Before writing a line of Haven I spent a day reading the market instead of the code: what already exists in this category, what people actually complain about in reviews, and what the research says about the one mechanic every wellness app reaches for. Some of what came back changed the product. Some of it stopped me building things I had already planned.
This is that scan, including the parts that argue against me.
The Chrome extension category is, functionally, empty
This is the finding I least expected. Search the Chrome Web Store for mood tracking and you find extensions with install counts in the twenties. One had 28 users. Another had 29. The most complete one — a journal with coaching and habits — has 209 users and has not shipped an update since 2023, and it is cloud-based with personally identifiable information collected.
The closest competitor by positioning is AB Plugins, whose tagline is literally “mental health, one tab away”: six free local-first extensions covering mood logging, breathing prompts, feed blocking, crisis hotlines and intrusive-thought capture. They are good, and they matter for one specific reason — they already own the local-storage claim. “Your data never leaves your device” is not a differentiator in this category any more.
What nobody has is integration. Six separate single-purpose extensions is six separate records. There is no diary card, no thought record, no exposure hierarchy, no safety plan, and no export in any of them. Thirteen instruments over one local store, cross-referenced and exportable as one document, is the actual gap.
On mobile, the DBT diary card is iOS-only
Every credible DBT diary card app I could find runs on iPhone and nowhere else. The best of the new ones is on-device by default, needs no account to make a first entry, and makes therapist-ready date-range PDF export its headline feature — behind a subscription. There is no serious browser-based diary card at all, which is odd given that typing a week of skills, urges and emotions on a keyboard is materially better than doing it on a phone.
What people actually complain about
Feature lists are written by product managers. Review sections are written by people who paid. The complaints cluster hard, and almost none of them are about missing features.
- Paying to read your own history. The single most resented pattern in the category, stated almost identically across several products. One reviewer: they had a good experience until the app wanted money to see the data they had recorded, and then they left.
- Backup behind a paywall. One market leader gates automatic backup behind its premium tier, so free users must remember to export manually — and the reviews are full of people who did not, and lost everything.
- Exports that quietly lose information. One app drops entries marked “none”, which destroys absence-of-symptom data. Another logs in six-hour blocks, so event ordering is gone and you cannot tell a warning sign from a trigger. A third has day-only timestamps and silently omits photos. A fourth exports “basic text lists” that reviewers call messy and hard to read. Export is where these products are weakest, almost universally.
- Setup burden. The most configurable tracker in the category is also the one whose reviews say it is complex to set up and takes two or three check-ins a day to maintain.
- Entries too personal to share. Several reviewers say plainly that they want to bring their log to an appointment and cannot, because the writing in it is far more personal than the ratings. Nobody offers a partial export.
That last one became a feature. Haven has a single switch that keeps your written answers out of anything you share while keeping every rating, date and ticked skill. As far as I can find, no competitor offers it.
Streaks: the evidence says do not
I had planned achievements. The research is unusually direct about this, and it points the other way.
Silverman and Barasch, Journal of Consumer Research 49(6), 2023 measured four things: an intact streak increases engagement regardless of actual past performance; the streak displaces the underlying goal, because people come to treat maintaining the logged streak as the goal itself; the damage on a break is worst when the person attributes the break to themselves; and it is reduced when the streak can be repaired.
A February 2026 systematic review of gamification in digital mental-health interventions is blunter. Its documented harms include psychological distress from constant performance feedback, anxiety from social comparison, and demotivation from failed achievement streaks — in exactly the population these apps serve. It ranks points and badges (used by around 70% of the apps it reviewed) as extrinsic-motivation traps, and streak mechanics as fostering dependency through loss aversion.
The counterweight is real and I will not pretend otherwise: two randomised controlled trials show positive outcomes from gamified mental-health tools. Both use narrative and skill-building mechanics, not loss. That is the line the literature actually supports — progress and mastery framing is fine; loss, competition and variable reward are not.
