card four panels down.
Product noteActing on a live review within a day keeps the feedback loop between product lead and build tight.
Every surface says Active Investments, not In Addepar
Stage labels now come from one shared map, so the compact header badge and Deal meta card show Active Investments and Rejected Investments like the board does, while the underlying stage keys stay unchanged in the database and code.
Product noteA single source of truth for user-facing vocabulary removes an entire class of inconsistency bugs.
Board reconciled against Gross Labs evidence: 27 matcher-created records archived, emails re-linked to their source deals, Playground renamed Playground Golf (#432).
The live board now holds exactly the 20 deals in Naomi Stein's evidence: 15 in Active Investments and 5 in Rejected Investments. The 28 records the email matcher had created from company names inside real deals' correspondence were soft archived (27) or flagged as test data (1, synthetic origin), reversible by run id, with every email first re-linked to the deal whose folder it came from. Counts exclude test data (is_test_deal = false).
Product noteA reusable dry-run-first runner with a run id and a revert path made a 29-record production correction reviewable on the issue before a single write, and undoable after.
Investment Lead is any eligible person: the picker now shares the Owner list, so a newly invited person is assignable without a code change (#451).
Investment Lead used to be a fixed list of three codes. It is now a reference to a person (deals.investment_lead_id) governed by the same eligibility rule as Owner: active, not a test user, and never Jay Stansell, enforced in the database and on every write path. Existing leads were carried over as a representation change with no new history rows; one legacy code with no person behind it was cleared on a test deal with a history row, as ruled on the issue. The legacy code column stays unwritten for one release as a rollback path.
Product noteA read only dry run of the backfill was posted on the issue before any code, and the one ambiguous row was ruled on there rather than guessed at in the migration.
A new gates:advance script walks every real open deal rightwards through the live stage gates until the first blocking gate, using the exact same gate evaluation as a board drag, defaulting to dry run and reversible by run id.
Product noteBulk operations that share the product's own rules and can be undone are safe enough to run on real data.
Pre-fill from evidence job and audited legacy gate rulings
Jay's gate rulings for legacy deals landed as idempotent, audited migrations with an N/A audit note and a scope editor for gate rules, and a reusable Pre-fill from evidence job now proposes fill-empty-only values from stored evidence behind the existing spend guard. No local script ever calls the model.
Product noteTurning a one-off unblock into a governed, repeatable job is how operational fixes become product capability.
The Pre-fill from evidence job can now read a company's public website as a second evidence source, using the same fill-empty-only rules, evidence-hash idempotency, dry-run mode and per-run call and spend caps as the email path.
Product noteExtending an existing governed pipeline rather than adding a parallel script keeps every safeguard intact by construction.
Bulk pre-fill runs one deal per request and no longer times out
The admin bulk Pre-fill action now walks open deals as a sequence of single-deal requests, so a run across all real deals completes instead of hitting the 300 second Server Action limit, while the per-run call and spend caps still apply to the whole run.
Product noteDesigning long operations to fit platform limits prevents a timeout from masquerading as a product bug.
Legacy deals assigned real owners and Investment Leads
Twenty-six real deals that had defaulted to Jay as owner and forty-two with no Investment Lead now carry their correct assignments, and a database guard prevents the system-owner fallback from ever landing a deal on a person who should never own one. Every change was posted with before-and-after counts on the issue.
Product noteFixing the routing default alongside the data it corrupted stops the same drift from returning next week.
Values written by AI, import or backfill paths now show as unverified until a team member keeps, clears or replaces them, with every decision recorded in the field history. Verification state is derived from that history rather than stored, so it can never fall out of sync with the record.
Product noteMaking machine-written data visibly provisional keeps human judgement in the loop without slowing it down.
Clearing a pre-filled Investment Lead now works, and main is green again
Clearing a pre-filled value compares the stored code rather than its display text, so Investment Lead and owner fills can finally be cleared, each lead change writes exactly one human history row, and a failing pgTAP test on main was repaired in the same change.
Product noteA follow-up that restores a red trunk before adding anything else keeps the whole team shipping.
The pdf.js worker is now bundled explicitly so term sheet parsing succeeds on Vercel instead of failing on every deployed run with a missing worker module. Four of the five failed real-deal term sheets recovered on the first reprocessing pass.
Product noteReproducing a production failure in the build output before fixing it turns a guess into a verified root cause.
Test invites hidden from the team list and fenced out of production
End-to-end test invites are classified at write time and hidden from the invites list without being deleted, and a production guard now refuses to run the e2e suite against live data. The Users tab shows only real people, and fixture accounts can no longer leak into the production roster.
Product noteKeeping test artefacts visible to auditors but invisible to users protects both the metrics and the experience.
Agentic changelog current through the week of 7 September
The public changelog gained the 5 and 6 September supplement, the full week of 7 September, and an editorial upgrade of the week-of-31-August product rows, delivered as a single guarded migration that cannot duplicate rows on re-run.
Product noteTreating the changelog as versioned data rather than prose keeps the public record reproducible and reviewable.
A footer icon and a Shift+A shortcut call the mascot immediately, released behind a new per-user feature flag audience with audited grant and revoke so the trigger reaches one person first with no DOM or listener rendered for anyone else. Feature flags gain a users audience the whole platform can reuse.
Product noteReleasing to a single named user before widening exposure keeps the feedback loop tight and the blast radius near zero.
An invitation is marked Accepted only when the invited person actually signs in for the first time, ending the false Accepted stamp that fired the moment an invite was sent. Historical statuses were recomputed from real sign-in data, so the Users list finally tells the truth about who has arrived.
Product noteStatus fields earn trust when they are derived from observed behaviour rather than from the system's own actions.
Gemini spend brought under a ledger, circuit breaker and daily budget
Every Gemini call is now recorded in an attempt ledger, an email is parked after three failed attempts, a circuit breaker trips on credit depletion, and a daily budget caps spend, all visible on a new AI Spend admin panel. The runaway loop that re-sent one oversized email every minute is structurally closed.
Product noteCost controls belong in the request path itself, where a guard can count what it is protecting.
Incident co-investor values restored to their prior state
A service-role RPC with a dry-run mode reverts the peers values the enrichment job wrote during the one-hour incident window, restoring each deal's earlier value rather than blanking it and leaving any field a person edited since untouched. The nine affected deals read as they did before the job ran.
Product noteA clean revert respects two orders of precedence at once: human edits over machine writes, and prior state over empty state.
Animal always loads fresh assets and says when it cannot
Sprite URLs now carry a content hash so a browser can never hold a stale frames contract after a swap, load status is published and shown on the status page with a Show Animal now control, reduced-motion visitors get a static poster and the appearance timer persists across navigation. The invisible-mascot report is resolved at its root cause rather than patched.
Product noteMaking failure states visible to users is as much a part of a fix as removing the failure itself.
A deployed guard now runs on each Vercel preview and production deployment, verifying that every sprite asset answers, the frames contract is the current version, all four roaming surfaces mount with real size and a driven appearance paints on screen. A regression in the mascot is caught by CI before a person notices.
Product noteReading production first and then automating the exact check that would have caught the report converts a one-off investigation into a standing safeguard.
Co-investors are recorded only when the words say it invested
Extraction now classifies each candidate co-investor as fill, suggest or reject based on whether the source snippet states that the entity invested, rejecting subject-line tags, senders, outside counsel and celebrity names that previously passed on shape alone. Ambiguous firms land in the review queue as suggestions instead of becoming values.
Product notePrecision on a decision-grade field comes from requiring evidence of the claim, not from recognising the shape of a name.
Each swimlane header total now opens a tooltip stating what is summed, how many deals in the lane carry a proposed allocation and how the figure is rounded, with live numbers drawn from the same labels and formatting the deal page uses. A headline like 33 deals and $8.5M is no longer read as a claim about every deal in the lane.
Product noteEvery aggregate on a dashboard should disclose its coverage, because a number without its denominator quietly misleads.
A runtime-read kill switch stops the enrichment job from writing the peers field until the extraction meets the precision bar, flippable without a deploy and covered by tests that prove other fields keep flowing. The one field known to produce false positives is contained while the rest of the pipeline keeps working.
Product noteShipping a narrow, reversible hold for the failing part preserves the value of everything that works instead of pausing the whole feature.
Extraction now asks for the fields the gates actually check
The email extraction schema covers all six gate-field gap groups, from classification and founders to ARR, historical revenue and fund terms, with every figure carrying its period and source text so a forward number can never land in a trailing column. Correspondence that plainly states a fact now fills the field instead of leaving it to be typed by hand.
Product noteClosing the loop between what the pipeline demands and what the intake collects removes manual work at its source rather than at its symptom.
Fix a blocked stage move without leaving the board
The Move Blocked dialog now renders the same field controls as the deal page for every unmet gate requirement, with a live remaining count and a Save and move action that re-runs the gate server-side before the card advances. Deal owners clear a gate in place instead of bouncing between the board and the detail page.
Product noteMeeting users at the moment of friction, rather than redirecting them to a form elsewhere, is how a control turns from a blocker into a workflow.
AI summary, risk flags and similar deals are removed from every gate, with the removal audited per gate and the trigger and editor hardened so those checks cannot be re-added. A deal move is now blocked only by facts a person can supply.
Product noteA gate is only legitimate when the people it stops can satisfy it themselves; anything else is a system failure dressed as governance.
Every deal carries the fund and asset class from the categorisation sheet
The Animal AI Training Categorization sheet was applied to production as 40 field changes across 18 deals under a named system actor with pre-image guards, while three mappings the database cannot express were reported for Jay rather than guessed.
Product noteA data pass against real records earns trust by declaring exactly how many changes it expects and refusing to run if the count differs.
The changelog record is current through the week of 31 August
The agentic record gained its 29 and 30 August supplement and the first six entries for the week of 31 August, and the product changelog for the week of 24 August was backfilled from 19 to 38 published entries with six plumbing pull requests recorded as deliberate exclusions.
Product noteA changelog is only useful as a record if the gaps are closed the week they appear.
The OKR agent can look up the key results it writes to
A read-only endpoint now returns the full objective and key-result ID map behind the agent token, unblocking the nightly dual-write that had been sheet-only since 27 August because the write endpoints required UUIDs the platform exposed nowhere.
Product noteAn automation that cannot discover its own targets is one that silently stops writing.
Animal walks with a proper arm swing, on a machine-readable contract
The v9 asset pack redraws the walk cycle with contralateral arm swing and the drumstick redrawn per grip, and every placement number moves out of the player into the frames file, so the sprite's feet and peek edges come from the contract rather than a runtime canvas measurement.
Product noteWhen the art file declares its own anchors, a re-cut of the pack changes one JSON file instead of a rendering engine.
The matcher recognises a company however the email spells it
An email that says Echo, Putt Shack or The Familie V 2.0 now auto-attaches to Echo Soccer LLC, Puttshack or The Familie when exactly one deal answers to that name, using the same same-company rules that previously only broke ties between two candidates.
Product noteA signal that is safe as a tie-break becomes safe as evidence once it is gated on uniqueness.
HTML-to-text conversion prints a linked address twice, sometimes glued together, which made one correspondent look like two strangers to the relationship matcher; header and attribution lines are now normalised at the source and on the 16 stored messages already affected.
Product noteIdentity keys have to be canonical before they are compared, or the matcher learns nothing from the history it has.
The accuracy replay scores the second look production already gives
The shadow replay had been counting every email the extractor called non-deal as an unreachable miss, when production gives those messages an attach-only rescue pass; the replay now runs the same rescue, lifting the measured curated-history accuracy from 60.1 to 62.8 percent and shrinking the extractor miss bucket from 54 to 39.
Product noteA benchmark that models a pipeline nobody runs steers the roadmap toward the wrong fixes.
The mascot is called Animal, and its tooltips no longer get clipped
Every user-facing sentence about the mascot now says Animal, and the shared tooltip is portalled to the page body with viewport-aware placement so it can no longer be cut off by a bordered panel on System Config, AI settings or the OKR pages.
Product noteA rename done in copy rather than code keeps routes, assets and tests stable while the product says the right thing.
The mascot is checked on the deployed build, not just in the repo
Two Playwright suites now run against a live deployment: one proves every sprite sheet is served byte-for-byte as the manifest describes, with transparent corners, and the other walks the mascot through its behaviour checklist, from walk timing to staying out of the way of a dialog.
Product noteA feature that lives in the browser is only verified where the browser sees it, so the QA runs against the URL people use.
Deal notes, the intelligence pipeline and the save bar are live
Deal comments and files now flow through the intelligence pipeline, the deal edit form has a save bar with an unsaved-change guard, and typing during an in-flight save is no longer lost; two earlier merges had landed on each other's branches rather than main, so this is the first time the work reached users.
Product noteWhen a rescue is needed, target main directly and verify the ancestor, because the branch graph does not care what the merge button said.
Hovering or focusing a field label now reveals a clock button that opens that field's change history, backed by an append-only record of every edit; the feature had merged into a dead branch and never reached the live product until now.