So Haven has milestones, and they are constrained by construction: off by default, cumulative only, and a test in the build fails if the words “streak”, “keep it up”, “missed”, “congratulations” or “better than” appear anywhere in their copy. The audit also changed code that already existed — the dashboard used to show a check-in streak, and it now shows a longest run, which is the same information in a number that cannot fall.
Demo data: the pattern is fine, the usual implementation is not
Mainstream usability guidance treats “explore with demo data” as uncontroversial; Nielsen Norman Group cites it approvingly as an empty-state pattern. The counter-argument, from onboarding practitioners, is that demo data is always slightly unconvincing because it is chosen to make the product look good.
I could find no published research on the specific failure I was worried about — users confusing seeded sample entries with their own, or sample data reaching an export — and I am recording that as an evidence gap rather than pretending otherwise. But the absence of published evidence is not the absence of risk, and here the risk is asymmetric: a fabricated low mood appearing in a PDF someone hands to a clinician is a serious harm, and a local-only tool has no undo and no support ticket.
The tell is that every mature product which seeds a real store eventually publishes a “how do I delete the demo data” support article. So Haven does not seed a store. The example week is generated in memory and read through a layer that refuses every write and every export.
Free features that competitors charge for
The useful output of a scan like this is a list of things that cost a solo developer nothing and cost the incumbents a subscription.
- Clinician-grade printable export, every module, any date range. Paywalled at two major trackers, absent entirely at a third, and the headline paid feature of the best DBT app. It is a print stylesheet.
- Full-fidelity data export. Entry id, second-precision timestamp and explicit nulls in the CSV — which fixes the ordering and granularity complaints above by simply not having the bug.
- Correlation statistics with no tier. Arithmetic over a few thousand local rows.
- A documented, re-importable schema. Everyone in this category is a roach motel. Publishing the format costs a document.
- Redacted export. Novel, cheap, and directly answers a stated unmet need.
- A safety plan that works. The reference implementation of the Stanley-Brown protocol is a free PDF; the flagship app for it has 3.8 stars from eight ratings and a sign-up flow that a reviewer reported broken in 2017.
Four things I decided not to build
These are the decisions the research settled, and they are permanent rather than deferred.
- AI insights or an LLM journaling coach. It needs network access, which the content security policy forbids by design — and that policy is the strongest claim the product has. Competitors charge for this feature precisely because it costs them money per request.
- Sync, cross-device, or backup to a server. The moment any of it exists there is a server bill, a breach surface, and mental-health data on it. Having none of those is the differentiator.
- Social, accountability or sharing features. Social comparison is a named harm in the 2026 ethics review. The one integrated competitor that built it has 209 users and has not shipped since 2023.
- A pet or creature companion. The market leader for this mechanic has reviewers reporting that the pet becomes burdensome over time, it is the exact design the ethics literature warns about, and it is not winnable against a funded art team.
One more, held separately: standardised instruments — PHQ-9, GAD-7, Y-BOCS. Technically free, since they are static questionnaires. But displaying a response to PHQ-9 item nine, which asks about thoughts of self-harm, with no clinical wrapper around it, is a safety problem rather than a feature. That decision predates this scan and the scan did not change it.
The gap in this research, stated plainly
Reddit was unreachable from the environment the scan ran in — every request returned a 403. That means the community sentiment I most wanted (streak guilt in the bipolar and ADHD communities, paper-versus-digital diary card preferences among DBT clients, ERP app experiences) is unverified. The peer-reviewed sources carried the argument on their own, and the community voices would have been corroboration rather than the case itself. But a research summary that quietly omits what it could not reach is worse than one that names it, so: that is the hole.
What it changed
Six things were built out of this: the redacted export, the read-only example week, cumulative-only milestones, the existing streak display being replaced, the full-fidelity CSV, and the printable pack. Four were declined permanently and three deferred with reasons written down. The whole disposition — sixteen findings, each with what was done about it — ships in the repository rather than living only in my head.
The product it describes is on the Haven project page, and the design reasoning behind it is in a journal that does not flinch. Nothing is on the Chrome Web Store yet.