Product noteMerged is not the same as shipped, and the only test that matters is whether the change is on the branch the product deploys from.
The replayed mailbox is filed where Jay's Drive says it belongs
Three approved production passes refiled 203 replayed emails onto the deals a human had already filed them under, archived the 22 records a passing mention had invented, and left every move as an auditable reassignment the timeline and accuracy read model understand.
Product noteA correction run ships as reviewable scripts and plans, not as a console session nobody can replay.
Delivery Performance gains an Updates tab that shows the same product and agentic changelog feed as the public Updates page, deep-linkable and reading from one shared query so the two surfaces cannot drift apart.
Product noteOne feed rendered in two places beats two feeds that have to be kept in step.
The deal records the replay fragmented (Echo beside Echo Soccer LLC, RX3 beside RX3 Growth II, a domain and a truncated name posing as companies) were consolidated from a plan Jay approved, with every step naming the record by id and current name so a stale plan aborts instead of merging the wrong company.
Product noteA production edit script should refuse to derive its own scope; the human approves exactly what will move and the run proves it moved exactly that.
Mail is filed with the relationship, not the company it mentions
When an email's correspondent has a history with one deal, that continuity can now outrank a company the body merely names, so a Breakaway update that mentions a portfolio company stays on Breakaway; the rule fires only on strong, unambiguous evidence and never on shared intermediary addresses.
Product noteWho the mail is with is often better evidence than who the mail is about, provided the system is strict about what counts as a relationship.
A System Config page explains the mascot and the pipeline
A new System Config tab introduces the mascot with a live gallery of its five moves, an animated explainer of how an email becomes a deal (attach, review, or open a new deal) and a section on what it fills in, all rendered through the same frozen art contract the roaming sprite uses.
Product noteExplaining the automation in the product, in the words of the pipeline that actually runs, is part of earning trust in it.
Naming a company in passing no longer creates a deal record for it; unless the company is a party to the correspondence, states economics or appears in the subject, the email goes to review instead, closing the source of eight junk records and the attachments they were stealing.
Product noteThe cheapest way to keep a pipeline clean is to stop it inventing entities it cannot justify.
The rematch pass that runs when a deal is created now checks whether a message already has a terminal or human decision, so mail attached to one deal is no longer re-suggested for an unrelated new one.
Product noteEvery work list that asks about a message must ask the same question, or the newest one will relitigate the rest.
A re-run creates one deal per company, not one per email
The settings re-run now remembers deals it created moments earlier so a burst of mail from a new company yields a single deal, intermediaries can no longer claim a company's attachments, and a cleanup script retires the duplicates the old behaviour left behind.
Product noteIdempotence is a property of the whole run, not of each item in it.
A 16-bit drummer mascot now wanders four surfaces of the app with a sprite engine and state machine driven by a checksummed asset pack, always behind the controls and never in the way.
Product notePersonality ships under the same discipline as everything else: scoped stories, a QA checklist and a verified asset manifest.
Bumping the routing configuration version makes every email decided under the old rules eligible for re-evaluation, so the fixes for legal-entity conflicts and intermediary domains apply to mail already in the review queue, not just what arrives next.
Product noteA rule change that only affects future input leaves the backlog wrong for as long as it lives.
Signature packets, subscription agreements, wire confirmations and capital-call notices were being classified as not deal-related and never reached routing; the extractor now treats an investment's execution lifecycle as deal mail.
Product noteA definition written for opportunity mail silently discards the mail that proves the deal happened.
Synthetic deals are now marked as test data by code at creation, including those created from email intake, and a backfill flagged the ones that had slipped onto the live board with a record of how they were identified.
Product noteA flag the seed can forget to set is not a boundary; a flag the compiler enforces is.
Intelligence Config gathers Extraction, Accuracy and AI Settings in pipeline order, and System Config gathers Gate Rules, Users and Feature Flags, both on the same keyboard-navigable tab strip the Delivery Performance page uses.
Product noteGrouping settings in the order the system runs them teaches the pipeline while you configure it.
Shared law-firm and fund-admin domains stop hijacking matches
Replaying 518 archived emails through production matched 136 of 471; the four causes behind most misses are fixed, starting with intermediary domains that appear on many deals no longer counting as company identity.
Product noteA replay against real mail is the only benchmark that tells you which rules are wrong rather than which tests are green.
The fund router now recognises putting, putt-putt and mini-golf vocabulary, so golf-entertainment deals route to Brekky Golf instead of the catch-all, and a deal can be held back from a backfill run.
Product noteClassification vocabulary should be widened from real misses, one honest example at a time.
Moving a deal into another Fund, Owner, Lead investor or Deal type lane now writes that attribute, and a combined lane-and-stage drop writes the lane first so stage gates evaluate the new value rather than the old one.
Product noteTwo axes on one drop only work if the order of writes matches the order the rules are checked in.
Swimlanes by age and last activity, and Backlog stays put
Deals can be laned by how long they have been in the pipeline or how recently anyone corresponded, youngest first, and Backlog deals no longer disappear when a grouping is active.
Product noteTime-based lanes answer the question a pipeline review actually asks: what has gone quiet.
The board now groups deals into lanes by Fund, Owner, Lead investor or Deal type, with a picker that remembers your choice per browser; lane counts and totals exclude test deals and say so.
Product noteGrouping logic lives in one pure module so every lane answers the same question the same way.
Automation no longer defaults new deals to whoever was created first (in practice Jay, on 61 of 65 real deals); owner defaults are now configuration with a never-owner rule, and each deal's fund is assigned by deterministic vocabulary rules with a stated rationale.
Product noteA default that nobody chose is still a decision, so make it explicit and make it auditable.
Stage gates are edited in the product, not by migration
Admins can now open /admin/stage-gates, see the five gates in pipeline order, switch any criterion on or off, and add checklist items that someone must tick before a deal can move; every change keeps its history.
Product noteWhen the rule that blocks a workflow can be changed by the person who owns the workflow, engineering stops being the bottleneck.
Synthetic test deals are now hidden behind an admin-only preference enforced by row-level security rather than component filters, so no current or future read path can leak them, and the data survives intact for when it is wanted.
Product noteChoosing a reversible visibility boundary over an irreversible purge preserves options while solving the problem.
The reject gate stopped demanding reviewed risk flags, a requirement AI generation often never produced, by switching the criterion off through a new active flag rather than deleting it, so deals can be rejected today and the rule's history survives.
Product noteSoft-deactivating a gate criterion keeps the audit trail while unblocking the workflow the same day.
Synthetic accounts created by automated testing are now flagged at the database and excluded from owner pickers, analytics and board counts, so the people you can assign a deal to are the people who actually work here.
Product noteA boundary that lives in the schema holds for every screen; one that lives in a query holds for that screen only.
Deal pages now show both the Fund and Company fieldsets at all times, with the inapplicable one locked behind a stated reason and a one-click path to classify the deal, ending the era of 21 undiscoverable fields; Correspondence also moved above Matched emails.
Product noteShowing a locked control with its reason teaches the data model; hiding it just hides the work.
New deals sweep the backlog for mail that already arrived
Creating a deal, or changing its name or domain, now triggers a bounded rescan of unmatched mail so earlier correspondence lands in the review queue instead of being lost to timing, and a reply into an already-matched thread is treated as conclusive.
Product notePutting the trigger in the database means every creation path present and future inherits it for free.
The July benchmark cost-per-point card is removed from Delivery Economics, leaving the extra-value card to carry the section while the benchmark itself continues to anchor the spend chart's reference line.
Product noteRemoving a redundant number is product work too; every figure on a page should earn its place.
Extraction, embedding, and routing now run on one-minute schedules guarded by database leases that prevent overlapping runs, and routing waits for vector signals before making a terminal decision on any email.
Product noteTightening a feedback loop is only safe once concurrency and ordering are guaranteed first.
The agentic development changelog gained its week of 24 August entries plus a 22 and 23 August supplement, seeded through a guarded migration that was rehearsed twice against production inside a rolled-back transaction before landing.
Product noteRehearsing a data migration in a rolled-back transaction gives production evidence without production risk.
Empty contact fields fill themselves from the thread
A scheduled pass now reads a deal's whole correspondence and fills in contact name, email and phone where they were blank, and only where they were blank; nothing a person entered is ever overwritten.
Product noteFill-empty-only is enforced in the database so the guarantee does not depend on any caller remembering it.
Cards say why a deal was rejected and which fund it belongs to
Rejected deals show their rejection reason as a red label on the card, and every card carries a fund-coloured line and badge (GL red, BG green, KV blue) so a glance across the board tells you whose deal it is.
Product noteThe board is read in seconds, so the two facts people ask about most belong on the card itself.
Six of six emails Jay matched by hand carried the deal name in the subject and still parked in review; subject-name and thread signals now count as strong evidence, so those emails attach on their own.
Product noteA manual review session is the best training data there is, provided someone reads it as a pattern rather than a to-do list.
The spend-versus-traditional trajectory moved from the OKR tab to Delivery Economics, so the tab reads as one narrative from cost per point through extra value to cumulative spend, and the descoped O4 KR3 no longer weighs on Q3 scoring.
Product noteRemoving a dead key result from the scoring keeps the scorecard describing reality.
Post-merge verification of the new visibility boundary against production found paths where synthetic deals still reached matching metrics and email search; both leaks are closed in a single migration.
Product noteVerifying a security boundary against production after merge is what separates enforced from intended.
Every gate requirement can be satisfied in the product
Walking one deal by hand through all five stage gates exposed seven required fields with no control anywhere in the product; each now has one, so a deal can clear every gate without a SQL write.
Product noteWalking the happy path by hand before declaring a workflow done is how hidden dead ends get found.
Deals now pass through five seeded gates, from Backlog exit to Reject entry, enforced in the database rather than the UI, with human-only not-applicable marks and a new flag that separates test deals from real ones.
Product noteEncoding the workshop's gate criteria as data means process changes ship as migrations, not rewrites.
The Extra Value card on Delivery Performance now reflects the retro-scored points for all fifteen unplanned items, so the dollar figure is computed from per-item evidence, and the OKR application itself is marked delivered.
Product noteClaiming value only after each item carries a score keeps the delivery narrative auditable.
Delivery Performance holds up at 320px and 200% zoom
A manual visual pass at 320px width, 200% zoom, and grayscale found real defects on the Delivery Performance page, including a tooltip that forced horizontal scroll at every viewport width; all are fixed in the presentation layer.
Product noteDeclaring a page done only after the constrained-viewport checks is the difference between shipped and verified.
Delivery Performance is reorganized into four tabs, with Delivery Economics as the default view, so the cost-and-value story opens the page and OKRs, the roadmap, and the extra-work ledger each get their own space.
Product noteLeading with the economics tab puts the question stakeholders actually ask on the first screen.
A new archive-then-purge script clears the 153 synthetic deals that were still live match candidates for inbound email, defaulting to a dry run and refusing to delete unless the operator states the expected deal count.
Product noteDestructive tooling that demands the operator predict its effect converts a risky cleanup into a checked one.
A new Roadmap tab lays out July 2026 through January 2027 as a month-by-month checklist under a computed AHEAD, ON TRACK, or BEHIND verdict, with an API that keeps it current as work ships.
Product noteA roadmap verdict computed from shipped work stays honest in a way a hand-updated slide never does.
The agentic development changelog gained its week of 17 August entries plus a supplement covering 15 and 16 August, seeded through a guarded migration that cannot duplicate rows on re-application.
Product noteA changelog that backfills its own gaps is a record, not a marketing page.
The OKR surface became Delivery Performance: objectives roll up by quarter, and a new financial layer shows cost per story point, quarterly spend rebalancing, and the statement-of-work link, with every figure read from the database rather than hardcoded.
Product notePutting delivery cost beside delivery output lets the spend conversation happen on evidence.
End-to-end testing of the new Roadmap tab surfaced four presentation defects, now fixed: month abbreviations no longer vary by ICU build, tooltips dismiss with Escape, and the tabs move with arrow keys.
Product noteAccessibility follow-through in the same week as the feature is what keeps it from becoming debt.
The security warning list is now an audited surface
All 44 SECURITY DEFINER advisor warnings were audited against production grants: one genuine over-grant, which let any signed-in user rewrite deal receipt dates, was revoked, and the remaining 43 are recorded as the platform's accepted RPC surface with written rationale.
Product noteAuditing findings to a documented baseline turns a warning count into a security posture.
The match review page now shows the original message body directly beneath the extracted output, visible by default and expandable, so a doubted score can be checked against its source.
Product noteShowing the evidence beside the inference is what makes a review more than an approval.
Deal detail now shows documents grouped by category with the current version prominent, prior versions one expander away, and drag-drop upload that rolls back cleanly instead of leaving a phantom file.
Product noteAn upload that fails visibly and completely earns more trust than one that half succeeds.
Deals now carry a document library with immutable versions, private storage, per-deal checksum dedupe and automatic promotion of email attachments by reference, so no attachment bytes are duplicated.
Product noteServer-chosen storage paths keep an upload surface safe without trusting anything the client sends.
An IC pack now composes from records the deal already holds, frozen as a snapshot alongside a completeness checklist and a target IC date, so the committee reviews a fixed version of the deal.
Product noteFreezing a snapshot at assembly time makes a committee decision reviewable against exactly what was presented.
Every error path in the email pipeline now passes through one redaction boundary, so addresses, phone numbers and quoted message text no longer reach production logs or admin-visible failure records.
Product noteA single boundary for sensitive text is auditable in a way that scattered discipline never is.
Replies are now stripped of their quoted chain before extraction, with the reply graph reconstructed so each message contributes its own words and only identity context from those above it.
Product noteRemoving duplicated input at the source improves accuracy more reliably than asking a model to ignore it.
A forwarded email is now split at its boundaries and each segment extracted with its own sender and date, so one company's figures no longer merge with another's and a deal is not filed under the introducer.
Product notePreserving provenance through every transformation is what keeps attribution correct downstream.
Runtime switches now gate product surfaces, administered from a page restricted to the product lead by an explicit email allowlist rather than the broader admin role.
Product noteSeparating who can release from who can administer keeps rollout decisions with one accountable owner.
Statements are attributed to the person who made them
Extraction now returns commitments, next steps, metrics, terms and claims bound to the message author who wrote them, with identity taken from parsed provenance rather than from the model.
Product noteDeriving identity from structure rather than inference keeps the record defensible.
The AI settings page now covers 113 audited variables across nine feature groups, each with its current value, default, source, editability and a plain-language note on what good looks like.
Product noteMaking configuration legible to non-engineers turns a tuning surface into a governance one.
The SMS channel now updates fields, moves stages and reports deal status, with every reply carrying a deep link and every call recorded on the deal's communication history.
Product notePre-checking a stage gate before writing means a blocked move changes nothing at all.
Summary and risk quality is measured, not asserted
Summaries and risk flags are now scored against a hand-labelled set of twelve historical deals, and a prompt or model change cannot go active without a new version id and a passing evaluation.
Product noteA measurable baseline turns model changes from a matter of opinion into a matter of evidence.
Each deal now assembles a read-only brief from what it already knows, drawing its summary, risk flags and confirmed term-sheet fields into a single meeting document that cannot contradict the deal card.
Product noteRe-presenting existing records instead of re-deriving them removes an entire class of inconsistency.
Add agentic development changelog entries for week of 10 August 2026
This week, the product moved forward with add agentic development changelog entries for week of 10 august 2026. Summary Seeds seven `agentic` changelog entries for `week_start = '2026-08-10'` (entry_date 2026-08-14) in a new migration, following the same shape as `20260807235900_agentic_changelog_week_0803.sql`: editorial rows with `pr_number`/`pr_url` null, `published` left to its default, `sort_index` 1..7, and a `not exists` guard on `(category = 'agentic', week_start)` so re-application is idempotent.
A deal created over SMS now shows the exchange that created it, and the agent actions taken on it, exactly as an email-created deal shows its source thread.
Product noteModelling a new channel on the proven shape of an existing one keeps the audit trail uniform across both.
Every agent action over SMS now carries the inbound text as a required input, so the communication record stores what the sender wrote rather than a reconstruction, and a call without it is refused.
Product noteAn audit trail is only as strong as its weakest entry, so incomplete provenance is better declined than stored.
PDF term sheets are now parsed field by field with confidence, page and verbatim excerpt attached, and every value waits in review rather than entering deal data.
Product noteHolding machine output in a pending state by default stops an extraction error from becoming a recorded fact.
Each deal card now carries a digest of what changed in its correspondence, timestamped and linked to the exact messages behind it, so a discussion can start without re-reading the thread.
Product noteRegenerating only once a burst of activity settles keeps a summary both current and inexpensive.
Parsed term-sheet fields now pass through a review screen where each value is confirmed by hand, with amendment history preserved so a later sheet never silently overwrites an earlier one.
Product noteNaming a single system of record removes the ambiguity of two places holding the same number.
A deal now shows the risk flags drawn from its own email evidence, each traceable to the message that raised it, and advisory only so nothing quietly moves a stage.
Product noteConstraining model output to a governed taxonomy keeps judgement with the people who own the decision.
ANIMAL gained a machine interface, so a text can create a deal, look for existing matches, or attach a note, with duplicate detection handled by the same matcher the email pipeline uses.
Product noteReusing logic already proven in production keeps a new channel consistent with the one users trust.
The Summary and Dense controls on a deal's correspondence timeline now change how much of each message is shown, and a reader's own expansions hold until the next switch.
Product noteA control that looks interactive but changes nothing costs more confidence than an absent one.
Replays now show page content as it appeared on screen and name the person behind the session, while typed values stay masked and recording remains off unless a deployment enables it.
Product notePrivacy controls serve better when tuned surface by surface than when applied wholesale to everything.
The duplicated migration that halted every production deploy was removed, so database changes flow through without disturbing the history already applied.
Product noteChoosing the remedy that preserves recorded state avoids trading a short outage for a long reconciliation.
Beyond clearing the leftover migration, filename validation now rejects two migrations with identical bodies, so the copy-instead-of-rename shortcut cannot recur.
Product noteEach incident is worth one automated check, so the lesson outlives the moment it was learned.
A version collision between two concurrently merged schema changes was resolved, restoring the checks that guard every pull request and the route to production.
Product noteTreating a red shared branch as the highest-priority item stops finished work from queueing behind it.
Session replay became an environment-level switch rather than a code decision, staying off until a deployment explicitly enables it. Input masking ships as part of that configuration, so an enabled replay still cannot leak what was typed.
Product noteObservability tooling earns its place faster when the privacy default is settled before the feature is turned on.
Seed agentic changelog entries for the week of 3 August 2026
This week, the product moved forward with seed agentic changelog entries for the week of 3 august 2026. Summary Seeds the 7 agentic-development changelog entries for `week_start = '2026-08-03'` (entry_date 2026-08-07), following the same pattern as the week-of-27-July seed (PR #170 / `20260731010000_agentic_changelog_week_0727.sql`): `category = 'agentic'`, null `title`/`pr_number`/`pr_url`, `sort_index` 1..7, wrapped in an idempotency guard.
Search across deals and correspondence from anywhere
A keyboard-summoned palette now searches deals and email together from any screen, grouping results and returning only what the viewer is permitted to see.
Product noteA visibility boundary becomes dependable once a named test gates it, so cross-fund isolation now fails the build rather than a review.
A single message or a whole thread can be re-assigned to the right deal from the correspondence timeline, and every move is kept in an append-only history of who moved what and when.
Product noteCorrections deserve an audit trail, since the record of a change is what makes trusting automation reasonable.
Board chrome collapses into one row and a bottom nav, lifting the visible column from 43 to 74 percent of a phone screen, and cards now lift on a long press instead of hijacking every scroll.
Product noteMeasuring the constrained case before and after keeps a mobile fix from becoming a matter of opinion.
Submitting a deal without a company name or fund returns a clear message in the form instead of failing silently behind a native browser prompt that mobile users never saw.
Product noteValidation that lives only in the browser leaves the failure path untested.
Dragging a deal highlights the destination column across its whole area rather than only the thin gaps between cards, so the drop target matches what the board looks like.
Product noteInteraction affordances have to agree with the visual boundaries people already read.
The review queue can turn several selected emails into a single new deal in one confirmed step, and the batch either completes in full or leaves nothing behind.
Product noteBulk actions become safe once partial success is designed out of them.
Extraction now runs in parallel on a tighter cron cadence, removing a 25-emails-per-hour ceiling that had held mean email-to-deal latency near 105 minutes.
Product noteMeasuring where the time actually goes prevents optimizing a stage that was never the constraint.
The Gmail admin view shows how many ingested emails are attached to no deal, derived from existing routing data so every attach and re-assign moves the number with no counter to drift.
Product noteInstrumenting a gap before building tooling for it keeps the next decision grounded in evidence.
Hovering the match confidence bars on a thread header now describes what the weighting means, matching the explanation individual messages already carried.
Product noteAn indicator that needs outside explanation is not yet finished.
The correspondence panel on a deal is now a threaded timeline, newest activity first, with collapsible threads, messages in sent order, and match confidence shown as visual weight. A weak match can be corrected in a single tap.
Product noteBuilding on an existing read model rather than beside it keeps visibility rules in one place.
The public changelog again carries every week from 13 July onward, and a reconciliation check compares merged work against published entries so a silently failing weekly job cannot thin the record.
Product noteA reporting pipeline needs a check that fails loudly, because a page that merely looks quiet reveals nothing.
The spend chart now plots logged actuals instead of the contracted retainer forecast, removing $280,000 of reported spend that was never incurred, and the savings headline follows the corrected figure. Months with nothing logged simply end the line.
Product noteA metric that quietly mixes forecast into actuals is less useful than no metric at all.
The platform became inspectable and searchable in the same week. The AI that reads inbound mail stopped being a black box — its settings are readable in plain English, every extraction can be examined, and impersonation attempts are flagged before they reach a deal. Alongside that, deal correspondence and its attachments became first-class: documents are browsable behind short-lived private links, and search now reaches deal names, companies and email content while still honouring who is entitled to see which fund.
Traditional team
2 senior full-stack engineers, 1 search/data engineer, 1 security engineer for the access matrix and audit trail, plus fractional QA and PM
Person-hours
~340 hrs5 people
Calendar time
~4 weeks
AssumptionsMid-to-senior contractor rates at ~6 productive hours/day, covering full-text search and ranking, signed-URL storage access, an admin observability surface, and the QA cycle a fund-scoped permission matrix demands.
Agentic deltaActual: one product lead directing agentic tooling across ~3 working days — a ~4-week, ~340-person-hour agency scope, including the security review a search feature over confidential data normally triggers.
See exactly what the extractor did
An observability page shows recent extractions, what was pulled from each message and where the process fell short.
Product noteConfidence in a model comes from being able to inspect its output, one message at a time.
Migration mistakes caught before they reach a branch
Continuous integration now rejects duplicate or malformed migration filenames in about a second, replacing an opaque database failure that used to surface thousands of log lines deep.
Product noteThe cheapest place to catch a class of failure is the first gate it can pass through.
Threads and their messages are read through a single query behind the deal emails endpoint, so a deal's mail arrives in sent order without a round trip per message.
Product noteShaping the read model before the interface keeps performance a design decision rather than a later repair.
The agentic development record covers the week of 27 July
The Updates screen now carries the week of 27 July in the agentic development record, keeping the account of how the work gets done as current as the product itself.
Product noteA delivery record only informs decisions while it is published on the same cadence as the work.
An admin panel now lays out the thresholds and models behind extraction and matching, each explained in ordinary language rather than configuration keys.
Product noteMaking automation legible to the people who own its outcomes is a prerequisite for tuning it.
This week, the product moved forward with seed agentic changelog for week of 20 july 2026. Summary Adds the **Agentic Development** changelog entries for the week of **20 July 2026 → 26 July 2026** (week ending 26 July; Friday run, evidence covers 20–24 July).
The redundant stage badge is gone from deal cards and the drag overlay, leaving the column to carry the stage and the card to carry owner and age at a glance.
Product noteRemoving duplicated signal is as much product work as adding features, because every element on a card competes for the same attention.
The email-to-deal pipeline turned into something the team can supervise. Confident matches attach themselves, borderline ones queue for review with their reasoning on show, genuinely new companies open their own deal in a holding state, and mail the system set aside is now visible and reclaimable. Accuracy stopped being a claim and became a dashboard. In parallel the deal record settled on its agreed field set, the board gained sorting that persists, and a database security advisory was closed with a verified fix — the pipeline became measurable and the record underneath it became stable.
Traditional team
2 senior full-stack engineers, 1 ML/data engineer on matching and thresholds, 1 designer for the review and board surfaces, plus fractional QA and PM
Person-hours
~420 hrs5 people
Calendar time
~5 weeks
AssumptionsMid-to-senior contractor rates at ~6 productive hours/day, covering confidence-threshold routing, a review UX with audit trail, an accuracy dashboard, a locked data-model migration and mobile layouts, with standard review and QA cycles — scoped toward the low end.
Agentic deltaActual: one product lead directing agentic tooling across ~5 working days — roughly a five-week, 420-person-hour agency engagement delivered inside a single week.
Privileged database routines locked down
Internal routines that run with elevated rights can no longer be invoked by anonymous callers, closing a gap flagged by the database security advisor.
Product noteSecurity findings get treated as delivery work with a verified fix, not as a note for later.
Deals now carry the agreed field set for funds and direct companies, with classification, ownership and reporting fields named and ordered the way the investment team uses them.
Product noteLocking the data model early is what keeps later reporting from becoming an archaeology exercise.
Deal type is now derived from the subclass instead of being entered twice, the classification fields follow the order people actually fill them in, and the stage is visible on the card.
Product noteEvery field that can be derived is a field nobody has to maintain.
This week, the product moved forward with gl-e15: idempotent okr sheet catch-up script. Summary The `/okrs` command center was backfilled from the Google Sheet on 2026-07-10 and hadn't been synced since, while the tracking agent kept writing to the sheet. This adds `scripts/catchup-okrs.mjs`, an **idempotent** catch-up that upserts only the sheet deltas into the `okr_*` tables (service role via `SUPABASE_SECRET_KEY`). It never deletes or reorders existing rows. The data catch-up was run directly against the DB; this PR just commits the reusable tool alongside `scripts/seed-okrs.mjs`.
This week, the product moved forward with fix skipped review_create_as_new_deal migration (version collision). Summary The "Create as new deal" review action failed in production with `Could not find the function public.review_create_as_new_deal(p_deal_id, p_decision_id) in the schema cache`. Root cause: **two migrations shared the same version prefix `20260721000000`**, having been authored on separate branches:
Guard realtime deal mapper against partial UPDATE payloads
This week, the product moved forward with guard realtime deal mapper against partial update payloads. Summary End-to-end testing of the needs-fund board (#148, GL-E05-S10) surfaced a crash that takes down the **entire** Kanban board when a realtime `postgres_changes` UPDATE arrives with a partial `new` record. In `patchDealFromRow` (`src/hooks/useRealtimeDeals.ts`):
Each review item now explains what its confidence score means in a tooltip, with the underlying match signals tucked into a collapsed panel for anyone who wants the detail.
Product noteProgressive disclosure keeps a screen calm while still making the reasoning available on demand.
When an email describes a company with no deal behind it, the platform now recognises the shape of a new opportunity and creates the deal, holding it aside until someone assigns the fund.
Product noteAuto-creation is only safe with a quarantine state, so nothing enters the pipeline unowned.
Messages from colleagues are no longer mistaken for deal correspondence, and a new Auto assigned tab shows exactly what the pipeline attached without being asked.
Product noteShowing what automation did unprompted is the fastest way to notice when it is wrong.
The platform's core promise — turning a CC'd inbox into structured deal flow — went from concept to a working ingestion spine. Inbound mail now lands securely, survives interruptions, and preserves its attachments in private storage, so no deal signal is silently lost between an email arriving and the pipeline seeing it. With access tokens encrypted at rest and the Gmail watch renewing itself daily, the confidential deal inbox became a dependable data source rather than a fragile one-off connection.
Traditional team
2 senior backend engineers (Gmail API, OAuth, Pub/Sub), 1 full-stack engineer for the admin surface, plus fractional QA and PM
Person-hours
~200 hrs4 people
Calendar time
~3 weeks
AssumptionsIntegration-heavy, security-sensitive work at mid-to-senior contractor rates, ~6 productive hours/day, including OAuth handling, dedup and resumable backfill, and QA of the failure paths.
Agentic deltaActual: one product lead directing agentic tooling across ~4 working days — roughly an order of magnitude less effort and calendar time than the ~200 person-hours over ~3 weeks a traditional shop would book.
Emails find the deal they belong to
Extracted details are now matched against existing deals by name, domain and meaning rather than exact spelling, and each candidate carries a score explaining why it ranked where it did.
Product noteRanked candidates with visible reasoning let a person overrule the system instead of guessing at it.
Anything the matcher was unsure about lands in a review screen where it can be accepted, rejected or reassigned to another deal, and every decision is recorded.
Product noteCapturing the human decision is what turns a queue into training data for the next round of matching.
Confident matches attach themselves; the rest wait for a human
Mail now routes by confidence: a strong match attaches to its deal automatically, a borderline one goes to review, and a clearly new company is flagged as a new deal.
Product noteAutomation is safest when its threshold is explicit and everything below it is handed to a person.
The hourly job that turns new mail into structured deal data was being redirected to the login page before it could run; it now executes as intended, with a check that keeps every scheduled job reachable.
Product noteA scheduled job that fails silently is worse than one that is missing, so the fix ships with a test that would have caught it.
Each week on the changelog carries a short Week in Review: the business outcome of the work, a conservative estimate of what a traditional team would have spent on the same scope, and how the actual delivery compared.
Product notePublishing the comparison alongside the work keeps delivery claims measurable rather than rhetorical.
Deal emails read themselves into structured fields
Incoming mail is now read for the details that matter — company, fund, stage, amounts, people — so a message arrives as the beginnings of a deal record instead of text somebody has to retype.
Product noteExtraction earns its place only when the output is structured enough to act on, which is why the fields mirror the deal model exactly.
The admin Gmail page now carries plain-language tooltips on each operational field, so connection health, capture timing, and watch expiry read clearly without external documentation.
Product noteExplaining operational state in place lowers the expertise required to keep an integration healthy.
This week, the product moved forward with chore: remove gmail-e2e-token helper; document oauth playground minting. Summary Removes the throwaway `scripts/gmail-e2e-token.mjs` helper now that both `GMAIL_E2E_REFRESH_TOKEN`s (animal@ → prod, test@ → preview) have been minted and set. As agreed in #104, the script was a temporary utility to decrypt a stored refresh token from `gmail_integrations`; it's no longer needed.
Nothing in the deal inbox gets processed twice — or lost
Inbound mail is now handled exactly once even when Gmail redelivers it, and anything that cannot be read is parked in a holding queue for review instead of disappearing.
Product noteA pipeline is only trustworthy when its failures are visible, so parking bad messages beats dropping them quietly.
ci: manual workflow to run the real Gmail staging-inbox e2e (GL-E04-S4)
This week, the product moved forward with ci: manual workflow to run the real gmail staging-inbox e2e (gl-e04-s4). Summary Wires the real send-to-self Gmail staging-inbox e2e (`e2e/gmail-staging-inbox.spec.ts`, GL-E04-S4 / AC3) to actually run — it previously only ever **skipped** because the standard `Playwright e2e` CI job passes placeholder Supabase creds and no Gmail secrets. This test sends and deletes real email against the production/preview mailbox + DB, so it must not run on every PR; instead it gets a dedicated **manual** workflow.
This week, the product moved forward with align footer logout with nav items using a list. Summary The footer logout button was visually out of line with the other nav links because the nav was a bare flex row and logout was wrapped in a `<form>` (block element) rather than sharing the same layout as the links.
Add agentic changelog entries for week of 13 July 2026
This week, the product moved forward with add agentic changelog entries for week of 13 july 2026. Summary Seeds the `agentic` changelog track for the week of 13-19 July 2026 (week_start `2026-07-13`, entry_date `2026-07-17`). These are authored editorial rows, not PR-sourced, so `pr_number`/`pr_url` stay null and `published` uses its `true` default, matching the existing `20260710190000_agentic_changelog_seed.sql` pattern.
Inbound attachments now extract into a private, service-role-only bucket with a metadata row each, and blocked or oversize files are recorded with a reason rather than silently dropped.
Product noteRecording rejected inputs instead of discarding them keeps a complete, auditable trail of what actually arrived.
The weekly changelog job now tracks a resumable cursor and backfilled the entries an earlier gap had missed, so shipped work reaches the public page without manual catch-up.
Product noteMaking a recurring job resumable ensures a single missed run repairs itself rather than compounding.
Email capture recovers from interruptions on its own
Ingestion now checkpoints against already-saved messages and skips ones that have vanished, so an interrupted mailbox backfill resumes where it left off instead of starting over.
Product noteDesigning for resumability up front keeps a long-running sync from becoming an all-or-nothing gamble.
Inbound deal email is captured into structured records
Verified Gmail notifications now flow through history sync and MIME parsing into a deduplicated store of raw emails, turning the CC'd inbox into the first stage of the deal pipeline.
Product noteMaking the capture path write-once and replay-safe means the same message can never create duplicate work downstream.
Gross Labs users invited to the platform are now pre-provisioned so their Google sign-in links to the assigned role immediately, removing the rejection that used to block a first login.
Product noteClosing the gap between invitation and first sign-in keeps onboarding friction from eroding early trust.
An admin can now link the dedicated animal@grosslabs.com mailbox through Google sign-in, with access tokens encrypted at rest and a daily job renewing the Gmail watch so capture never lapses.
Product noteAutomating credential renewal removes the silent expiry that quietly breaks most email integrations.
This week, the product moved forward with hold unsupported typescript and eslint majors. Summary Dependabot currently proposes two majors that the Next.js lint stack cannot load:
This was the week the product became real. Deals moved off ad-hoc notes onto a secured, audited Supabase model with role-level access enforced and tested on every change, and the kanban pipeline went live on production data with realtime sync, precise drag-and-drop, and stage gates that stop incomplete deals from advancing. In parallel a full OKR command center, privacy-safe usage analytics, supply-chain scanning, and the public changelog all shipped. The prototype became a governed, decision-grade platform the fund team can actually operate on.
Traditional team
2 senior full-stack engineers, 1 security/data engineer (RLS, auth, supply-chain hardening), 1 designer, plus fractional QA and PM
Person-hours
~600 hrs5 people
Calendar time
~6 weeks
AssumptionsA small delivery team at mid-to-senior contractor rates, ~6 productive hours/day, covering data modeling, RLS test matrices, realtime, and the OKR suite with standard review and QA cycles — scoped toward the low end.
Agentic deltaActual: one product lead directing agentic tooling in ~6 working days — a roughly 6-week, 600-person-hour agency effort compressed into a single week, and the gap compounds week over week.
OKR scores now read as percentages
Key result and objective progress now shows as clear percentages across the command center, replacing raw scores so status is legible at a glance.
Product noteSmall legibility choices compound: a number people can read instantly is a number they will actually act on.
A new overview chart ranks every objective's rolled-up score in one view, so the quarter's leaders and laggards are obvious before drilling into any single goal.
Product noteA comparison view turns a list of scores into a decision aid for where attention should go next.
A dedicated OKRs space brings the whole quarter into one view — every objective, the key results beneath it, the daily activity and evidence behind each score, and spend measured against benchmarks.
Product notePutting goals and the evidence for them in one place turns OKRs from a quarterly ritual into a living instrument the team can steer by.
Every third-party GitHub Action in the build pipeline is now pinned to an exact commit, so a repointed upstream tag can no longer slip unreviewed code into a release.
Product notePinning the supply chain converts a class of invisible risk into something reviewed on every change.
Changelog gains a dedicated agentic-development track
The public changelog now separates product releases from the story of how the agent itself is evolving, giving stakeholders two clear narratives on a single page.
Product noteSeparating streams of work by audience keeps each update legible instead of blurring the signal together.
Daily updates now flow into the command center through a secure, dedicated write path, so scores and activity stay fresh without anyone editing the tracker by hand.
Product noteAutomating the update path removes the reporting tax that quietly lets most goal trackers drift out of date.
The objectives, key results, evidence, and spend behind the OKR command center now sit in a normalized model with row-level security and immutable audit history from the first migration.
Product noteEnforcing access rules at the data layer before the feature ships makes confidentiality a property of the system rather than a later patch.
Magic-link authentication with role-aware sessions now protects every route, so confidential deal data is only ever visible to the people it belongs to.
Product noteBuilding auth before features means security is a property of the system, not a retrofit.
The venture began with the unglamorous work that determines everything after it: a typed, secure Next.js and Supabase foundation, a walkable kanban prototype the team could react to before any backend was committed, and a complete ANIMAL brand system with editorial typography, motion, and light and dark theming. The payoff was direction — stakeholders shaped the pipeline against real screens, and every later feature inherited type safety, managed auth, and a distinct identity instead of having them bolted on afterward.
Traditional team
1 senior full-stack engineer, 1 brand/UI designer, plus a fractional PM/tech lead
Person-hours
~140 hrs3 people
Calendar time
~2.5 weeks
AssumptionsDiscovery-phase pace with a design-then-build handoff at mid-to-senior contractor rates, ~6 productive hours/day; a full brand system with theming alone typically consumes 1-2 designer-weeks.
Agentic deltaActual: one product lead directing agentic tooling in ~5 working days — matching a ~2.5-week, ~140-person-hour agency engagement in the project's very first week.
ANIMAL brand system lands with light and dark themes
A complete brand system — monochrome palette, editorial typography, motion and theming — now gives the product a distinct, premium identity across every screen.
Product noteEstablishing design tokens early makes consistency the default rather than a cleanup task.
Kanban pipeline takes shape with a working prototype
A mock-data kanban board and deal views let the team walk the full pipeline — Backlog through Addepar — and react to real screens before any backend was committed.
Product notePrototyping the workflow with mock data surfaces design feedback while changes are still cheap.
Platform foundation laid with typed, secure scaffolding
The deal intelligence platform starts on a Next.js, Supabase and TypeScript foundation, giving every future feature type safety, managed auth and a database with row-level security from day one.
Product noteChoosing boring, well-supported infrastructure up front keeps later sprints focused on user problems rather than plumbing.